Csi feedback method, apparatus, device, and storage medium

Abstract

A CSI feedback method, device, equipment and storage medium, relate to the field of mobile communication. The method comprises: encoding CSI using an encoder (12) to obtain CSI feedback information; the encoder (12) is trained by a real training set and a first supplementary training set, the first supplementary training set is generated in a generator of a generative adversarial network, and the generative adversarial network is trained based on the real training set; and sending the CSI feedback information to an access network device (20).

Description

Technical Field

[0001] This application relates to the field of mobile communications, and in particular to a method, apparatus, device and storage medium for channel state information (CSI) feedback. Background Technology

[0002] In current New Radio (NR) systems, for CSI feedback schemes, terminals typically employ codebook-based eigenvector feedback to enable the base station to acquire the CSI of the downlink channel. Specifically, the base station sends downlink CSI Reference Signals (CSI-RS) to the user. The terminal uses the CSI-RS to estimate the CSI of the downlink channel and performs eigenvalue decomposition on the estimated downlink channel to obtain the corresponding eigenvector. Furthermore, NR provides two codebook design schemes: Type 1 and Type 2. Type 1 codebooks are used for CSI feedback of normal accuracy and for Single-User MIMO (SU-MIMO) and Multi-User MIMO (MU-MIMO) transmissions, while Type 2 codebooks are used to improve the transmission performance of MU-MIMO. Summary of the Invention

[0003] This application provides a CSI feedback method, apparatus, device, and storage medium, and proposes a CSI feedback scheme based on generative adversarial networks. The technical solution is as follows.

[0004] According to one aspect of this application, a CSI feedback method is provided, applied in a terminal, the method comprising:

[0005] The CSI is encoded using an encoder to obtain CSI feedback information; the encoder is trained using a real training set and a first supplementary training set, the first supplementary training set being generated by a generator in an adversarial generative network, and the adversarial generative network being trained based on the real training set.

[0006] The CSI feedback information is sent to the access network equipment.

[0007] According to one aspect of this application, a CSI feedback method is provided, applied in an access network device, the method comprising:

[0008] The terminal receives CSI feedback information, which is obtained by the terminal through CSI encoding using an encoder.

[0009] The CSI feedback information is decoded using a decoder to obtain the CSI measured by the terminal; the encoder and the decoder are trained using a real training set and a first supplementary training set, the first supplementary training set being generated by a generator in an adversarial generative network, and the adversarial generative network being trained based on the real training set;

[0010] The CSI feedback information is sent to the access network equipment.

[0011] According to one aspect of this application, a CSI feedback device is provided, the device comprising:

[0012] An encoding module is used to encode CSI using an encoder to obtain CSI feedback information; the encoder is trained on a real training set and a first supplementary training set, the first supplementary training set being generated by a generator in an adversarial generative network, the adversarial generative network being trained based on the real training set;

[0013] The sending module is used to send the CSI feedback information to the access network equipment.

[0014] According to one aspect of this application, a CSI feedback device is provided, the device comprising:

[0015] The receiving module is used to receive CSI feedback information sent by the terminal, wherein the CSI feedback information is obtained by the terminal through CSI encoding by an encoder;

[0016] The decoding module is used to decode the CSI feedback information using the decoder to obtain the CSI measured by the terminal; the encoder and the decoder are trained from a real training set and a first supplementary training set, the first supplementary training set being generated by the generator in the adversarial generative network, and the adversarial generative network being trained based on the real training set.

[0017] According to one aspect of this application, a terminal is provided, the terminal comprising: a processor; a transceiver connected to the processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to load and execute the executable instructions to implement the CSI feedback method as described above.

[0018] According to one aspect of this application, a network device is provided, the network device comprising: a processor; a transceiver connected to the processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to load and execute the executable instructions to implement the CSI feedback method as described above.

[0019] According to one aspect of this application, a computer-readable storage medium is provided that stores executable instructions which are loaded and executed by the processor to implement the CSI feedback method as described above.

[0020] According to one aspect of this application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium, a processor of a computer device reading the computer instructions from the computer-readable storage medium, and the processor executing the computer instructions, causing the computer device to perform the CSI feedback method described above.

[0021] According to one aspect of this application, a chip is provided, the chip including programmable logic circuitry or a program, the chip being used to implement the CSI feedback method as described in the above aspect.

[0022] The technical solutions provided in this application have at least the following beneficial effects:

[0023] When there are few training samples in the real training set, adversarial generative networks are used to generate supplementary training sets, thereby training high-performance encoders and decoders. Using these encoders and decoders to complete CSI feedback can improve the accuracy of CSI feedback between terminals and network devices, and use less feedback data to represent more complete and detailed channel information. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is an architecture diagram of a CSI feedback method provided in an exemplary embodiment of this application;

[0026] Figure 2 This is a flowchart of a CSI feedback method provided in an exemplary embodiment of this application;

[0027] Figure 3 This is a flowchart of an adversarial generative network provided in an exemplary embodiment of this application;

[0028] Figure 4 This is a schematic diagram of training an adversarial generative network provided in an exemplary embodiment of this application;

[0029] Figure 5This is a flowchart of a training method for an encoder and decoder provided in an exemplary embodiment of this application;

[0030] Figure 6 This is a training diagram of the encoder and decoder provided in an exemplary embodiment of this application;

[0031] Figure 7 This is a flowchart of a CSI feedback method provided in an exemplary embodiment of this application;

[0032] Figure 8 This is a flowchart of a model update method provided in an exemplary embodiment of this application;

[0033] Figure 9 This is a flowchart of a model update method provided in an exemplary embodiment of this application;

[0034] Figure 10 This is a flowchart of a model update method provided in an exemplary embodiment of this application;

[0035] Figure 11 This is a flowchart of a model update method provided in an exemplary embodiment of this application;

[0036] Figure 12 This is a flowchart of a model update method provided in an exemplary embodiment of this application;

[0037] Figure 13 This is a structural block diagram of a CSI feedback device provided in an exemplary embodiment of this application;

[0038] Figure 14 This is a structural block diagram of a CSI feedback device provided in an exemplary embodiment of this application;

[0039] Figure 15 This is a schematic diagram of the structure of a communication device provided in an exemplary embodiment of this application. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0041] Figure 1 A schematic diagram of a mobile communication system provided in one embodiment of this application is shown. The mobile communication system may include: a terminal 10 and an access network device 20.

[0042] The number of terminals 10 is typically multiple, with one or more terminals 10 distributed within the cell managed by each access network device 20. Terminals 10 may include various handheld devices, vehicle-mounted devices, wearable devices, computing devices, or other processing devices connected to a wireless modem, as well as various forms of user equipment (UE), mobile station (MS), etc. For ease of description, in this embodiment, the devices mentioned above are collectively referred to as terminals.

[0043] Access network device 20 is a device deployed in an access network to provide mobile communication functions for terminal 10. Access network device 20 may include various forms of macro base stations, micro base stations, relay stations, access points, Location Management Function (LMF) entities, etc. In systems employing different wireless access technologies, the name of the device with access network device functions may differ; for example, in a 5G NR system, it is called gNodeB or gNB. As communication technologies evolve, the name "access network device" may change. For ease of description, in this embodiment, the aforementioned device providing mobile communication functions for terminal 10 is collectively referred to as access network device. Access network device 20 and terminal 10 can establish a connection via an air interface, thereby communicating through this connection, including signaling and data exchange. There can be multiple access network devices 20, and two adjacent access network devices 20 can communicate via wired or wireless means. Terminal 10 can switch between different access network devices 20, that is, establish connections with different access network devices 20.

[0044] The "5G NR system" in this disclosure can also be referred to as a 5G system or an NR system, but those skilled in the art will understand its meaning. The technical solutions described in this disclosure are applicable to 5G NR systems and also to subsequent evolution systems of 5G NR systems.

[0045] In this embodiment, terminal 10 is equipped with an encoder 12, and access network device 20 is equipped with a decoder 22. Access network device 120 sends CSI-RS to terminal 10 on the downlink channel. Terminal 10 measures the CSI of the downlink channel based on CSI-RS. Terminal 10 encodes the CSI using encoder 12 to obtain CSI feedback information. Terminal 10 reports the CSI feedback information to access network device 20. Access network device 20 decodes the CSI of terminal 10 using decoder 22.

[0046] Since encoder 12 and decoder 22 are based on artificial intelligence models, they need to be pre-trained using real training samples. However, since there are relatively few training samples in the real training set, this application also proposes a sample supplementation scheme based on Generative Adversarial Networks (GANs), which can supplement a sufficient number of training samples, and the supplemented training samples have a very high similarity to the real training samples.

[0047] Figure 2 A flowchart of a CSI feedback method provided in an exemplary embodiment of this application is shown. This embodiment applies the method to... Figure 1 The method is illustrated using terminal 10 and network device 20 as examples. The method includes:

[0048] Step 202: The terminal uses an encoder to encode the CSI to obtain CSI feedback information;

[0049] The encoder is an AI encoding model used to encode CSI into CSI feedback information. Access network devices transmit CSI-RS to terminals via the downlink channel; the CSI is obtained by the terminal after measuring the CSI-RS.

[0050] CSI feedback information is the feedback bit sequence or feedback codebook obtained after the encoder encodes the CSI. The terminal uses the encoder to encode or compress the CSI to obtain the CSI feedback information. That is, the AI ​​coding model has non-linear fitting capability, which is used to compress and feed back the CSI. The encoder is also called a channel encoder.

[0051] Indicatively, the CSI feedback information is at least one of the following: feedback codebook, feature vector, matrix, and bit sequence.

[0052] Step 204: The terminal sends CSI feedback information to the access network equipment;

[0053] The terminal sends CSI feedback information to the access network equipment through the uplink feedback channel. This uplink feedback channel can be the Physical Uplink Control Channel (PUCCH) or the Physical Uplink Shared Channel (PUSCH).

[0054] Step 206: The access network device receives the CSI feedback information sent by the terminal. The CSI feedback information is obtained by the terminal through the encoder encoding the CSI.

[0055] Access network equipment receives CSI feedback information sent by the terminal through the uplink feedback channel.

[0056] Step 208: The access network device uses a decoder to decode the CSI feedback information to obtain the CSI measured by the terminal.

[0057] The encoder and decoder are trained using a real training set and a first supplementary training set. The first supplementary training set is generated by the generator in the Generative Adversarial Network (GAN), which is trained based on the real training set. The decoder is also called the channel decoder. The encoder and decoder can be collectively referred to as the CSI autoencoder.

[0058] Access network equipment uses a decoder to decode or reconstruct the CSI feedback information to obtain the CSI of the downlink channel measured by the terminal.

[0059] The neural network structure of the generator and discriminator is illustrative and can be at least one of Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), Gate Recurrent Unit (GRU), Recurrent Neural Network (RNN), and any other possible neural network architecture. This embodiment does not limit the specific network architecture of the generator and discriminator.

[0060] In summary, the method provided in this embodiment utilizes a generative adversarial network to generate a supplementary training set when there are few training samples in the real training set, thereby training a high-performance encoder and decoder. Using this encoder and decoder to complete CSI feedback can improve the accuracy of CSI feedback between the terminal and network devices, and use less feedback data to represent more complete and detailed channel information.

[0061] It should be noted that, Figure 2 Steps 202 and 204 in the embodiment can be implemented independently as a CSI feedback method on the terminal side. Figure 2 Steps 206 and 208 in the embodiments can be implemented independently as CSI feedback methods on the access network device side. Similarly, steps executed by the terminal in other embodiments can be implemented independently as corresponding methods on the terminal side, and steps executed by the access network device in other embodiments can be implemented independently as corresponding methods on the access network device side.

[0062] Training process of adversarial generative networks:

[0063] Figure 3A flowchart illustrating a training method for a generative adversarial network provided in an exemplary embodiment of this application is shown. This method can be performed by an access network device, a terminal, or other devices, and includes:

[0064] Generative Adversarial Networks (GANs) include Generator Neural Networks (GNNs) and Discriminator Neural Networks (DNNs). GANs are also known as Generative Adversarial Networks. The Generator Neural Network is simply called the generator, and the Discriminator Neural Network is simply called the discriminator.

[0065] Inspired by zero-sum games in game theory, Generative Adversarial Networks (GANs) view the generation problem as an adversarial game between a discriminator and a generator network: the generator produces synthetic data from given noise (typically uniformly or normally distributed), while the discriminator distinguishes the generator's output from real data. The former attempts to generate data closer to reality, while the latter tries to more perfectly differentiate between real and generated data. Thus, the two networks improve through this adversarial process, and continue to compete after each improvement. The data obtained by the generative network becomes increasingly perfect, approximating real data, thereby enabling the generation of the desired data.

[0066] The neural network structure of the generator and discriminator is illustrative and can be at least one of DNN, CNN, LSTM, GRU, RNN and any other possible neural network architecture. This embodiment does not limit the specific network architecture of the generator and discriminator.

[0067] Step 302: Input the training samples from the real training set into the discriminator to obtain the first discrimination result;

[0068] The training samples in the real training set include at least one of the following:

[0069] CSI

[0070] • Paired CSI and CSI feedback information (such as feedback codebooks obtained using high-precision quantization).

[0071] This application does not limit the form of training samples. This embodiment uses CSI as the training sample in the real training set for illustration.

[0072] The discriminator inputs training samples from the real training set to obtain the first discrimination result. Illustratively, this first discrimination result can be 0 or 1, where 0 represents false and 1 represents true. Alternatively, the first discrimination result can be a percentage probability value, representing the probability that the discrimination result is true. For example, 80% means that the probability of the discrimination result being true is 80%. A result exceeding a 50% threshold is considered true, and a result less than 50% is considered false.

[0073] In this embodiment, the discriminator can also be called a channel discriminator.

[0074] Step 304: Input the noise signal into the generator to obtain supplementary training samples;

[0075] The noise signal can be random noise, such as noise signal that follows a Gaussian distribution, or noise signal that follows a uniform distribution, noise signal that follows a Bernoulli two-dimensional distribution, or noise signal that follows other distributions. The noise signal can also be a noise signal containing known information.

[0076] When a noise signal is input into the generator, the generator will generate supplementary training samples based on the noise signal. The goal of these supplementary training samples is to be as similar as possible to or identical to the real training samples in the real training set. The generator can also be called a channel generator.

[0077] Step 306: Input the supplementary training samples into the discriminator to obtain the second discrimination result;

[0078] The supplementary training samples are input into the discriminator to obtain the second discrimination result.

[0079] It should be noted that steps 302 and 304-306 can be executed alternately or simultaneously, and this application does not limit the order in which steps 302 and 304-306 are executed. For example, step 302 can be executed first, followed by steps 304-306; or steps 304-306 can be executed first, followed by step 302. Furthermore, steps 302 and 304-306 can be executed alternately based on different training samples; or step 302 can be executed once, followed by steps 304-306 multiple times; or steps 304-306 can be executed once, followed by steps 302 multiple times.

[0080] Reference Figure 4 Taking an access network device as an example, the access network device can collect CSI data from multiple terminals as a real training set 31, such as the CSI data from multiple terminals within the same geographical area. On one hand, the real training samples 32 from the real training set 31 are input to the discriminator D, which outputs a corresponding discrimination result 35, i.e., the first discrimination result. On the other hand, random noise 33 is input to the generator G, which outputs a false (supplementary) training sample 34. This false training sample 34 is input to the discriminator D, which outputs a corresponding discrimination result 35, i.e., the second discrimination result. Then, based on the loss function of the first and second discrimination results, the generator and discriminator are trained.

[0081] Step 308: Based on the first and second discrimination results, train the generator and discriminator.

[0082] The generator's loss function is set for a target where both the first and second discrimination results are true, while the discriminator's loss function is set for a target where the first discrimination result is true and the second discrimination result is false.

[0083] To illustrate, during the discriminator training phase, the generator's neural model parameters are kept constant, and the discriminator is trained based on the discriminator's loss function. During the generator training phase, the discriminator's neural model parameters are kept constant, and the generator is trained based on the generator's loss function. These two training processes are executed alternately until the training termination condition is met.

[0084] Indicatively, training termination conditions include: the number of training iterations reaching a threshold, or the loss function converging.

[0085] This embodiment does not limit the training method of the adversarial generative network. For example, it can also be implemented using improved forms such as Wasserstein GAN with gradient penalty (WGAN-GP).

[0086] In summary, the method provided in this embodiment can train an adversarial generative network based on real training samples in a real training set. This adversarial generative network can generate fake training samples that are as similar as possible to the real training samples. When the number of real training samples in the real training set is limited, this adversarial generative network can generate a sufficient number of fake training samples as supplementary training samples.

[0087] The training process of the encoder and decoder:

[0088] Figure 5 A flowchart illustrating a training method for an encoder and decoder provided in an exemplary embodiment of this application is shown. This method can be performed by an access network device, a terminal, or other device, and includes:

[0089] Step 402: Generate the first supplementary training set using the generator in the Generative Adversarial Network;

[0090] Suppose that the number of training samples required to train the encoder and decoder is M, and the number of real training samples in the real training set (or original dataset) is m. If there are limitations on the real training set, such as acquisition constraints, time constraints, or cost constraints, and the size of the real training set m is much smaller than the required training set size M (i.e., m << M), then directly using the real training set to train the encoder will not yield a coding model with good performance.

[0091] After the adversarial generative network (GCN) is trained, the generator has a good sample generation capability. Based on the generator in the GCN, a first supplementary training set can be generated. The size of the first supplementary training set is not less than (Mm).

[0092] Step 404: Mix the real training set and the first supplementary training set to obtain the joint training set;

[0093] By mixing the real training samples from the real training set with the supplementary training samples from the first supplementary training set, a joint training set can be obtained. The size of this joint training set is equal to or greater than M.

[0094] That is, the size of the joint training set is sufficient to support the number of training samples required by the encoder and decoder.

[0095] Step 406: Train the encoder and / or decoder using a joint training set to obtain the trained encoder and / or decoder.

[0096] The encoder and / or decoder are trained using a joint training set to obtain the trained encoder and / or decoder.

[0097] In one example, each training sample in the joint training set is a CSI, and the encoder and decoder are trained end-to-end. In another example, each training sample in the joint training set includes a set of CSIs and a CSI feedback codebook. Based on each training sample, the encoder can be trained independently, the decoder can be trained independently, or the encoder and decoder can be trained end-to-end jointly.

[0098] This embodiment does not limit the training method of the encoder and / or decoder. (Illustrative reference) Figure 6 The training device combines the real training set 41 and the supplementary training set 42 into a joint training set 43. This joint training set 43 includes multiple training samples, such as each training sample being a CSI (Content Instance Analysis). The CSIs in this joint training set 43 are used to train the encoder and decoder. Specifically, the training samples are input to the encoder, which trains itself to obtain CSI feedback information 44. The decoder decodes the CSI feedback information 44 and outputs the recovered channel 45. The error between the recovered channel 45 and the training samples is calculated based on the loss function 46, and the encoder and decoder are trained end-to-end using the error backpropagation method.

[0099] In summary, the method provided in this embodiment generates a first supplementary training set through an adversarial generative network, and then uses a joint training set obtained by mixing the real training set and the first supplementary training set. This enables the use of sufficient training samples to train an encoder and decoder with better performance, thereby improving the compression efficiency and feedback accuracy during CSI feedback.

[0100] CSI feedback process based on access network equipment as model training device:

[0101] Figure 7 A flowchart of a CSI feedback method provided in an exemplary embodiment of this application is shown. This method can be performed by an access network device or a terminal, and includes:

[0102] Step 502: The access network device trains an adversarial generative network based on the real training set;

[0103] The training process for generative adversarial networks can be referenced above. Figure 3 The training process illustrated in the example will not be described in detail again.

[0104] Step 504: The access network device trains the encoder and decoder based on the joint training set;

[0105] The training process for the encoder and decoder can be referred to the above. Figure 5 The training process illustrated in the example will not be described in detail again.

[0106] Step 506: The access network device sends the encoder to the terminal;

[0107] Access network devices send encoders or encoder model parameters to terminals via downlink signaling, and the terminals construct encoders themselves based on the encoder model parameters.

[0108] Schematic illustration: The downlink signaling includes at least one of Downlink Control Information (DCI), Radio Resource Control (RRC), and Medium Access Control Element (MAC CE). The downlink signaling may also be proprietary signaling and channel resources specifically for model delivery; this application does not limit this aspect.

[0109] Illustratively, the model parameters of the encoder and / or decoder include at least one of the following: neural network type, number of neural network layers, type of neural network layers, type of neurons in the neural network, number of neurons in the neural network, and matrix weights of neurons in the neural network.

[0110] Step 508: The terminal uses an encoder to encode the CSI to obtain CSI feedback information;

[0111] Step 510: The terminal sends CSI feedback information to the access network equipment;

[0112] The terminal uses the uplink feedback channel to send CSI feedback information to the access network equipment.

[0113] Step 512: The access network device receives the CSI feedback information sent by the terminal. The CSI feedback information is obtained by the terminal through the encoder encoding the CSI.

[0114] Step 514: The access network device uses a decoder to decode the CSI feedback information to obtain the CSI measured by the terminal.

[0115] In other embodiments, the training equipment described above can also be executed by a terminal. The terminal trains an adversarial generative network (GAN) based on a real training set; the terminal trains an encoder and decoder based on a joint training set; the terminal reports the decoder or its model parameters to the access network device. The terminal encodes the CSI using the encoder to obtain CSI feedback information. The terminal sends the CSI feedback information to the access network device. The access network device receives the CSI feedback information sent by the terminal, which is obtained by the terminal encoding the CSI using the encoder. The access network device decodes the CSI feedback information using the decoder to obtain the CSI measured by the terminal.

[0116] In other embodiments, the first device may train the Generative Adversarial Network (GAN) and send the GAN or its model parameters to the second device. The second device generates a supplementary training set based on the GAN and trains the encoder and decoder based on the joint training set. The first device is an access network device, and the second device is a terminal; or, the first device is a terminal, and the second device is an access network device.

[0117] In summary, the method provided in this embodiment, by leveraging the strong computing power of the access network device, can quickly train encoders and decoders with excellent performance. The access network device then sends the encoder to the terminal to complete the AI-based CSI feedback process, improving compression efficiency and feedback accuracy during CSI feedback.

[0118] The encoder and / or decoder update training process based on periodic feedback:

[0119] Limited by the encoder's generalization ability and the complex and variable channel environment, when the channel environment changes and the encoder becomes incompatible, leading to performance degradation, model parameters need to be updated. However, since the encoder of the CSI autoencoder is deployed on the UE side while the decoder is deployed on the network side, when model mismatch occurs, the network side cannot obtain a large amount of high-quality, effective channel data after significant changes through CSI feedback. Continuously using a high-precision codebook to compress and feed back changes in the channel introduces high feedback overhead. Therefore, this embodiment employs a generative adversarial network approach to reduce the feedback density of the high-precision codebook and generate a second supplementary dataset based on a small amount of channel feedback to complete the encoder model update.

[0120] Figure 8 A flowchart illustrating a model update method provided in an exemplary embodiment of this application is shown. This method can be performed by an access network device and a terminal, and includes:

[0121] Step 602: The terminal periodically sends the first CSI feedback codebook based on codebook quantization to the access network equipment;

[0122] The first CSI feedback codebook is obtained based on codebook quantization, rather than on AI models or encoder compression. Illustratively, the first CSI feedback codebook is a high-quality CSI feedback codebook capable of effectively representing the changed channel environment.

[0123] Since there can be multiple first CSI feedback codebooks, the first CSI feedback codebook can also be called the second original training set, the second real training set, the updated training set, etc.

[0124] Indicatively, the first CSI feedback codebook includes CSI feedback information from multiple sampling time points, such as multiple CSI feedback information within a time window W, or multiple CSI feedback information sampled at a specified period within a time window W.

[0125] Indicatively, the first CSI feedback codebook includes CSI feedback information on multiple broadband, subband, or frequency points, such as multiple CSI feedback information within frequency window B, or multiple CSI feedback information on multiple broadband and / or subbands of frequency window B.

[0126] As an illustration, the feedback period of the first CSI feedback codebook can be determined by the configuration parameter T. The configuration parameter T of the first CSI feedback codebook can also carry other configuration information required during the codebook feedback process.

[0127] In one example, prior to step 602, the access network device sends a first reporting configuration to the terminal, for example, via downlink signaling. The terminal receives the first reporting configuration sent by the access network device, which indicates the first reporting parameters of the first CSI feedback codebook. Illustratively, the first reporting parameters include at least one of the following: a time window W; a frequency window B; and configuration parameters T of the first CSI feedback codebook.

[0128] In another example, prior to step 602, the terminal determines a first reporting configuration, which indicates the first reporting parameters of the first CSI feedback codebook. The terminal sends the first reporting configuration to the access network device. For example, the terminal sends the first reporting configuration to the access network device via uplink signaling.

[0129] Step 604: The access network device receives the first CSI feedback codebook sent by the terminal;

[0130] Step 606: The access network device constructs a second supplementary training set based on the first CSI feedback codebook through the adversarial generative network;

[0131] The access network device constructs a second supplementary training set based on the first CSI feedback codebook using an adversarial generative network; or, the access network device determines the CSI corresponding to the first CSI feedback codebook and constructs a second supplementary training set based on the CSI corresponding to the first CSI feedback codebook using an adversarial generative network; or, the access network device determines the CSI corresponding to the first CSI feedback codebook and constructs a second supplementary training set based on the first CSI feedback codebook and the CSI corresponding to the first CSI feedback codebook using an adversarial generative network.

[0132] The method of training the generative adversarial network based on the first CSI feedback codebook, and... Figure 3 The embodiments shown are similar, and will not be described again in this embodiment.

[0133] In one example, the adversarial generative network is... Figure 3 The illustrated embodiment shows an adversarial generative network.

[0134] In one embodiment, the adversarial generative network is an adversarial generative network separately trained based on the first CSI feedback codebook, and this adversarial generative network is... Figure 3 The adversarial generative network shown in the embodiment is different.

[0135] Step 608: The access network device updates and trains at least one of the encoder and decoder using the first CSI feedback codebook and the second supplementary training set;

[0136] Indicatively, the access network device uses a first joint update training set, consisting of a first CSI feedback codebook and a second supplementary training set, to jointly update and train the encoder and decoder. The updated encoder can then be well adapted to the changed channel information.

[0137] Access network devices can update training only on the encoder; or only on the decoder; or update training on both the encoder and decoder.

[0138] Step 610: The access network device sends the updated encoder to the terminal.

[0139] When updating the encoder, the access network device sends the updated encoder to the terminal. The entire process can be referenced below. Figure 9As shown. In step 602, the quantization accuracy of the high-precision codebook can be further enhanced based on the Type 2 codebook to improve the recovery accuracy of CSI and ensure the accuracy of the first CSI feedback codebook. Simultaneously, the feedback-related configurations of the first CSI feedback codebook—period parameter T, time window W, frequency window B, etc.—can all be configured by the network side and notified to the UE via DCI. The process of generating the second supplementary dataset based on the first CSI feedback codebook in step 606 is related to... Figure 3 The method for generating the first supplementary dataset in the embodiments is the same. However, since it involves updating the model, the sample size of the generated second supplementary dataset is generally less than that of the first dataset. Figure 3 The required dataset sample size M in Example 1 is used to ensure that the model convergence of the Generative Adversarial Network and the rapid update of the CSI autoencoder model can be completed quickly.

[0140] In another embodiment, the device for updating the encoder and / or decoder can also be a terminal. The terminal periodically measures a first CSI to form an updated training set; the terminal constructs a second supplementary training set based on the updated training set using a generative adversarial network; the terminal updates and trains at least one of the encoder and decoder using the codebook fed back from the first CSI and the second supplementary training set. The terminal reports the updated decoder to the access network device. The specific process is similar to... Figure 8 The embodiments shown are similar and will not be described again.

[0141] In summary, the method provided in this embodiment, through a periodic update approach during the online deployment of the CSI autoencoder, utilizes a lower-density, high-precision codebook for feedback and dataset updates, and is combined with a generative adversarial network, to achieve online updates of the CSI autoencoder model, ensuring the accuracy of CSI feedback and recovery.

[0142] The update training process of the encoder and / or decoder based on non-periodic feedback:

[0143] Figure 10 A flowchart illustrating a model update method provided in an exemplary embodiment of this application is shown. This method can be performed by an access network device and a terminal, and includes:

[0144] Step 702: When the triggering conditions are met, the terminal sends a second CSI feedback codebook based on codebook quantization to the access network device;

[0145] The second CSI feedback codebook is obtained based on codebook quantization, rather than on AI models or encoder compression. Illustratively, the second CSI feedback codebook is a high-quality CSI feedback codebook capable of effectively representing the changed channel environment.

[0146] Since there can be multiple second CSI feedback codebooks, the second CSI feedback codebook can also be called the second original training set, the second real training set, the updated training set, etc.

[0147] The triggering conditions include: changes in channel state, channel information, or channel parameters exceeding a preset threshold. Channel state, channel information, or channel parameters include at least one of: Reference Signal Received Power (RSRP), Reference Signal Received Quality (RSRQ), Reference Signal Strength Indicator (RSSI), and CSI.

[0148] The preset threshold can be predefined by the communication protocol; or, the preset threshold can be preconfigured; or, the preset threshold can be configured by the access network device to the terminal.

[0149] Indicatively, the second CSI feedback codebook includes CSI feedback information from multiple sampling time points, such as multiple CSI feedback information within a time window W, or multiple CSI feedback information sampled at a specified period within a time window W.

[0150] Indicatively, the second CSI feedback codebook includes CSI feedback information on multiple broadband, subband, or frequency points, such as multiple CSI feedback information within frequency window B, or multiple CSI feedback information on multiple broadband and / or subbands of frequency window B.

[0151] As an illustration, the feedback period of the second CSI feedback codebook can be determined by the configuration parameter T. The triggering condition of the second CSI feedback codebook can also carry other configuration information required during the codebook feedback process.

[0152] In one example, prior to step 702, the access network device sends a second reporting configuration to the terminal, for example, via downlink signaling. The terminal receives the second reporting configuration sent by the access network device, which indicates the second reporting parameters of the second CSI feedback codebook. Illustratively, the second reporting parameters include at least one of the following: a time window W; a frequency window B; and a triggering condition for the second CSI feedback codebook, such as... Figure 11 As shown.

[0153] In another example, prior to step 702, the terminal determines a second reporting configuration, which indicates the second reporting parameters of the second CSI feedback codebook. The terminal sends the second reporting configuration to the access network device. For example, the terminal sends the second reporting configuration to the access network device via uplink signaling, such as... Figure 12As shown.

[0154] Step 704: The access network device receives the second CSI feedback codebook sent by the terminal;

[0155] Step 706: The access network device constructs a third supplementary training set based on the second CSI feedback codebook through the adversarial generative network;

[0156] The access network device constructs a third supplementary training set based on the second CSI feedback codebook using an adversarial generative network; or, the access network device determines the CSI corresponding to the second CSI feedback codebook and constructs a third supplementary training set based on the CSI corresponding to the second CSI feedback codebook using an adversarial generative network; or, the access network device determines the CSI corresponding to the second CSI feedback codebook and constructs a third supplementary training set based on the second CSI feedback codebook and the CSI corresponding to the second CSI feedback codebook using an adversarial generative network.

[0157] The method of training the adversarial generative network based on the second CSI feedback codebook, and... Figure 3 The embodiments shown are similar, and will not be described again in this embodiment.

[0158] In one example, the adversarial generative network is... Figure 3 The illustrated embodiment shows an adversarial generative network.

[0159] In one embodiment, the adversarial generative network is an adversarial generative network separately trained based on a second CSI feedback codebook, which in turn... Figure 3 The adversarial generative network shown in the embodiment is different.

[0160] Step 708: The access network device updates and trains at least one of the encoder and decoder using the second CSI feedback codebook and the third supplementary training set;

[0161] As an illustration, the access network device uses a second joint update training set, consisting of a second CSI feedback codebook and a third supplementary training set, to jointly update and train the encoder and decoder. The updated encoder can adapt well to the changed channel information.

[0162] Access network devices can update training only on the encoder; or only on the decoder; or update training on both the encoder and decoder.

[0163] Step 710: The access network device sends the updated encoder to the terminal.

[0164] When updating the encoder, the access network device sends the updated encoder to the terminal. In step 702, the quantization accuracy of the high-precision codebook can be further enhanced based on the Type 2 codebook to improve the recovery accuracy of CSI and ensure the accuracy of the second CSI feedback codebook. Simultaneously, the feedback-related configurations of the second CSI feedback codebook—period parameter T, time window W, frequency window B, etc.—can all be configured by the network side and notified to the UE via DCI. The process of generating the second supplementary dataset based on the second CSI feedback codebook in step 706 is related to... Figure 3 The method for generating the first supplementary dataset in the embodiments is the same. However, since it involves updating the model, the sample size of the generated second supplementary dataset is generally less than that of the first dataset. Figure 3 The required dataset sample size M in Example 1 is used to ensure that the model convergence of the Generative Adversarial Network and the rapid update of the CSI autoencoder model can be completed quickly.

[0165] In another embodiment, the device for updating the encoder and / or decoder can also be a terminal. The terminal periodically measures a second CSI to form an updated training set; the terminal constructs a third supplementary training set based on the updated training set using a generative adversarial network; the terminal updates and trains at least one of the encoder and decoder using the codebook fed back from the second CSI and the third supplementary training set. The terminal reports the updated decoder to the access network device. The specific process is similar to... Figure 8 The embodiments shown are similar and will not be described again.

[0166] In summary, the method provided in this embodiment, by updating the CSI autoencoder model online based on trigger conditions during the online deployment of the CSI autoencoder, can reduce the amount of communication data between the terminal and the server compared to the previous embodiment.

[0167] Figure 13 This illustration shows a block diagram of a CSI feedback device provided in an exemplary embodiment of this application. The device can be implemented as a terminal or a functional module within a terminal. The device includes:

[0168] The encoding module 1320 is used to encode the CSI using an encoder to obtain CSI feedback information; the encoder is trained on a real training set and a first supplementary training set, the first supplementary training set being generated by a generator in an adversarial generative network, and the adversarial generative network being trained based on the real training set.

[0169] The sending module 1340 is used to send the CSI feedback information to the access network equipment.

[0170] In an optional embodiment, the encoder is trained in the following manner:

[0171] The generator in the adversarial generative network is used to generate the first supplementary training set;

[0172] The real training set and the first supplementary training set are mixed to obtain a joint training set;

[0173] The encoder is trained using the joint training set to obtain the trained encoder.

[0174] In an optional embodiment, the adversarial generative network includes a generator and a discriminator, and the adversarial generative network is trained in the following manner:

[0175] The training samples in the real training set are input into the discriminator to obtain the first discrimination result;

[0176] The noise signal is input into the generator to obtain supplementary training samples; the supplementary training samples are input into the discriminator to obtain a second discrimination result;

[0177] Based on the first discrimination result and the second discrimination result, the generator and the discriminator are trained.

[0178] The loss function of the generator is set with the goal that both the first discrimination result and the second discrimination result are true, and the loss function of the discriminator is set with the goal that the first discrimination result is true and the second discrimination result is false.

[0179] In an optional embodiment, the apparatus further includes:

[0180] The receiving module 1360 is used to receive the encoder sent by the access network device, wherein the encoder is trained by the access network device.

[0181] In an optional embodiment, the sending module 1340 is further configured to periodically send a first CSI feedback codebook based on codebook quantization to the access network device; the receiving module 1360 is configured to receive an updated encoder sent by the access network device, wherein the updated encoder is obtained by the access network device after updating and training the encoder based on the first CSI feedback codebook and a second supplementary training set, and the second supplementary training set is constructed by the generative adversarial network based on the first CSI feedback codebook.

[0182] In an optional embodiment, the receiving module 1360 is further configured to receive a first reporting configuration sent by the access network device, the first reporting configuration being used to indicate the first reporting parameters of the first CSI feedback codebook;

[0183] or,

[0184] The sending module 1340 is further configured to send a first reporting configuration to the access network device, wherein the first reporting configuration is used to indicate the first reporting parameters of the first CSI feedback codebook.

[0185] In an optional embodiment, the first reporting parameter includes at least one of the following: time window W; frequency window B; configuration parameter T of the first CSI feedback codebook.

[0186] In an optional embodiment, the sending module 1340 is further configured to send a second CSI feedback codebook based on codebook quantization to the access network device when a triggering condition is met; the receiving module 1360 is further configured to receive an updated encoder sent by the access network device, wherein the updated encoder is obtained by the access network device after updating and training the encoder based on the second CSI feedback codebook and a third supplementary training set, wherein the third supplementary training set is constructed by the generative adversarial network based on the second CSI feedback codebook.

[0187] In an optional embodiment, the triggering condition includes:

[0188] The change in channel parameters exceeds a preset threshold;

[0189] The channel state, channel information, or channel parameters include at least one of RSRP, RSRQ, RSSI, and CSI.

[0190] In an optional embodiment, the apparatus further includes:

[0191] The receiving module 1360 is further configured to receive a second reporting configuration sent by the access network device, the second reporting configuration being used to indicate the second reporting parameters of the second CSI feedback codebook; or, the sending module 1340 is further configured to send a second reporting configuration to the access network device, the second reporting configuration being used to indicate the second reporting parameters of the second CSI feedback codebook.

[0192] In an optional embodiment, the second reporting parameter includes at least one of the following: time window W; frequency window B; triggering conditions of the second CSI feedback codebook.

[0193] Figure 14 This application shows a block diagram of a CSI feedback device provided in an exemplary embodiment. The device can be implemented as an access network device or a functional module within an access network device. The device includes:

[0194] The receiving module 1420 is used to receive CSI feedback information sent by the terminal, wherein the CSI feedback information is obtained by the terminal through CSI encoding by an encoder;

[0195] The decoding module 1440 is used to decode the CSI feedback information using the decoder to obtain the CSI measured by the terminal; the encoder and the decoder are trained from a real training set and a first supplementary training set, the first supplementary training set being generated by the generator in the adversarial generative network, and the adversarial generative network being trained based on the real training set.

[0196] In an optional embodiment, the encoder and the decoder are trained in the following manner:

[0197] The generator in the adversarial generative network is used to generate the first supplementary training set;

[0198] The real training set and the first supplementary training set are mixed to obtain a joint training set;

[0199] The encoder and the decoder are trained using the joint training set to obtain the trained encoder and the decoder.

[0200] In an optional embodiment, the adversarial generative network includes a generator and a discriminator, and the adversarial generative network is trained in the following manner:

[0201] The training samples in the real training set are input into the discriminator to obtain the first discrimination result;

[0202] The noise signal is input into the generator to obtain supplementary training samples; the supplementary training samples are input into the discriminator to obtain a second discrimination result;

[0203] Based on the first discrimination result and the second discrimination result, the generator and the discriminator are trained.

[0204] The loss function of the generator is set with the goal that both the first discrimination result and the second discrimination result are true, and the loss function of the discriminator is set with the goal that the first discrimination result is true and the second discrimination result is false.

[0205] In an optional embodiment, the sending module 1460 is used to send the encoder to the terminal.

[0206] In an optional embodiment, the receiving module 1420 is used to periodically receive a first CSI feedback codebook based on codebook quantization sent by the terminal;

[0207] The training module 1480 is used to construct a second supplementary training set based on the first CSI feedback codebook through the adversarial generative network; and to update and train at least one of the encoder and the decoder using the first CSI feedback codebook and the second supplementary training set.

[0208] In an optional embodiment, the receiving module 1420 is configured to receive a first reporting configuration sent by the terminal, the first reporting configuration being used to indicate the first reporting parameters of the first CSI feedback codebook; or, the sending module 1460 is configured to send a first reporting configuration to the terminal, the first reporting configuration being used to indicate the first reporting parameters of the first CSI feedback codebook.

[0209] In an optional embodiment, the first reporting parameter includes at least one of the following:

[0210] Time window W;

[0211] Frequency window B;

[0212] The configuration parameter T of the first CSI feedback codebook.

[0213] In an optional embodiment, the receiving module 1420 is configured to receive a second CSI feedback codebook based on codebook quantization sent by the terminal when the triggering condition is met;

[0214] The training module 1480 is used to construct a second supplementary training set based on the first CSI feedback codebook through the adversarial generative network; and to update and train at least one of the encoder and the decoder using the first CSI feedback codebook and the second supplementary training set.

[0215] In an optional embodiment, the sending module 1460 is configured to send an updated encoder to the terminal when the encoder is being updated and trained.

[0216] In an optional embodiment, the triggering condition includes: the change in channel parameters exceeding a preset threshold. Channel state, channel information, or channel parameters include at least one of RSRP, RSRQ, RSSI, and CSI.

[0217] In an optional embodiment, the receiving module 1420 is configured to receive a second reporting configuration sent by the terminal, the second reporting configuration being used to indicate a second reporting parameter of the second CSI feedback codebook; or, the sending module 1460 is configured to send a second reporting configuration to the terminal, the second reporting configuration being used to indicate a second reporting parameter of the second CSI feedback codebook.

[0218] In an optional embodiment, the second reporting parameter includes at least one of the following: time window W; frequency window B; triggering conditions of the second CSI feedback codebook.

[0219] Figure 15 The diagram shows a schematic representation of a communication device (terminal or access network device) provided in an exemplary embodiment of this application. The communication device includes a processor 101, a receiver 102, a transmitter 103, a memory 104, and a bus 105.

[0220] The processor 101 includes one or more processing cores. The processor 101 executes various functional applications and information processing by running software programs and modules.

[0221] The receiver 102 and the transmitter 103 can be implemented as a communication component, which can be a communication chip.

[0222] The memory 104 is connected to the processor 101 via the bus 105.

[0223] The memory 104 can be used to store at least one instruction, and the processor 101 can execute the at least one instruction to implement the various steps in the above method embodiments.

[0224] Furthermore, the memory 104 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, including but not limited to: magnetic disks or optical disks, electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), static random access memory (SRAM), read-only memory (ROM), magnetic storage, flash memory, and programmable read-only memory (PROM).

[0225] In an exemplary embodiment, a computer-readable storage medium is also provided, which stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the CSI feedback method executed by a first terminal, a second terminal, or a network device provided in the above-described method embodiments.

[0226] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a communication device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the communication device to perform the CSI feedback method performed by a first terminal, a second terminal, or a network device as described above.

[0227] The above description is merely an optional embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A Channel State Information (CSI) feedback method, characterized in that, When applied in a terminal, the method includes: The CSI is encoded using an encoder to obtain a CSI feedback codebook; the encoder is trained using a joint training set, which is based on a mixture of a real training set and a first supplementary training set, wherein the first supplementary training set is generated by a generator in an adversarial generative network, and the adversarial generative network is trained based on the real training set, wherein the real training set includes at least one of the following: CSI; paired CSI and CSI feedback codebook; Send the CSI feedback codebook to the access network equipment; The first CSI feedback codebook based on codebook quantization is periodically sent to the access network device. The first CSI feedback codebook is used to express the changed channel environment. The access network device receives an updated encoder, which is obtained by updating and training the encoder based on the first CSI feedback codebook and the second supplementary training set. The second supplementary training set is constructed by the adversarial generative network based on the first CSI feedback codebook. The updated encoder has higher adaptability to the changed channel information. The CSI feedback codebook in the real training set is further enhanced based on the type 2 codebook; When the joint training set includes CSI, the encoder and decoder are trained using an end-to-end training method; when the joint training set includes paired CSI and CSI feedback codebooks, the encoder can be trained independently, or the encoder and decoder can be trained using an end-to-end training method.

2. The method according to claim 1, characterized in that, The encoder was trained in the following manner: The first supplementary training set is generated using the generator in the adversarial generative network; The real training set and the first supplementary training set are mixed to obtain a joint training set; The encoder is trained using the joint training set to obtain the trained encoder.

3. The method according to claim 1, characterized in that, The adversarial generative network includes a generator and a discriminator, and is trained in the following manner: The training samples in the real training set are input into the discriminator to obtain the first discrimination result; The noise signal is input into the generator to obtain supplementary training samples; the supplementary training samples are input into the discriminator to obtain a second discrimination result; Based on the first discrimination result and the second discrimination result, the generator and the discriminator are trained. The loss function of the generator is set with the goal that both the first discrimination result and the second discrimination result are true, and the loss function of the discriminator is set with the goal that the first discrimination result is true and the second discrimination result is false.

4. The method according to any one of claims 1 to 3, characterized in that, The method further includes: The encoder is received from the access network device, and the encoder is trained by the access network device.

5. The method according to claim 1, characterized in that, The method further includes: Receive the first reporting configuration sent by the access network device, wherein the first reporting configuration is used to indicate the first reporting parameters of the first CSI feedback codebook; or, Send a first reporting configuration to the access network device, the first reporting configuration being used to indicate the first reporting parameters of the first CSI feedback codebook.

6. The method according to claim 5, characterized in that, The first reported parameter includes at least one of the following: Time window W; Frequency window B; The configuration parameter T of the first CSI feedback codebook.

7. The method according to any one of claims 1 to 3, characterized in that, The method further includes: When the triggering conditions are met, a second CSI feedback codebook based on codebook quantization is sent to the access network device; The access network device receives an updated encoder, which is obtained by updating and training the encoder based on the second CSI feedback codebook and the third supplementary training set. The third supplementary training set is constructed by the generative adversarial network based on the second CSI feedback codebook.

8. The method according to claim 7, characterized in that, The triggering conditions include: The change in channel parameters exceeds the preset threshold.

9. The method according to claim 7, characterized in that, The method further includes: Receive a second reporting configuration sent by the access network device, the second reporting configuration being used to indicate the second reporting parameters of the second CSI feedback codebook; or, Send a second reporting configuration to the access network device, the second reporting configuration being used to indicate the second reporting parameters of the second CSI feedback codebook.

10. The method according to claim 9, characterized in that, The second reported parameter includes at least one of the following: Time window W; Frequency window B; The configuration parameter T of the second CSI feedback codebook.

11. A Channel State Information (CSI) feedback method, characterized in that, Applied in access network equipment, the method includes: The terminal receives a CSI feedback codebook, which is obtained by the terminal encoding the CSI using an encoder. The CSI feedback codebook is decoded using a decoder to obtain the CSI measured by the terminal; the encoder and the decoder are trained using a joint training set, which is based on a mixture of a real training set and a first supplementary training set. The first supplementary training set is generated by a generator in an adversarial generative network, which is trained based on the real training set. The real training set includes at least one of the following: CSI; paired CSI and CSI feedback codebook. The first CSI feedback codebook based on codebook quantization is periodically received by the terminal. The first CSI feedback codebook is used to express the changed channel environment. The adversarial generative network constructs a second supplementary training set based on the first CSI feedback codebook; At least one of the encoder and the decoder is updated and trained using the first CSI feedback codebook and the second supplementary training set. When the encoder is updated and trained, the updated encoder is sent to the terminal. The updated encoder has a higher adaptability to the changed channel information. The CSI feedback codebook in the real training set is further enhanced based on the type 2 codebook; When the joint training set includes CSI, the encoder and decoder are trained using an end-to-end training method; when the joint training set includes paired CSI and CSI feedback codebooks, the decoder supports individual training, or the encoder and decoder are trained using an end-to-end training method.

12. The method according to claim 11, characterized in that, The encoder and the decoder are trained in the following manner: The first supplementary training set is generated using the generator in the adversarial generative network; The real training set and the first supplementary training set are mixed to obtain a joint training set; The encoder and the decoder are trained using the joint training set to obtain the trained encoder and the decoder.

13. The method according to claim 11, characterized in that, The adversarial generative network includes a generator and a discriminator, and is trained in the following manner: The training samples in the real training set are input into the discriminator to obtain the first discrimination result; The noise signal is input into the generator to obtain supplementary training samples; the supplementary training samples are input into the discriminator to obtain a second discrimination result; Based on the first discrimination result and the second discrimination result, the generator and the discriminator are trained. The loss function of the generator is set with the goal that both the first discrimination result and the second discrimination result are true, and the loss function of the discriminator is set with the goal that the first discrimination result is true and the second discrimination result is false.

14. The method according to any one of claims 11 to 13, characterized in that, The method further includes: The encoder is sent to the terminal.

15. The method according to claim 11, characterized in that, The method further includes: Receive the first reporting configuration sent by the terminal, the first reporting configuration being used to indicate the first reporting parameters of the first CSI feedback codebook; or, Send a first reporting configuration to the terminal, the first reporting configuration being used to indicate the first reporting parameters of the first CSI feedback codebook.

16. The method according to claim 15, characterized in that, The first reported parameter includes at least one of the following: Time window W; Frequency window B; The configuration parameter T of the first CSI feedback codebook.

17. The method according to any one of claims 11 to 13, characterized in that, The method further includes: Receive the second CSI feedback codebook based on codebook quantization sent by the terminal when the triggering conditions are met; The adversarial generative network constructs a third supplementary training set based on the second CSI feedback codebook; At least one of the encoder and the decoder is updated and trained using the second CSI feedback codebook and the third supplementary training set.

18. The method according to claim 17, characterized in that, The triggering conditions include: The change in channel parameters exceeds the preset threshold.

19. The method according to claim 17, characterized in that, The method further includes: The terminal sends a second reporting configuration, which is used to indicate the second reporting parameters of the second CSI feedback codebook; or, A second reporting configuration is sent to the terminal, the second reporting configuration being used to indicate the second reporting parameters of the second CSI feedback codebook.

20. The method according to claim 19, characterized in that, The second reported parameter includes at least one of the following: Time window W; Frequency window B; The triggering conditions for the second CSI feedback codebook.

21. A Channel State Information (CSI) feedback device, characterized in that, The device includes: An encoding module is used to encode the CSI using an encoder to obtain a CSI feedback codebook; the encoder is trained on a joint training set, which is based on a mixture of a real training set and a first supplementary training set, wherein the first supplementary training set is generated by a generator in an adversarial generative network, and the adversarial generative network is trained on the real training set, wherein the real training set includes at least one of the following: CSI; paired CSI and CSI feedback codebook; The sending module is used to send the CSI feedback codebook to the access network equipment; The sending module is also used to periodically send a first CSI feedback codebook based on codebook quantization to the access network device. The first CSI feedback codebook is used to express the changed channel environment. The receiving module is used to receive the updated encoder sent by the access network device. The updated encoder is obtained by the access network device after updating and training the encoder based on the first CSI feedback codebook and the second supplementary training set. The second supplementary training set is constructed by the adversarial generative network based on the first CSI feedback codebook. The updated encoder has higher adaptability to the changed channel information. The CSI feedback codebook in the real training set is further enhanced based on the type 2 codebook; When the joint training set includes CSI, the encoder and decoder are trained using an end-to-end training method; when the joint training set includes paired CSI and CSI feedback codebooks, the encoder can be trained independently, or the encoder and decoder can be trained using an end-to-end training method.

22. The apparatus according to claim 21, characterized in that, The encoder was trained in the following manner: The first supplementary training set is generated using the generator in the adversarial generative network; The real training set and the first supplementary training set are mixed to obtain a joint training set; The encoder is trained using the joint training set to obtain the trained encoder.

23. The apparatus according to claim 21, characterized in that, The adversarial generative network includes a generator and a discriminator, and is trained in the following manner: The training samples in the real training set are input into the discriminator to obtain the first discrimination result; The noise signal is input into the generator to obtain supplementary training samples; the supplementary training samples are input into the discriminator to obtain a second discrimination result; Based on the first discrimination result and the second discrimination result, the generator and the discriminator are trained. The loss function of the generator is set with the goal that both the first discrimination result and the second discrimination result are true, and the loss function of the discriminator is set with the goal that the first discrimination result is true and the second discrimination result is false.

24. The apparatus according to any one of claims 21 to 23, characterized in that, The receiving module is further configured to receive the encoder sent by the access network device, wherein the encoder is trained by the access network device.

25. The apparatus according to claim 21, characterized in that, The device further includes: The receiving module is further configured to receive a first reporting configuration sent by the access network device, wherein the first reporting configuration is used to indicate the first reporting parameters of the first CSI feedback codebook; or, The sending module is further configured to send a first reporting configuration to the access network device, wherein the first reporting configuration is used to indicate the first reporting parameters of the first CSI feedback codebook.

26. The apparatus according to claim 25, characterized in that, The first reported parameter includes at least one of the following: Time window W; Frequency window B; The configuration parameter T of the first CSI feedback codebook.

27. The apparatus according to any one of claims 21 to 23, characterized in that, The device further includes: The sending module is also used to send a second CSI feedback codebook based on codebook quantization to the access network device when the triggering conditions are met; The receiving module is further configured to receive the updated encoder sent by the access network device. The updated encoder is obtained by the access network device after updating and training the encoder based on the second CSI feedback codebook and the third supplementary training set. The third supplementary training set is constructed by the adversarial generative network based on the second CSI feedback codebook.

28. The apparatus according to claim 27, characterized in that, The triggering conditions include: The change in channel parameters exceeds the preset threshold.

29. The apparatus according to claim 27, characterized in that, The device further includes: The receiving module is further configured to receive a second reporting configuration sent by the access network device, wherein the second reporting configuration is used to indicate the second reporting parameters of the second CSI feedback codebook; or, The sending module is further configured to send a second reporting configuration to the access network device, the second reporting configuration being used to indicate the second reporting parameters of the second CSI feedback codebook.

30. The apparatus according to claim 29, characterized in that, The second reported parameter includes at least one of the following: Time window W; Frequency window B; The triggering conditions for the second CSI feedback codebook.

31. A Channel State Information (CSI) feedback device, characterized in that, The device includes: The receiving module is used to receive the CSI feedback codebook sent by the terminal, wherein the CSI feedback codebook is obtained by the terminal through an encoder encoding the CSI; A decoding module is used to decode the CSI feedback codebook using a decoder to obtain the CSI measured by the terminal; the encoder and the decoder are obtained from a joint training set, which is based on a mixture of a real training set and a first supplementary training set, wherein the first supplementary training set is generated by a generator in an adversarial generative network, and the adversarial generative network is trained based on the real training set, wherein the real training set includes at least one of the following: CSI; paired CSI and CSI feedback codebook; The receiving module is used to periodically receive a first CSI feedback codebook based on codebook quantization sent by the terminal. The first CSI feedback codebook is used to express the channel environment after the change. The training module is used to construct a second supplementary training set based on the first CSI feedback codebook through the adversarial generative network; and to update and train at least one of the encoder and the decoder using the first CSI feedback codebook and the second supplementary training set. The sending module is used to send an updated encoder to the terminal when the encoder is being updated and trained. The updated encoder has a higher adaptability to the changed channel information. The CSI feedback codebook in the real training set is further enhanced based on the type 2 codebook; When the joint training set includes CSI, the encoder and decoder are trained using an end-to-end training method; when the joint training set includes paired CSI and CSI feedback codebooks, the decoder supports individual training, or the encoder and decoder are trained using an end-to-end training method.

32. The apparatus according to claim 31, characterized in that, The encoder and the decoder are trained in the following manner: The first supplementary training set is generated using the generator in the adversarial generative network; The real training set and the first supplementary training set are mixed to obtain a joint training set; The encoder and the decoder are trained using the joint training set to obtain the trained encoder and the decoder.

33. The apparatus according to claim 31, characterized in that, The adversarial generative network includes a generator and a discriminator, and is trained in the following manner: The training samples in the real training set are input into the discriminator to obtain the first discrimination result; The noise signal is input into the generator to obtain supplementary training samples; the supplementary training samples are input into the discriminator to obtain a second discrimination result; Based on the first discrimination result and the second discrimination result, the generator and the discriminator are trained. The loss function of the generator is set with the goal that both the first discrimination result and the second discrimination result are true, and the loss function of the discriminator is set with the goal that the first discrimination result is true and the second discrimination result is false.

34. The apparatus according to any one of claims 31 to 33, characterized in that, The device further includes: The sending module is used to send the encoder to the terminal.

35. The apparatus according to claim 31, characterized in that, The device further includes: The receiving module is used to receive the first reporting configuration sent by the terminal, wherein the first reporting configuration is used to indicate the first reporting parameters of the first CSI feedback codebook; or, The sending module is used to send a first reporting configuration to the terminal, wherein the first reporting configuration is used to indicate the first reporting parameters of the first CSI feedback codebook.

36. The apparatus according to claim 35, characterized in that, The first reported parameter includes at least one of the following: Time window W; Frequency window B; The configuration parameter T of the first CSI feedback codebook.

37. The apparatus according to any one of claims 31 to 33, characterized in that, The device further includes: The receiving module is used to receive the second CSI feedback codebook based on codebook quantization sent by the terminal when the triggering conditions are met; The training module is used to construct a third supplementary training set based on the second CSI feedback codebook through the adversarial generative network; and to update and train at least one of the encoder and the decoder using the second CSI feedback codebook and the third supplementary training set.

38. The apparatus according to claim 37, characterized in that, The triggering conditions include: The change in channel parameters exceeds the preset threshold.

39. The apparatus according to claim 37, characterized in that, The device further includes: The receiving module is used to receive the second reporting configuration sent by the terminal, the second reporting configuration being used to indicate the second reporting parameters of the second CSI feedback codebook; or, The sending module is used to send a second reporting configuration to the terminal, the second reporting configuration being used to indicate the second reporting parameters of the second CSI feedback codebook.

40. The apparatus according to claim 39, characterized in that, The second reported parameter includes at least one of the following: Time window W; Frequency window B; The triggering conditions for the second CSI feedback codebook.

41. A terminal, characterized in that, The terminal includes: processor; A transceiver connected to the processor; Memory for storing the executable instructions of the processor; The processor is configured to load and execute the executable instructions to implement the CSI feedback method as described in any one of claims 1 to 10.

42. A network device, characterized in that, The network device includes: processor; A transceiver connected to the processor; Memory for storing the executable instructions of the processor; The processor is configured to load and execute the executable instructions to implement the CSI feedback method as described in any one of claims 11 to 20.

43. A computer-readable storage medium, characterized in that, The readable storage medium stores executable instructions that are loaded and executed by a processor to implement the CSI feedback method as described in any one of claims 1 to 20.

44. A computer program product, characterized in that, The computer program product stores executable instructions that are loaded and executed by a processor to implement the CSI feedback method as described in any one of claims 1 to 20.

45. A chip, characterized in that, The chip includes programmable logic circuitry and is used to implement the CSI feedback method as described in any one of claims 1 to 20.

Images