Model adjustment method and communication apparatus

Abstract

The present application provides a model adjustment method and a communication apparatus. Upon completing the adjustment of an auto-encoder model at a network side, a network device processes, on the basis of a first processing mode, a data set provided by a terminal side and used for the adjustment of the auto-encoder model, and provides, to the terminal side, the processed data set, and an output of an encoder in the adjusted auto-encoder model when the processed data set is used as an input to the adjusted auto-encoder model. Upon collecting the data set, the terminal side can also process the data set, and provide the processed data set for the adjustment of the auto-encoder model at the network side. Then, a data set used for the adjustment of an encoder at the terminal side is received from the network side. The method facilitates performance optimization of an encoding model at a terminal side. In addition, in a pairing process of a dual-end model based on data set transfer, the overhead of data transmission is reduced.

Description

Methods for adjusting the model and communication devices

[0001] This application claims priority to Chinese Patent Application No. 202411641796.9, filed on November 15, 2024, entitled "Method and Communication Apparatus for Adjusting a Model", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of artificial intelligence (AI), and more specifically, to a method for adjusting a model and a communication device. Background Technology

[0003] In AI-based channel state information (CSI) feedback, when the model is deployed at the base station, the base station obtains the estimated results of the channel state information-reference signal (CSI-RS) at the user equipment (UE) side as labels for model training. An auto-encoder (AE) model can generally refer to a network structure consisting of two sub-models, such as an encoder and a decoder. AE models can also be called bilateral models, cooperative models, etc. The encoder and decoder of an AE model are generally trained together and can be used in a paired manner. CSI feedback can be implemented based on the AE model. In one implementation, the base station locally trains the encoder and decoder based on historically collected CSI data. To distribute the trained encoder to the UE side to achieve pairing of the two-end models, the base station constructs a dataset based on the encoder's input (i.e., measured CSI) and output (i.e., fed-back CSI) and transmits this dataset to the UE side. The UE side uses this dataset to train the encoder sub-model, enabling pairing of the two-end models on the base station and UE sides. However, the encoder sub-model trained using this method may contain significant discrepancies between the information obtained by the base station decoder from the UE's output of this sub-model in a real channel environment and the UE's encoder input. Therefore, a method is urgently needed to optimize this UE-side encoder sub-model. Summary of the Invention

[0004] This application provides a method and communication device for adjusting the model, which is beneficial for optimizing the encoding model of the transmitting end.

[0005] In a first aspect, a method for adjusting a model is provided, executed by a communication device or a module for the communication device (e.g., a processor, chip, circuit, AI entity, etc., or a logic module, hardware, and / or software capable of implementing all or part of the functions of the communication device), the communication device corresponding to the terminal device in the method embodiment. The method includes: sending a first dataset, the first dataset including measurement information of a reference signal, the first dataset being used for adjusting a self-model, the self-model including a first encoding model and a first decoding model; receiving a second dataset, the second dataset being related to a first processing method, the first processing method including a processing method for the first dataset, the second dataset being used for adjusting a third encoding model to obtain a fourth encoding model, the fourth encoding model being matched with a second decoding model in the adjusted self-model.

[0006] In the technical solution of this application, the terminal device sends a newly collected first dataset for network-side self-model adjustment to the network device. After completing the self-model adjustment, the network device sends a second dataset obtained by processing the first dataset based on a first processing method to the terminal device. The second dataset varies depending on the first processing method. Since the second dataset is determined based on the first dataset reported by the terminal device (for example, a third dataset is determined based on the first dataset, and then the second dataset is determined based on the third dataset), and the first dataset is obtained by the terminal device through measurement of a reference signal, compared to the network device recovering information from the output of the terminal-side encoding model through the decoding model in the self-model to obtain information for self-model adjustment, in this application, when the network device adjusts the self-model based on the terminal device's measurement dataset (i.e., the first dataset) and feeds back a portion based on the measurement dataset (i.e., the third dataset) as input to the self-model, the output of the encoding model in the adjusted self-model (i.e., the second dataset) is used for the adjustment of the terminal-side encoding model, which can optimize the performance of the terminal-side encoding model. Furthermore, applying the terminal-side coding model to the feedback of reference signal measurement information, such as channel state information (CSI), can improve the compression performance of the reference signal measurement information on the terminal side and the recovery performance on the network side.

[0007] Furthermore, when the first processing method is certain, the second dataset sent from the network side to the terminal side can reduce data transmission overhead compared to the dataset constructed by the network side based on the input and output of the encoder in the adjusted self-model.

[0008] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: adjusting the third encoding model based on the second dataset.

[0009] In this implementation, the terminal device sends a first dataset for network-side self-model adjustment to the network device, and receives a second dataset from the network device. Based on the second dataset, the terminal-side encoding model is adjusted to match the adjusted self-model on the network side.

[0010] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: obtaining a third dataset by receiving first information, the first information indicating one or more of the following: a third dataset, or one or more reference signals corresponding to the third dataset, or a first processing method, or parameters related to the first processing method; or, the first processing method is predefined, obtaining the third dataset based on the first processing method and the first dataset, wherein the third dataset and the second dataset are used together for adjusting the third coding model.

[0011] In this implementation, the network device sends first information, which indicates a third dataset or one or more reference signals corresponding to the third dataset. Compared to the network device sending the third dataset directly, the network device indicates one or more reference signals corresponding to the third dataset through the first information, so that the terminal device can determine the third dataset based on the first information and then use it for adjusting the third coding model. The overhead of the network device sending the first information is smaller; for example, the first information can be a bitmap or an index of the one or more reference signals, which can further reduce the data transmission overhead.

[0012] Optionally, the first information may indicate all or part of the parameters related to the first processing method and / or the first processing method.

[0013] For example, when the first information indicates the parameters corresponding to the first processing method, it implicitly indicates the first processing method. For instance, if the first information indicates the filtering ratio and / or augmentation method corresponding to SAS, it means the first processing method is SAS. Alternatively, if the parameter information corresponding to the first processing method is predefined, it is sufficient to indicate only the first processing method. For example, for multiple processing methods, such as data distillation, SAS, or core set-based methods, different indexes can be used to distinguish them. If the filtering ratio and augmentation method corresponding to SAS are predefined or specified by a protocol, then the first information only needs to indicate the index corresponding to SAS, i.e., only indicating that the first processing method is SAS. Furthermore, parameters related to the first processing method not indicated by the first information can be predefined or obtained through other means, and are not limited here.

[0014] In conjunction with the first aspect, in some implementations of the first aspect, adjusting the third coding model based on the second dataset includes: adjusting the third coding model according to one or more output data of the second coding model contained in the second dataset, and one or more measurement information corresponding to the one or more output data.

[0015] In this implementation, the terminal adjusts its encoding model based on a second dataset received from the network and the corresponding measurement information. The measurement information corresponding to the second dataset can be determined by first information sent from the network, or it can be sent from the network to the terminal if the terminal does not store the measurement information of the reference signal. The measurement information corresponding to the second dataset is the third dataset. The terminal sends the first dataset to the network, and the network processes the first dataset to obtain the third dataset. To reduce data transmission overhead, the network does not send the third dataset but instead indicates the reference signal corresponding to the third dataset through the first information. The terminal, based on the first information, determines the measurement information that matches the reference signal indicated by the first information from its local measurement information, thus also obtaining the third dataset. Furthermore, the terminal adjusts its encoding model using both the third and second datasets. Specifically, the third dataset is used as the input to the terminal's encoding model, and the second dataset is used as the output to adjust the terminal's model.

[0016] In conjunction with the first aspect, in some implementations of the first aspect, the first processing method includes subsets maximizing augmentation similarity (SAS), and the parameters related to the first processing method include the augmentation method and / or the screening ratio corresponding to SAS.

[0017] In this implementation, the network side filters the first dataset based on SAS and instructs the terminal side on the corresponding augmentation method and / or filtering ratio. The terminal side uses the same augmentation method and filtering ratio as the network side to augment and reconstruct the second and third datasets from the network side, and adjusts the encoding model based on the augmented and reconstructed datasets.

[0018] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: augmenting and reconstructing the second dataset and one or more measurement information corresponding to the second dataset based on the augmentation method and / or the screening ratio to obtain augmented and reconstructed data; the adjustment of the third coding model based on the second dataset includes: adjusting the second coding model based on the augmented and reconstructed data.

[0019] In conjunction with the first aspect, in some implementations of the first aspect, the first processing method also includes a core set-based method. The parameters related to the first processing method include the weights associated with the data in the third dataset when adjusting the model. The data in the third dataset includes one or more measurement information corresponding to the second dataset.

[0020] In conjunction with the first aspect, in some implementations of the first aspect, adjusting the third encoding model based on the second dataset includes: adjusting the third encoding model based on the weights associated with the data in the second and third datasets when used for model adjustment.

[0021] In the above implementation, when the first processing method is based on the core set, the weights associated with the data in the third dataset when used for model adjustment can be predefined, protocol-specified, or indicated by the network side to the terminal side. The data in the third dataset are also the measurement information of one or more reference signals corresponding to the second dataset. Since the second dataset uses the third dataset as input to the adjusted self-model and as output to the encoding model, the second and third datasets correspond to the same set of reference signals; in other words, the second and third datasets correspond. When adjusting the model, the terminal side adjusts the third encoding model based on the second dataset, the third dataset, and the weights associated with the data in the third dataset.

[0022] In conjunction with the first aspect, in some implementations of the first aspect, the first processing method is a data distillation method, the third dataset includes a generated dataset of the first dataset, and one or more output data of the second encoding model include: the output of the second encoding model when the generated dataset is used as the input of the adjusted self-model, and the second encoding model is an encoding model obtained by adjusting the first encoding model.

[0023] Furthermore, the adjustment of the third coding model based on the second dataset includes: adjusting the third coding model based on one or more output data of the second coding model and one or more measurement information corresponding to the one or more output data.

[0024] It should be understood that the second dataset includes one or more output data of the second coding model, and the third dataset includes one or more measurement information corresponding to one or more output data of the second coding model. For example, the second dataset is one or more output data of the second coding model, and the third dataset is one or more measurement information corresponding to one or more output data of the second coding model.

[0025] In this implementation, the network side processes the first dataset provided by the terminal side for self-model adjustment using a data distillation method. The network side then sends a generated dataset, obtained from the first dataset using the distillation method and the first dataset, to the terminal side for adjusting the terminal's coding model. When channel conditions are favorable, distilling the first dataset using the distillation method results in a significantly smaller generated dataset compared to the first dataset, thereby reducing the overhead of dataset transmission.

[0026] Secondly, a method for adjusting a model is provided, executed by a communication device or a module for the communication device (e.g., a processor, chip, circuit, AI entity, etc., or a logic module, hardware, and / or software capable of implementing all or part of the functions of the communication device), the communication device corresponding to the network device in the method embodiment. The method includes: receiving a first dataset including measurement information of a reference signal, the first dataset being used for adjusting a self-model, the self-model including a first encoding model and a first decoding model; and transmitting a second dataset related to a first processing method for the first dataset, the second dataset being used for adjusting a third encoding model to obtain a fourth encoding model, the fourth encoding model being matched with a second decoding model in the adjusted self-model.

[0027] In conjunction with the second aspect, in some implementations of the second aspect, a first message is sent, which indicates one or more of the following: a third dataset, or one or more reference signals corresponding to the third dataset, or a first processing method, or parameters related to the first processing method.

[0028] In conjunction with the second aspect, in some implementations of the second aspect, the first processing method includes a core set-based approach, which further includes sending a fourth message indicating the weights associated with the data in the third dataset when adjusting the model.

[0029] In conjunction with the second aspect, in some implementations of the second aspect, the method further includes: sending third information, the third information being used to indicate the first processing method.

[0030] In some implementations of the first or second aspect, the second dataset includes at least one or more output data of the second coding model, wherein the second coding model is a coding model obtained by adjusting the first coding model.

[0031] In some implementations of the first or second aspect, the first processing method includes one or more of the following: meta-learning, reinforcement learning, core set-based methods, data distillation methods, or augmented similarity subclass selection (SAS); one or more output data of the second encoding model include: output data of the second encoding model when a third dataset is used as input to the adjusted self-model, wherein the third dataset is obtained by processing the data in the first dataset based on the first processing method.

[0032] In some implementations of the first or second aspect, the first processing method includes a data distillation method; the third dataset includes a generated dataset of the first dataset, and one or more output data of the second coding model include: the output of the second coding model when the generated dataset is used as the input of the adjusted two-end model, wherein the second coding model is a coding model obtained by adjusting the first coding model.

[0033] Among these implementation methods, the first processing method is meta-learning, reinforcement learning, SAS, or data distillation. Since the dataset used for adjusting the encoding model on the terminal side is reduced, the computational power requirements for training the model on the terminal side are reduced, thus lowering the training cost.

[0034] In some implementations of the first or second aspect, the first processing method includes SAS, and the parameters related to the first processing method include: the augmentation method and / or screening ratio corresponding to SAS.

[0035] In some implementations of the first or second aspect, the first processing method includes a core set-based approach, and the parameters associated with the first processing method include: the weights associated with each data point in the third dataset when the model is adjusted.

[0036] In some implementations of the first or second aspect, the first information indicating the first processing method includes: the first information being used to indicate that the first dataset is not processed; the second dataset includes: the output data of the second encoding model when the first dataset is used as the input of the adjusted self-model, the second encoding model being an encoding model obtained by adjusting the first encoding model.

[0037] The second aspect is the network-side method corresponding to the first aspect. The beneficial technical effects of the second aspect and its various implementations can be found in the description of the technical effects of the first aspect or the corresponding implementations of the first aspect.

[0038] Thirdly, a method for adjusting a model is provided, executed by a communication device or a module for the communication device (e.g., a processor, chip, circuit, AI entity, etc., or a logic module, hardware, and / or software capable of implementing all or part of the functions of the communication device), the communication device corresponding to the terminal device in the method embodiment. The method includes: acquiring a first dataset, the first dataset including measurement information of a reference signal; and transmitting a dataset #1, the dataset #1 being related to a processing method #1 for the first dataset, the dataset #1 being used for adjusting a self-model, the self-model including a first encoding model and a first decoding model.

[0039] In this technical solution, after acquiring the first dataset, the terminal processes it using processing method #1 to obtain dataset #1, and then sends dataset #1 to the network side. Since the dataset #1 sent by the terminal is obtained by processing the first dataset using processing method #1, and the first dataset is real measurement information obtained by the terminal through measuring reference signals, similar to the effect of the first aspect, adjusting the self-model based on this measurement information dataset, and thus adjusting the encoding model on the terminal side, is beneficial for optimizing the encoding model on the terminal side. Furthermore, when processing method #1 is certain, sending dataset #1 from the terminal side for adjusting the self-model on the network side reduces data transmission overhead compared to the terminal side directly sending the first dataset.

[0040] In conjunction with the third aspect, in some implementations of the third aspect, the method further includes: processing the first dataset based on processing method #1 to obtain dataset #1.

[0041] In conjunction with the third aspect, in some implementations of the third aspect, the method further includes: receiving a dataset #2, the dataset #2 being used to adjust the third encoding model to obtain a fourth encoding model, the fourth encoding model being matched with the second decoding model in the adjusted self-model, the dataset #2 being the output data of the second encoding model when the dataset #1 is used as the input of the adjusted self-model, the dataset #1 being obtained by processing the data in the first dataset based on processing method #1, and the second encoding model being the encoding model obtained by adjusting the first encoding model.

[0042] In this implementation, the terminal processes the first collected dataset, sends the processed dataset #1 to the network side, and receives dataset #2 from the network side. Dataset #2 is used to adjust the encoding model on the terminal side.

[0043] In conjunction with the third aspect, in some implementations of the third aspect, processing method #1 is to select SAS for augmenting similarity subclasses. This method further includes sending second information, which is used to indicate the augmentation method and / or screening ratio corresponding to the SAS.

[0044] In this implementation, when the terminal side uses SAS to process the first dataset, it may also send the corresponding augmentation method and / or filtering ratio to the network side. The network side then uses the same augmentation method and filtering ratio to augment and reconstruct the dataset #1 provided by the terminal side, and then uses the augmented and reconstructed dataset to adjust the self-model. SAS is more suitable for processing datasets on the terminal side because the terminal side only needs to transmit the dataset #1 obtained by processing the first dataset consisting of measurement CSIs to the network side. In some implementations of the first aspect, if the network side uses SAS to filter the dataset sent to the terminal side for adjusting the encoding model, the network side, in addition to providing the filtered measurement CSI dataset to the terminal side, also needs to provide the output of the encoding model (second encoding model) in the adjusted self-model when the measurement CSI dataset is used as input, i.e., the dataset of feedback CSIs corresponding to the filtered measurement CSI dataset. In the pairing process of the adjusted model, compared with the network side using SAS to process the dataset, the terminal side using SAS to process the dataset results in lower data transmission overhead.

[0045] In conjunction with the third aspect, in some implementations of the third aspect, the method further includes: sending a third message, the third message indicating processing method #1.

[0046] In conjunction with the third aspect, in some implementations of the third aspect, the method further includes: receiving a dataset #3, the dataset #3 being used to adjust the third encoding model to obtain a fourth encoding model, the fourth encoding model being matched with the second decoding model in the adjusted self-model, the dataset #3 being the output of the second encoding model when the dataset #4 is used as the input of the adjusted self-model, the dataset #4 being determined based on the dataset #1 and the first processing method.

[0047] In this implementation, the terminal processes the collected first dataset (e.g., filtering or distilling) to obtain dataset #1, and provides dataset #1 to the network side. The network side can adjust its own model based on dataset #1 provided by the terminal side. The network device processes dataset #1 based on the first processing method and determines dataset #3 based on dataset #4 obtained from processing dataset #1. The network side sends dataset #3 to the terminal side and also instructs dataset #4 to be used for adjusting the encoding model on the terminal side. In this implementation, both the terminal side and the network side process the sent datasets simultaneously. Both the datasets sent from the terminal side and the datasets sent from the network side can reduce transmission overhead, thereby reducing the data transmission overhead in the pairing process of the adjusted model.

[0048] In conjunction with the third aspect, in some implementations of the third aspect, the first processing method includes one or more of the following: meta-learning, reinforcement learning, core set-based methods, data distillation methods, or augmented similarity subclass selection (SAS).

[0049] In conjunction with the third aspect, in some implementations of the third aspect, the method further includes: receiving information #1, the information #1 indicating one or more of the following: the dataset #4, or one or more reference signals corresponding to the dataset #4, or the first processing method, or parameters related to the first processing method.

[0050] In conjunction with the third aspect, in some implementations of the third aspect, the first processing method includes the SAS, and the method further includes: receiving information #2, the information #2 indicating the augmentation method and / or filtering ratio corresponding to the SAS.

[0051] In conjunction with the third aspect, in some implementations of the third aspect, the method further includes: receiving the data in the dataset #4 for the respective associated weights when adjusting the model.

[0052] In conjunction with the third aspect, in some implementations of the third aspect, the method further includes: receiving information #3, the information #3 being used to indicate the first processing method.

[0053] Fourthly, a method for adjusting a model is provided, executed by a communication device or a module for the communication device (e.g., a processor, chip, circuit, AI entity, etc., or a logic module, hardware, and / or software capable of implementing all or part of the functions of the communication device), the communication device corresponding to the network device in the method embodiment. The method includes: receiving a dataset #1, which is used for adjusting a self-model, the self-model including a first encoding model and a first decoding model, the dataset #1 relating to a processing method #1 for the first dataset, the first dataset including measurement information of a reference signal; and acquiring a dataset #2 or a dataset #3, which is used for adjusting a third encoding model to obtain a fourth encoding model, the fourth encoding model being matched with a second decoding model in the adjusted self-model, wherein the dataset #2 is determined based on the dataset #1 and the adjusted self-model, and the dataset #3 relating to the first processing method for the dataset #1.

[0054] In conjunction with the fourth aspect, in some implementations of the fourth aspect, when the dataset #2 is obtained, the method further includes: sending the dataset #2.

[0055] In conjunction with the fourth aspect, in some implementations of the fourth aspect, processing method #1 includes augmenting similarity subclass selection (SAS), and the method further includes receiving second information, which indicates the augmentation method and / or screening ratio corresponding to the SAS.

[0056] In conjunction with the fourth aspect, in some implementations of the fourth aspect, the method further includes: receiving third information, the third information indicating processing method #1.

[0057] In conjunction with the fourth aspect, in some implementations of the fourth aspect, when the dataset #3 is obtained, the method further includes: sending the dataset #3, which is the output of the second encoding model when the dataset #4 is used as the input of the adjusted self-model, and the dataset #4 is determined based on the dataset #1 and the first processing method.

[0058] In conjunction with the fourth aspect, in some implementations of the fourth aspect, the method further includes: processing the dataset #1 based on the first processing method to obtain the dataset #4, wherein the first processing method includes one or more of the following: meta-learning, reinforcement learning, core set-based methods, data distillation methods, or augmented similarity subclass selection (SAS).

[0059] In conjunction with the fourth aspect, in some implementations of the fourth aspect, the method further includes: sending information #1, which indicates one or more of the following: the dataset #4, or one or more reference signals corresponding to the dataset #4, or the first processing method, or parameters related to the first processing method.

[0060] In conjunction with the fourth aspect, in some implementations of the fourth aspect, the first processing method includes the SAS, and the method further includes: sending information #2, the information #2 indicating the augmentation method and / or filtering ratio corresponding to the SAS.

[0061] In conjunction with the fourth aspect, in some implementations of the fourth aspect, the first processing method includes the core set-based method, and the parameters related to the first processing method include: the weights associated with the data in dataset #4 when adjusting the model.

[0062] In conjunction with the fourth aspect, in some implementations of the fourth aspect, the method further includes: sending data from said dataset #4 for the respective associated weights during model adjustment.

[0063] In conjunction with the fourth aspect, in some implementations of the fourth aspect, the method further includes: sending information #3, which is used to indicate the first processing method.

[0064] In some implementations of the third or fourth aspect, processing method #1 includes augmented similarity subclass selection (SAS); dataset #1 includes: the dataset obtained after processing the first dataset based on processing method #1.

[0065] In some implementations of the third or fourth aspect, processing method #1 is SAS, and dataset #2 includes one of the following:

[0066] When dataset #1 is used as the input to the adjusted self-model, the output data of the second encoding model; or,

[0067] The data includes the output data of the second encoder when dataset #1 is used as the input of the adjusted self-model, the data obtained by augmenting and reconstructing the data in dataset #1 based on the augmentation method and screening ratio corresponding to SAS, and the output data of the second encoder when the data obtained by augmentation and reconstruction is used as the input of the adjusted self-model.

[0068] In some implementations of the third or fourth aspect, processing method #1 includes a data distillation method; dataset #1 includes: a generated dataset of the first dataset, which is the output of the distillation method when the distillation method is applied to the first dataset.

[0069] In some implementations of the third or fourth aspect, processing method #1 is a data distillation method, and dataset #2 includes: output data of the second encoding model when the generated dataset of the first dataset is used as the input of the adjusted self-model.

[0070] In some implementations of the third or fourth aspect, the third information indicates that the processing method #1 includes: the third information indicates that the first dataset is not processed; the dataset #1 includes the first dataset; and the dataset #2 includes the output data of the second encoding model when the dataset #1 is used as the input of the adjusted self-model, the second encoding model being an encoding model obtained by adjusting the first encoding model.

[0071] In some implementations of the third or fourth aspect, the first processing method includes the SAS, and the parameters related to the first processing method include the augmentation method and / or screening ratio corresponding to the SAS; or, the first processing method includes the core set-based method, and the parameters related to the first processing method include the weights associated with the data in dataset #4 when they are used for model adjustment.

[0072] The fourth aspect is the network-side method corresponding to the third aspect. The beneficial technical effects of the fourth aspect and its various implementations can be found in the description of the technical effects of the third aspect or its corresponding implementations.

[0073] Fifthly, a communication device is provided, the communication device having the function of implementing the methods of the first aspect or the third aspect, or any possible implementation of these aspects. The function can be implemented by hardware, by software, or by hardware executing corresponding software. The hardware or software includes one or more units corresponding to the above-described functions.

[0074] Sixthly, a communication device is provided, the communication device having the function of implementing the methods of the second or fourth aspect, or any possible implementation of these aspects. The function can be implemented by hardware, by software, or by hardware executing corresponding software. The hardware or software includes one or more units corresponding to the above-described functions.

[0075] A seventh aspect provides a communication device including at least one processor configured to cause the communication device to perform a method of the first aspect or the third aspect, or any possible implementation thereof; or to perform a method of the second aspect or the fourth aspect, or any possible implementation thereof. Optionally, the at least one processor is coupled to at least one memory for storing a computer program or instructions, the at least one processor being configured to call and execute the computer program or instructions from the at least one memory, causing the communication device to perform a method of the first aspect or the third aspect, or any possible implementation thereof; or to perform a method of the second aspect or the fourth aspect, or any possible implementation thereof. Optionally, the at least one processor may be included in the communication device or configured externally to the communication device. Optionally, the communication device further includes the at least one memory. Optionally, the communication device further includes a communication interface.

[0076] Eighthly, a communication device is provided, comprising a communication circuit and a processing circuit. The communication circuit is configured to receive a signal to be processed and transmit the signal to the processing circuit. The processing circuit is configured to process the signal to perform a method as described in the first or third aspect, or any possible implementation thereof; or to perform a method as described in the second or fourth aspect, or any possible implementation thereof. Optionally, the communication circuit is further configured to output the processed signal. As an example, the communication circuit may be a transceiver, hardware circuit, bus, module, pin, or other type of communication interface. The signal includes information and / or data. Optionally, the communication device may be a chip or a chip system.

[0077] A ninth aspect provides a computer-readable storage medium storing computer program code or instructions that, when executed on a computer, cause the method of the first or third aspect, or any possible implementation thereof, to be implemented; or the method of the second or fourth aspect, or any possible implementation thereof, to be implemented.

[0078] In a tenth aspect, a computer program product is provided, the computer program product comprising computer program code or instructions that, when executed on a computer, cause the method of the first aspect or the third aspect, or any possible implementation thereof, to be implemented; or, as in the second aspect or the fourth aspect, or any possible implementation thereof, to be implemented.

[0079] Eleventh aspect: A wireless communication system is provided, including the communication device as described in the fifth aspect and the communication device as described in the sixth aspect. Attached Figure Description

[0080] Figure 1 is a schematic diagram of a communication system applicable to an embodiment of this application.

[0081] Figure 2 is another schematic diagram of a communication system applicable to an embodiment of this application.

[0082] Figure 3 is a schematic diagram of a possible application framework in a communication system.

[0083] Figure 4 is a schematic diagram of another possible application framework in a communication system.

[0084] Figure 5 is a schematic diagram of AI-CSI feedback based on the AE model.

[0085] Figure 6 is a schematic flowchart of two-end model pairing based on dataset transmission.

[0086] Figure 7 is a schematic diagram of fine-tuning of the two-end model based on dataset transmission.

[0087] Figure 8 is a schematic flowchart of the method 200 for adjusting the model provided in this application.

[0088] Figure 9 is a schematic diagram of the adjustment and matching of the dual-end model based on network-side data processing according to Scheme 1 of this application.

[0089] Figure 10 is a schematic flowchart of the method 500 for adjusting the model provided in this application.

[0090] Figure 11 is a schematic diagram of the adjustment and matching of the dual-end model based on terminal-side data processing according to Scheme 2 of this application.

[0091] Figure 12 is a schematic flowchart of the method 800 for adjusting the model provided in this application.

[0092] Figure 13 is a schematic block diagram of the communication device 1000 provided in this application.

[0093] Figure 14 is a schematic block diagram of another communication device 1100 provided in this application.

[0094] Figure 15 is a schematic structural diagram of the chip provided in this application.

[0095] Figure 16 is a schematic diagram of the system architecture of the communication device provided in this application. Detailed Implementation

[0096] The technical solutions in this application will now be described with reference to the accompanying drawings.

[0097] The technical solutions provided in this application can be applied to various communication systems, such as 5th generation (5G) or new radio (NR) systems, long term evolution (LTE) systems, LTE frequency division duplex (FDD) systems, LTE time division duplex (TDD) systems, wireless local area network (WLAN) systems, and satellite communication systems. Furthermore, they can be applied to device-to-device (D2D) communication, vehicle-to-everything (V2X) communication, machine-to-machine (M2M) communication, machine-type communication (MTC), as well as Internet of Things (IoT) communication systems, future communication systems, or integrated systems of multiple systems.

[0098] In a communication system, one network element can send signals to or receive signals from another network element. These signals can include information, signaling, or data. A network element can also be replaced by an entity, network entity, device, communication device, communication module, node, or communication node; this embodiment uses a device as an example. For instance, a communication system can include at least one terminal device and at least one network device. The network device can send downlink signals to the terminal device, and / or the terminal device can send uplink signals to the network device.

[0099] Figure 1 is a schematic diagram of a communication system applicable to an embodiment of this application. As shown in Figure 1, the communication system 100 may include at least one network device, such as network device 110 shown in Figure 1; the communication system 100 may also include at least one terminal device, such as terminal device 120 and terminal device 130 shown in Figure 1. Network device 110 and terminal devices (such as terminal devices 120 and 130) can communicate via a wireless link. The communication devices in this communication system, for example, network device 110 and terminal device 120, can communicate via multi-antenna technology.

[0100] Optionally, the communication system may also include at least one AI node.

[0101] Figure 2 is another schematic diagram of a communication system applicable to embodiments of this application. Compared to the communication system 100 shown in Figure 1, the communication system 100 shown in Figure 2 further includes an AI node 140. The AI ​​node 140 is used to perform AI-related operations, such as building training datasets, training or inferring AI models, etc.

[0102] In one implementation, network device 110 can send data related to AI model training to AI node 140, whereby AI node 140 constructs a training dataset and trains the AI ​​model. As an example, the data related to AI model training may include data reported by terminal devices. AI node 140 can send the results of AI model-related operations to network device 110, which then forwards them to the terminal devices. For example, the results of AI model-related operations may include at least one of the following: a trained AI model, model evaluation results, or test results. Exemplarily, a portion of the trained AI model may be deployed on network device 110, and another portion on the terminal devices. Optionally, the trained AI model may be deployed on network device 110, or it may be deployed on the terminal devices.

[0103] It should be understood that Figure 2 is only used as an example of AI node 140 being directly connected to network device 110. In other scenarios, AI node 140 can also be connected to terminal device. Alternatively, AI node 140 can be connected to both network device 110 and terminal device simultaneously. Alternatively, AI node 140 can also be connected to one or more of network device 110 and terminal device through a third-party network element. This application embodiment does not limit the connection relationship between AI network element and other network elements.

[0104] Alternatively, in another implementation, the AI ​​node 140 can also be configured as a module in a network device and / or a terminal device, for example, in the network device 110, terminal device 120, or terminal device 130 shown in FIG1.

[0105] It should be noted that Figures 1 and 2 are schematic diagrams for ease of understanding only. The communication system may also include other devices, such as wireless relay devices and / or wireless backhaul devices, which are not shown in Figures 1 and 2. Furthermore, in practical applications, the communication system may include multiple network devices or multiple terminal devices. This application does not limit the number of network devices and terminal devices.

[0106] In the embodiments of this application, the terminal device may also be referred to as user equipment (UE), access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent, or user apparatus. The terminal device can be a device that provides voice / data, such as a handheld device or vehicle-mounted device with wireless connectivity. Currently, examples of terminals include: mobile phones, tablets, laptops, PDAs, mobile internet devices (MIDs), wearable devices, virtual reality (VR) devices, augmented reality (AR) devices, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in remote medical surgery, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, wireless terminals in smart homes, cellular phones, cordless phones, session initiation protocol (SIP) phones, wireless local loop (WLL) stations, personal digital assistants (PDAs), handheld devices with wireless communication capabilities, computing devices or other processing devices connected to wireless modems, wearable devices, terminal devices in 5G networks, or terminal devices in future communication systems, etc., and this application does not limit these examples.

[0107] In this embodiment, the device used to implement the functions of the terminal device can be the terminal device itself, or any device capable of supporting the terminal device in implementing corresponding functions, such as a processor, circuit, or chip. This device can be configured in the terminal device or used in conjunction with the terminal device. In this embodiment, the terminal device is used as an example to illustrate the function of the terminal device, and this does not constitute a limitation on the solution of this embodiment.

[0108] The network devices in this application embodiment may include radio access network (RAN) nodes that connect terminal devices to wireless networks, such as base stations. Base stations can broadly encompass various names as follows, or be replaced by the following names: NodeB, evolved NodeB (eNB), next-generation NodeB (gNB), relay station, access point, transmitting and receiving point (TRP), transmitting point (TP), master station, auxiliary station, motor slide retainer (MSR) node, home base station, network controller, access node, wireless node, access point (AP), transmission node, transceiver node, baseband unit (BBU), remote radio unit (RRU), active antenna unit (AAU), remote radio head (RRH), central unit (CU), distributed unit (DU), radio unit (RU), etc. A base station can be a macro base station, micro base station, relay node, donor node, or a combination thereof. A base station can also refer to a communication module, modem, or chip installed within the aforementioned equipment or apparatus. A base station can also be a mobile switching center, equipment performing base station functions in D2D, V2X, and M2M communications, or equipment performing base station functions in future communication systems. A base station can support networks using the same or different access technologies. Optionally, a RAN node can also be a server, wearable device, vehicle, or in-vehicle equipment. For example, the access network equipment in vehicle-to-everything (V2X) technology can be a roadside unit (RSU). The embodiments of this application do not limit the specific technologies or equipment forms used in the network equipment.

[0109] Base stations can be fixed or mobile. For example, a helicopter or drone can be configured to act as a mobile base station, and one or more cells can move depending on the location of the mobile base station. In other examples, a helicopter or drone can be configured as a device to communicate with another base station.

[0110] In some deployments, the network devices mentioned in the embodiments of this application may be devices including CU, DU, or CU and DU, or devices with control plane CU nodes (central unit-control plane (CU-CP)) and user plane CU nodes (central unit-user plane (CU-UP)) and DU nodes. For example, the network devices may include gNB-CU-CP, gNB-CU-UP, and gNB-DU.

[0111] In some deployments, multiple RAN nodes collaborate to assist terminals in achieving wireless access, with different RAN nodes each implementing some of the base station's functions. For example, RAN nodes can be CUs, DUs, CU-CPs, CU-UPs, or RUs. CUs and DUs can be configured separately or included in the same network element, such as a BBU. RUs can be included in radio frequency equipment or radio frequency units, such as RRUs, AAUs, or RRHs.

[0112] In one possible design, the processing unit in the BBU used to implement baseband functions is called the baseband high (BBH) unit, and the processing unit in the RRU / AAU / RRH used to implement baseband functions is called the baseband low (BBL) unit.

[0113] In different systems, CU (or CU-CP and CU-UP), DU, or RU may have different names, but those skilled in the art will understand their meaning. For example, in an open radio access network (ORAN / O-RAN) system, CU can also be called O-CU (open CU), DU can also be called O-DU, CU-CP can also be called O-CU-CP, CU-UP can also be called O-CU-UP, and RU can also be called O-RU. Any of the units among CU (or CU-CP, CU-UP), DU, and RU in this application can be implemented through software modules, hardware modules, or a combination of software modules and hardware modules.

[0114] In this embodiment, the device used to implement the functions of the network device can be a network device itself; it can also be a device capable of supporting the network device in implementing corresponding functions, such as a processor, circuit, or chip. This device can be configured within the network device or used in conjunction with the network device. In this embodiment, the network device is used as an example to illustrate the function of the network device, and this does not constitute a limitation on the solutions described in this embodiment.

[0115] Network devices and / or terminal devices can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; they can also be deployed on water; and they can also be deployed in the air on airplanes, balloons, and satellites. This application does not limit the scenario in which the network devices and terminal devices are located. Furthermore, terminal devices and network devices can be hardware devices, or software functions running on dedicated hardware or general-purpose hardware, such as virtualization functions instantiated on a platform (e.g., a cloud platform), or entities that include dedicated or general-purpose hardware devices and software functions. This application does not limit the specific form of the terminal devices and network devices.

[0116] Optionally, the AI ​​node can be deployed in one or more of the following locations within the communication system: access network equipment, terminal equipment, or core network equipment, etc. Alternatively, the AI ​​node can be deployed independently, for example, in a location other than any of the aforementioned devices, such as in the host or cloud server of an over-the-top (OTT) system. The AI ​​node can communicate with other devices in the communication system, which can be, for example, one or more of the following: access network equipment, terminal equipment, or core network equipment, etc.

[0117] This application does not limit the number of AI nodes. For example, when there are multiple AI nodes, they can be divided based on function, such as different AI nodes being responsible for different functions.

[0118] Optionally, AI nodes can be independent devices, integrated into the same device to implement different functions, or they can be network elements in hardware devices, software functions running on dedicated hardware, or virtualization functions instantiated on a platform (e.g., a cloud platform). This application does not limit the specific form of the AI ​​nodes described above. AI nodes can also be called AI network elements or AI modules.

[0119] Figure 3 illustrates a possible application framework in a communication system. As shown in Figure 3, network elements in the communication system are connected via interfaces (e.g., NG, Xn) or air interfaces. These network elements, such as core network equipment, access network equipment (RAN nodes), terminals, or one or more devices in the operation administration and maintenance (OAM) system, are equipped with one or more AI modules. Access network equipment can be a single RAN node or can include multiple RAN devices, such as CUs and DUs. The CUs and / or DUs can also be equipped with one or more AI modules. Optionally, the CU can be further divided into CU-CP and CU-UP. One or more AI models are configured in the CU-CP and / or CU-UP.

[0120] The AI ​​module is used to implement corresponding AI functions. AI modules deployed in different network elements can be the same or different. Depending on the parameter configuration, the AI ​​module can implement different functions. The AI ​​module model can be configured based on one or more of the following parameters: structural parameters (e.g., at least one of the following: number of neural network layers, neural network width, inter-layer connections, neuron weights, neuron activation function, or bias in the activation function), input parameters (e.g., type and / or dimension of input parameters), or output parameters (e.g., type and / or dimension of output parameters). The bias in the activation function can also be referred to as the neural network bias.

[0121] An AI module can have one or more models. A model can infer an output, which includes one or more parameters. The learning, training, or inference processes of different models can be deployed on different nodes or devices, or they can be deployed on the same node or device.

[0122] The network device can be a network device equipped with one or more AI modules. For example, the network device can be one or more devices in the core network, access network, or OAM as shown in Figure 3. The AI ​​module can be the RAN intelligent controller (RIC) shown in Figure 4, such as a near real-time RIC or a non-real-time RIC. For example, a near real-time RIC is set in a RAN node (e.g., in a CU or DU), while a non-real-time RIC is set in the OAM, cloud server, core network device, or other network device.

[0123] Figure 4 illustrates another possible application framework in a communication system. As shown in Figure 4, the communication system includes a Resource Interchange (RIC). For example, the RIC could be the AI ​​module in the RAN device shown in Figure 1, used to implement AI-related functions. RICs include near-real-time RICs (near-RT RICs) and non-real-time RICs (non-RT RICs). Non-real-time RICs primarily process non-real-time information, such as data that is not sensitive to latency, with latency in the order of seconds. Real-time RICs primarily process near-real-time information, such as data that is relatively sensitive to latency, with latency in the order of tens of milliseconds.

[0124] Near real-time RICs are used for model training and inference. For example, they are used to train AI models and then use those models for inference. Near real-time RICs can obtain network-side and / or terminal-side information from RAN devices (e.g., CU, CU-CP, CU-UP, DU, and / or RU) and / or terminal devices. This information can be used as training data or as data for inference.

[0125] Optionally, near real-time RIC can deliver inference results to RAN devices and / or terminal devices.

[0126] Optionally, inference results can be exchanged between CU and DU, and / or between DU and RU. For example, near real-time RIC submits inference results to DU, and DU sends them to RU.

[0127] Non-real-time RICs are also used for model training and inference. For example, they can be used to train AI models and then use those models for inference. Non-real-time RICs can obtain network-side and / or terminal-side information from RAN devices (e.g., CU, CU-CP, CU-UP, DU, and / or RU) and / or terminals. This information can be used as training data or as inference data, and the inference results can be delivered to RAN nodes and / or terminals.

[0128] Optionally, inference results can be exchanged between CU and DU, and / or between DU and RU. For example, a non-real-time RIC can submit inference results to DU, which in turn can send them to RU.

[0129] Near real-time RICs and non-real-time RICs can also be configured as separate devices. Alternatively, near real-time RICs and non-real-time RICs can also be part of other devices. For example, near real-time RICs can be configured in RAN nodes (e.g., CU, DU), while non-real-time RICs can be configured in OAM, cloud servers, core network devices, or other devices.

[0130] Optionally, the AI ​​model can be implemented as hardware circuitry, software, or a combination of both, without limitation. Non-limiting examples of software include: program code, program, subroutine, instruction, instruction set, code, code segment, software module, application program, or software application, etc.

[0131] The following section introduces some related technologies involved in the technical solution of this application.

[0132] 1. AI-CSI Feedback

[0133] In existing LTE, NR, and other communication systems, base stations acquire channel state information (CSI) to determine one or more of the following configurations for scheduling the downlink data channel of the UE: resources, modulation and coding scheme (MCS), and precoding. In TDD systems, due to the reciprocity of uplink and downlink channels, the base station can obtain the uplink CSI by measuring the uplink reference signal and then infer a more accurate downlink CSI, for example, using the uplink CSI as the downlink CSI. In FDD systems, uplink and downlink reciprocity cannot be guaranteed. The downlink CSI is obtained by the UE measuring the downlink reference signal, such as the CSI-RS or synchronization signal block (SSB) (i.e., the synchronization signal / physical broadcast signal block, SS / PBCH block). Therefore, the UE needs to generate a CSI report according to the protocol predefined method or the base station configuration and feed the CSI back to the base station so that it can obtain the downlink CSI.

[0134] In AI-based CSI (Computer-Assisted Instruction) feedback, when the model is deployed at the base station, the base station obtains the CSI-RS estimation results at the UE side as labels for model training. The AE (Auto-Edge) model consists of two sub-models: an encoder and a decoder. The encoder and decoder of the AE are usually trained together and can be used in a mutually compatible manner; they can also be called self-models. CSI feedback can be implemented based on the AE model.

[0135] Figure 5 illustrates the AI-CSI feedback based on the AE model. As shown in Figure 5, the UE side compresses and quantizes the CSI using an encoder (i.e., the encoding model), while the base station recovers the CSI using a decoder (i.e., the decoding model). For the base station, the model's input is the fed-out CSI, and its output is the recovered CSI; the model's training uses the CSI measured by the UE side as the label (or ground truth) of the recovered CSI.

[0136] 2. Pairing of two-end models based on dataset transmission

[0137] Figure 6 is a schematic flowchart of the pairing of two-end models based on dataset transmission. As shown in Figure 6, in stage 1, the base station locally trains an encoder and decoder based on historically collected CSI data. This encoder and decoder constitute the network-side self-model. To distribute the encoder from the trained self-model to the UE to achieve pairing of the two-end models on the terminal and network sides, the base station constructs a dataset based on the input (i.e., measured CSI) and output (i.e., feedback CSI) of the trained encoder and transmits this dataset to the UE side. In stage 2, the UE side uses the received dataset to train its own encoder sub-model, enabling pairing of the two-end models on the base station and UE sides. The two-end model can also be called a bilateral model or a collaborative model, etc.

[0138] 3. Adjustments to the two-end model based on dataset transmission

[0139] In the aforementioned techniques, the dual-end model is trained based on the base station-side data distribution. When the UE-side data distribution is inconsistent with the base station-side data distribution—for example, if the UE-side data distribution is denoted as p1 and the base station-side data distribution as p2, as shown in Figure 6—the dataset received by the UE for encoder training is based on the base station-side data distribution. Therefore, the trained model is not optimal for the UE side, and the model's compression and recovery performance will be compromised. Consequently, the dual-end model can be further adjusted based on the UE-side data distribution. The adjustment process for the dual-end model based on the UE-side data distribution is shown in Figure 7.

[0140] Figure 7 illustrates the adjustment of the dual-end model based on dataset transmission. As shown in Figure 7, the self-model trained by the base station on data distribution p2 includes an encoder and a decoder, denoted as the basic encoder and basic decoder, respectively. The encoder sub-model trained by the UE on the dataset provided by the base station is called the basic encoder. When adjusting the dual-end model to adapt to the data portion p1 on the UE side, optionally, the UE sends a data collection request to the base station to request the collection of measurement information for CSI-RS used for model adjustment. The base station issues CSI-RS according to the request, and the UE collects local measurement CSIs according to the CSI-RS issued by the base station and constructs a dataset to transmit to the base station. The base station updates the dual-end model based on the newly collected measurement CSIs by the UE to adapt to the data distribution on the UE side. After the dual-end model on the base station side is adjusted, the encoder and decoder are re-paired on the base station side and the UE side. The adjusted dual-end model on the base station side includes an adjusted encoder and an adjusted decoder, and the basic encoder on the UE side, after being paired with the adjusted dual-end model on the base station side, is also called the adjusted encoder. It should be understood that, in Figure 7, the adjustment of the dual-end model actually refers to the further training of the dual-end model based on newly collected measurement CSI from the UE side after obtaining the basic dual-end model. This can also be called optimization, updating, or training. The term "adjustment of the dual-end model" is relative to the "basic dual-end model." Similarly, the "adjusted encoder" or "adjusted decoder" in the adjusted dual-end model is also relative to the "basic encoder" or "basic decoder" before the adjustment.

[0141] During the aforementioned model adjustment process, the UE sends the entire amount of newly collected data to the base station for the base station to adjust the dual-end model, resulting in high transmission overhead. Furthermore, after the base station completes its self-model adjustment, the pairing of the adjusted models still follows the transmission process of the basic model. That is, it uses the input and output data of the encoded model in the adjusted self-model to construct a dataset, and then transmits the constructed dataset to the UE for adjusting the encoded model deployed on the UE side, resulting in high dataset transmission overhead.

[0142] Therefore, this application provides a model tuning method that helps optimize the performance of the encoding model on the terminal side. Furthermore, in some implementations, data transmission overhead can be reduced during the two-way model tuning process based on dataset transmission.

[0143] First, the technical solution of this application will be summarized as follows:

[0144] (1) One approach is for the network device to adjust its own model based on the dataset provided by the terminal side. After adjusting the self-model, the network device processes the dataset provided by the terminal device and returns different datasets to the terminal device based on the different processing methods, which are used to adjust the encoding model on the terminal side. Finally, the adjusted encoding model on the terminal side and the adjusted self-model on the network side can be matched (or aligned or paired, etc.), which is referred to as Scheme 1 below.

[0145] (2) Another approach is that after collecting the dataset for adjusting the self-model, the terminal device first processes the dataset, and then provides the processed dataset to the network device based on the different processing methods. This is beneficial for optimizing the performance of the encoding model on the terminal side. In addition, it also helps to reduce the transmission overhead of the dataset. The network device adjusts the self-model based on the dataset provided by the terminal device, and after completing the adjustment, returns the data for adjusting the encoding model on the terminal side to the terminal device. The terminal device adjusts the encoding model based on the data returned by the network device, realizing the matching between the encoding model on the terminal side and the self-model on the network side. This is referred to as Scheme 2.

[0146] It is evident that Scheme 1 helps reduce the transmission overhead of the dataset used for terminal-side model adjustment when the network side transmits it to the terminal device; Scheme 2 helps reduce the transmission overhead of the dataset used for self-model adjustment when the terminal device transmits it to the network device.

[0147] (3) Another approach is to: process the collected dataset on the terminal side for the first time and provide the processed dataset to the network side. The network side adjusts the self-model based on the dataset provided by the terminal side. When providing the dataset for adjusting the encoding model on the terminal side to the terminal device, the network side processes the dataset provided by the terminal side for the second time and sends the output of the encoding model, which is used as the input of the encoding model in the adjusted self-model, to the terminal device for adjusting the encoding model on the terminal side. This approach helps reduce the transmission overhead when the terminal side uploads data for self-model adjustment. On the other hand, it also helps reduce the transmission overhead when the network side provides the data for adjusting the encoder sub-model on the terminal side after completing the self-model adjustment. This is referred to as Approach 3.

[0148] In the embodiments of this application, the adjustment of the model can also be replaced by the description of model updating, optimization, training or fine-tuning, etc.

[0149] The above solutions will be described in detail below.

[0150] Option 1

[0151] Figure 8 is a schematic flowchart of the method 200 for adjusting a model provided in this application. Method 200 involves network devices and terminal devices, and can be implemented by each of the network devices and terminal devices performing corresponding steps. Optionally, one or more steps in method 200 performed by communication devices (e.g., network devices or terminal devices) can be replaced by a device (e.g., referred to as a first device) for these communication devices. The first device can be a chip, processor, circuit, or AI entity serving the communication device, etc., applied to the communication device. The AI ​​entity can be deployed on the communication device or outside the communication device. If the first device is deployed outside the communication device, air interface interaction between the first device and the communication device may also be involved, such as data exchange between the first device and the communication device, or information exchange related to model training and adjustment, etc., to facilitate the first device to perform relevant processing based on the data set and / or information provided by the communication device to complete the method for adjusting the model provided in this application. This description applies to any embodiment of this application, and will not be repeated below. As an example, when the communication device is a terminal device, the aforementioned AI entity can be a host or cloud server of an over-the-top (OTT) system; when the communication device is a network device, the aforementioned AI entity can be a deployment device for a network-side AI model, such as a near real-time RIC, collectively referred to as an intelligent network element. Therefore, the deployment of the network-side or terminal-side AI model may not be within the same physical entity as the network device or terminal device. The following embodiments use network devices and terminal devices as examples for description.

[0152] 210. The terminal device sends the first dataset.

[0153] The first dataset includes measurement information of the reference signal and is used for adjusting the self-model, which includes a first encoding model and a first decoding model.

[0154] The network device receives the first dataset from the terminal device.

[0155] Optionally, the reference signal may include any one or more of uplink reference signals, downlink reference signals, or known signals. The reference signal, also known as a pilot signal or pilot, is a known signal, such as one provided by the transmitter to the receiver for channel estimation, channel sounding, or data demodulation. The reference signal mainly includes the synchronization signal block (SSB) (also known as the synchronization signal / physical broadcast signal block, SS / PBCH block) and the channel state information-reference signal (CSI-RS). The CSI-RS is a cell broadcast signal, including the primary synchronization signal (PSS), secondary synchronization signal (SSS), physical broadcast channel (PBCH), and demodulation reference signal (DMRS). There can be multiple types of reference signals; as standards evolve, the names of the above reference signals may change, and more reference signals may appear. No specific limitations are made regarding this.

[0156] As an example, the measurement information of the reference signal included in the first dataset can be CSI. The meaning of CSI in this application is broader than that of traditional CSI, including but not limited to channel quality indication (CQI), precoding matrix indicator (PMI), rank indicator (RI), CSI-RS resource indicator (CRI), and may also include one or more of the following: channel response information (e.g., channel response matrix, frequency domain channel response information, time domain channel response information), weight information corresponding to the channel response, precoding matrix information corresponding to the channel response, reference signal receiving power (RSRP), reference signal receiving quality (RSRQ), or signal to interference plus noise ratio (SINR), etc.

[0157] In the two-ended model shown in Figure 5, the measured CSI refers to the CSI obtained by measuring the reference signal. It can also be called the channel measurement information, the measured channel information, the reference signal measurement information, etc.

[0158] CSI feedback, also known as CSI feedback information, includes CSI feedback information, feedback information of channel measurement results, feedback information of channel information, compressed CSI information, compressed information, compressed channel information, compressed channel information, or compressed CSI, etc.

[0159] The recovered CSI can also be called recovered CSI, reconstructed CSI, reconstructed channel information, decompressed CSI, decompressed channel information, etc.

[0160] 220. The network device sends the second dataset.

[0161] Accordingly, the terminal device receives the second dataset.

[0162] The second dataset is related to the first processing method, which includes processing methods applied to the first dataset. The second dataset is used to adjust the third encoding model to obtain a fourth encoding model. The fourth encoding model matches the second decoding model in the adjusted dual-end model. The second dataset includes at least one or more output data from the second encoding model. The second encoding model is the encoding model in the adjusted self-model, i.e., the encoding model obtained by adjusting the first encoding model.

[0163] As an example, the first processing method may include one or more of the following: meta-learning, reinforcement learning, core set-based methods, data distillation methods, or augmented similarity subclass selection (SAS). The data contained in the second dataset is related to the first processing method.

[0164] In one implementation, when the first processing method includes one or more of meta-learning, reinforcement learning, or subsets maximizing augmentation similarity (SAS), and the second dataset includes one or more output data of the second encoding model, wherein the second dataset is the output data of the second encoding model when the third dataset is used as the input of the adjusted dual-end model, and the third dataset is obtained by processing the data in the first dataset based on the first processing method. Meta-learning or reinforcement learning is a method based on replay sets.

[0165] In another implementation, the first processing method includes a data distillation method, where the second dataset includes a generated dataset of the first dataset and the output data of the second encoding model when the generated dataset is used as input to the adjusted two-end model. Exemplarily, the distillation method may include dataset condensation (DC), dataset distillation by representative matching (DREAM), etc.

[0166] The data distillation method refers to extracting a small synthetic dataset from a large real dataset so that a model trained on the small dataset can achieve similar performance to a model trained on the original large dataset.

[0167] In another implementation, the first processing method includes a coreset-based approach, such as gradient matching.

[0168] The core-set-based approach can involve selecting a weighted subset of samples from the original dataset for a given model, such that the weighted gradient of the model on this subset approximates the full gradient on the original dataset. Similar to the replay set, the core set consists of selected real samples (not generated samples). The difference is that each selected sample is associated with a weight. Subsequently, the terminal adjusts the encoding model based on these weights. Because the core-set approach relies on the complete dual-end model, it can be applied to the network side for processing the first dataset.

[0169] When the first processing method is a core set-based method, the second dataset includes: one or more output data of the second encoding model, wherein the second dataset is the output data of the second encoding model when the third dataset is used as the input of the adjusted dual-end model, and the third dataset is obtained by processing the data in the first dataset based on the first processing method (specifically, the core set-based method).

[0170] As an example, network devices can determine the first processing method based on the following principles: when transmission overhead is limited and / or channel conditions are poor, processing methods such as meta-learning, reinforcement learning, or SAS are used; when available transmission capacity is large and / or channel conditions are good, data distillation methods are used.

[0171] For example, meta-learning enables models to quickly adapt to and learn new task scenarios while maintaining the performance level of the original scenario. Replay techniques in this type of method retain knowledge from the current task by filtering a subset of historical data, thus serving as a means of compressing current scenario data. A representative method is incremental classifier and representation learning (iCaRL).

[0172] For example, data filtering methods based on reinforcement learning can achieve sample value assessment by constructing a data value evaluator that learns the probability of each sample being used in the trained model; based on sample value, effective data filtering can be performed. A representative method is data valuation using reinforcement learning (DVRL).

[0173] For example, the SAS method divides the dataset into several augmented similarity subclasses by using predefined data augmentation operators, and selects representative meta-samples in each similarity subclass to characterize the data features of that subclass, thereby compressing the dataset. By performing corresponding augmentation operations on the representative meta-samples, the subclasses can be approximately restored, and the original dataset can be reconstructed.

[0174] In addition, optionally, the network side is not limited to using one of the first processing methods mentioned above, but can also use a combination of multiple processing methods. For example, samples that have a greater impact on the model weight update can be screened first based on meta-learning, reinforcement learning, core set-based methods, or SAS, and then the data distillation method can be used to further compress the size of the dataset.

[0175] Optionally, method 200 may include step 230. Step 230 may precede step 220.

[0176] 230. Network devices acquire the second dataset.

[0177] As mentioned above, the second dataset is related to the first processing method. Specifically, the second dataset is determined based on the third dataset and the first processing method. Specifically, the first dataset is processed using the first processing method to obtain the third dataset; when the third dataset is used as input to the adjusted self-model, the output of the second encoding model is the second dataset. As an example, the third dataset can be obtained by processing the first dataset using the first processing method from a network device or an AI entity serving the network device. This AI entity (e.g., a smart network element) can be deployed on the network device or outside of the network device. The third dataset includes one or more measurement information corresponding to the second dataset.

[0178] In one implementation, the first processing method includes meta-learning, reinforcement learning, or augmented similarity subclass selection (SAS), and the third dataset includes a dataset obtained by processing the first dataset based on the first processing method; in another implementation, the first processing method includes a data distillation method, in which case the third dataset includes a generated dataset of the first dataset.

[0179] When the AI ​​entity is deployed outside the network device, and the third dataset is obtained by the network device processing the first dataset based on the first processing method, the network device further sends the third dataset to the AI ​​entity, and the AI ​​entity uses the third dataset as the input of the adjusted self-model to obtain the output of the second encoding model (i.e., the adjusted first encoding model) in the self-model (i.e., the second dataset), and then sends the second dataset to the network device.

[0180] Alternatively, in one implementation, method 200 further includes step 240.

[0181] 240. The network device sends first information, and the terminal device receives the first information. Alternatively, the terminal device determines a third dataset based on a predefined first processing method and / or parameters related to the first processing method.

[0182] The first information indicates one or more of the following: a third dataset, or one or more reference signals corresponding to the third dataset, or a first processing method, or parameters related to the first processing method.

[0183] Optionally, one or more reference signals corresponding to the third dataset are also one or more reference signals corresponding to one or more output data of the second coding model.

[0184] As an example, if the first processing method is meta-learning or reinforcement learning: based on the contribution of samples to the parameter updates of the encoded model during model adjustment (such as the gradient magnitude of the corresponding samples), a small number of samples are selected to construct a replay set, so that the model adjusted on the replay set based on the basic encoder can achieve similar performance to the model adjusted on the full sample dataset. Here, the small number of samples represents the measurement information corresponding to the one or more reference signals. Alternatively, if the first processing method is SAS: based on several predefined augmentation methods (such as rotation, translation, or unitary transformation), the first dataset is divided into several augmented equivalent subclasses, and a small number of representative samples (i.e., the measurement information corresponding to the one or more reference signals) are taken from each subclass to construct a small dataset. Alternatively, if the first processing method is a core set-based method, the network can select a weighted sample subset (i.e., the third dataset) from the first dataset, with each sample in the subset associated with a weight. This sample subset includes one or more samples, i.e., the measurement information corresponding to the one or more reference signals.

[0185] In an implementation where the first information indicates one or more reference signals corresponding to a third dataset, as an example, the first information is used to indicate the index of the one or more reference signals.

[0186] As another example, the first information may include a bitmap comprising multiple bits, each corresponding to a reference signal. Based on this bitmap, one or more reference signals can be selected from all the reference signals.

[0187] It should be understood that the one or more reference signals correspond to one or more output data of the second coding model, or in other words, the second dataset is the one or more output data of the second coding model when the measurement information of the one or more reference signals is used as the input of the adjusted dual-end model. The measurement information corresponding to the one or more reference signals is obtained by the network side processing the first dataset based on the first processing method, that is, the third dataset.

[0188] Optionally, in step 240, when the network device indicates one or more reference signals corresponding to the third dataset to the terminal device through the first information, the data transmission overhead can be reduced compared to the network side sending the third dataset to the terminal side. The implementation of the network side sending the third dataset to the terminal side can be applied when the terminal side does not store the measurement information of the reference signals.

[0189] In another implementation, the first processing method includes a data distillation method.

[0190] In this implementation, the second dataset includes a generated dataset from the first dataset and the output of the second encoding model when the generated dataset is used as input to the adjusted two-end model. The generated dataset is the output of the distillation method when applied to the first dataset. In other words, the input to the distillation method is the first dataset, and the output is the generated dataset from the first dataset. The generated dataset reflects the properties of the first dataset and can be directly used for model adjustment.

[0191] In this embodiment, the dual-end model is deployed on the network device side and includes a first encoding model and a first decoding model. The adjusted self-model includes a second encoding model and a second decoding model. The encoding model deployed on the terminal side is called the third encoding model. Before the model is adjusted, the third encoding model and the first decoding model are matched. The terminal device adjusts the third encoding model based on the second dataset and the third dataset to obtain a fourth encoding model, which is matched with the second decoding model.

[0192] After the terminal device obtains the second dataset, it adjusts the third encoding model based on the second dataset to obtain the fourth encoding model. Optionally, the adjustment of the third encoding model can be performed by the terminal device or by an AI entity serving the terminal device, such as an OTT server or a cloud server. The AI ​​entity can be deployed on the terminal device or outside the terminal device, without limitation. Optionally, method 200 also includes step 250.

[0193] 250. The terminal device adjusts the third encoding model based on the second dataset to obtain the fourth encoding model.

[0194] As described above, when the first processing method is meta-learning, reinforcement learning, or SAS, the second dataset contains one or more output data of the second encoding model. In this case, the terminal device adjusts the third encoding model using the one or more output data of the second encoding model and one or more measurement information corresponding to those output data. There is a one-to-one correspondence between the one or more output data of the second encoding model and one or more measurement information; in other words, each output data of the one or more output data of the second encoding model corresponds to one measurement information. The one or more output data of the second encoding model serves as the model's label (or ground truth), while the one or more measurement information serves as the input to the third encoding model. After adjustment, a fourth encoding model is obtained.

[0195] Optionally, when the first processing method is SAS, the terminal device can perform augmented reconstruction on the received second dataset and one or more measurement information corresponding to the second dataset to obtain augmented and reconstructed data. That is, augmented reconstruction is performed on one or more output data of the received second coding model and one or more measurement information corresponding to the one or more output data, and the corresponding augmented and reconstructed data is obtained respectively. In this implementation, the terminal device adjusts the third coding model based on the augmented and reconstructed data to obtain a fourth coding model. During the adjustment of the third coding model, the augmented and reconstructed data of one or more output data of the second coding model is used as the model label, and the augmented and reconstructed data of the one or more measurement information is used as the model input. In addition, the augmented reconstruction is based on a certain augmentation method and filtering ratio, and the augmentation method and filtering ratio used by the terminal device and the network device are the same. In one implementation, the augmentation method and filtering ratio can be pre-configured, or the augmentation method and / or filtering ratio can be indicated to the terminal device by the network device. In the latter implementation, the first information in step 240 above can be used to indicate the augmentation method and / or filtering ratio corresponding to SAS. Alternatively, the first information is used to indicate the third dataset, or one or more reference signals corresponding to the third dataset, and the augmentation method and / or filtering ratio corresponding to the SAS are indicated by the second information, as in optional step 260. In this way, while reducing the overhead of indicating the third dataset, augmentation and reconstruction of the third dataset can be performed based on the SAS, which can further improve the performance of terminal-side model adjustment.

[0196] 260. The terminal device receives the second information, which indicates the augmentation method and / or filtering ratio corresponding to the SAS.

[0197] Optionally, when the first processing method is a core set-based method, the first information in step 240 above can be used to indicate the weights associated with each data point in the third dataset when used for model adjustment. Alternatively, the first information can be used to indicate the third dataset, or one or more reference signals corresponding to the third dataset, and the weights associated with each data point in the third dataset when used for model adjustment can be indicated by fourth information, such as in optional step 270.

[0198] 270. The terminal device receives fourth information, which indicates the weights associated with one or more data points in the third dataset when used for model adjustment.

[0199] In the case where the first processing method is based on the core set, the network side, in addition to sending the second dataset for adjusting the encoding model on the terminal side, may also send the weights associated with each data point in the third dataset for model adjustment. The third dataset is obtained by processing the first dataset using the core set-based method; that is, in the core set-based method, the third dataset is the sample set selected by the network side. The second dataset is the output of the second encoding model when the third dataset is used as the input to the adjusted self-model. The data in the second dataset and the data in the third dataset have a corresponding relationship, for example, a one-to-one correspondence. The terminal side uses the third dataset as the input to the encoding model, the second dataset as the output, and combines the weights associated with each data point in the third dataset to adjust the third encoding model on the terminal side, resulting in the fourth encoding model. Optionally, the weights associated with each data point in the third dataset used for encoding model adjustment on the terminal side can also be aligned between the network side and the terminal side in other ways, in which case step 270 may be omitted.

[0200] In the case where the first processing method is the data distillation method, the terminal device adjusts the third encoding model based on the generated dataset of the first dataset to obtain the fourth encoding model.

[0201] In one implementation, the first processing method can be pre-configured. In this implementation, the first processing method adopted by the network device is known to the terminal device; for example, a certain first processing method is used by default, or the first processing method to be used in different scenarios is negotiated in advance between the terminal device and the network device. In another implementation, the network device processes the first dataset based on the first processing method, and the network device also notifies the terminal device of the first processing method. In the latter implementation, the first information in step 240 above can also be used to indicate the first processing method. Alternatively, in another implementation, the first information is used to indicate a third dataset, or one or more reference signals corresponding to the third dataset, while the first processing method is indicated by other information, such as by the third information, as in optional step 280.

[0202] 280. The terminal device receives third information, which is used to indicate the first processing method.

[0203] In one implementation, the third information indicates one or more of meta-learning, reinforcement learning, SAS, core set-based methods, or distillation methods. In another possible implementation, the third information indicates that the first dataset should not be processed. In this case, the aforementioned second dataset includes the output data of the second encoding model when the first dataset is used as input to the adjusted dual-end model.

[0204] In the model adjustment method shown in Figure 8, after the network device completes the self-model adjustment based on the first dataset provided by the terminal device, it processes the first dataset according to a first processing method. Based on the different processing methods, it then transmits a second dataset to the terminal device, which helps reduce the transmission overhead of the dataset. Optionally, when the network device indicates a third dataset to the terminal device, it can send first information indicating the third dataset. This allows the terminal device to determine the third dataset based on the first information and combine it with the second dataset for adjusting the third coding model. Compared to the network device directly sending the third dataset, sending the first information incurs less overhead. For example, the first information could include a bitmap or the indices of one or more reference signals corresponding to one or more data points in the third dataset. This also allows the terminal device to obtain the third dataset, further reducing transmission overhead.

[0205] Figure 9 is a schematic diagram of the adjustment and matching of the dual-end model based on network-side data processing according to Scheme 1 of this application.

[0206] 301. The pairing of the third encoder on the terminal side (such as encoder 3) with the self-model on the network side, wherein the self-model includes the first encoding model and the first decoding model (such as encoder 1 and decoder 1).

[0207] 302. Optionally, the terminal device sends a request message to the network device, the request message being used to request the collection of measurement information of the reference signal, the measurement information of the reference signal being used for model adjustment.

[0208] As an example, the reference signal includes CSI-RS.

[0209] 303. The terminal device receives the configuration information from CSI-RS.

[0210] 304. The terminal device measures CSI-RS to obtain the first dataset, which contains the measurement information of CSI-RS.

[0211] 305. The terminal device sends the first dataset.

[0212] Accordingly, the network device receives the first dataset.

[0213] 306. The network device adjusts the self-model on the network side based on the first dataset to obtain the adjusted self-model.

[0214] The adjusted self-model includes a second encoding model and a second decoding model (such as encoder 2 and decoder 2).

[0215] 307. Optionally, the network device processes the first dataset based on the first processing method to obtain the third dataset.

[0216] 308. The network device sends a second dataset to the terminal device.

[0217] 309. The network device sends first information to the terminal device, the first information indicating a third dataset or one or more reference signals corresponding to the third dataset.

[0218] As described above, when the first processing method is meta-learning, reinforcement learning, a core set-based method, or SAS, the second dataset may include one or more output data from the second encoding model. In step 309, the network device may further indicate to the terminal device one or more CSI-RS corresponding to the one or more output data, which are also one or more CSI-RS included in the third dataset. As an example, the first information includes the index of the one or more CSI-RS.

[0219] As described in the embodiments above, the second dataset is related to the first processing method. For example, when the first processing method includes meta-learning, reinforcement learning, or SAS, the second dataset includes: one or more output data of the second encoding model when the third dataset is used as input to the adjusted self-model; or, when the first processing method is a data distillation method, the second dataset includes: the generated dataset of the first dataset. Optionally, in one possible implementation, the first processing method may also include: not processing the first dataset, in which case the second dataset includes: the output data of the second encoding model when the first dataset is used as input to the adjusted dual-end model. In this implementation, method 300 may not include step 307.

[0220] The third and second datasets are used together to adjust the third coding model to obtain the fourth coding model. The third dataset serves as the input to the third coding model, and the second dataset serves as the output of the third coding model, thus adjusting the third coding model.

[0221] Optionally, method 300 may also include one or more of steps 310 to 313.

[0222] 310. The network device sends third information to the terminal device, the third information being used to indicate the first processing method.

[0223] Step 310 can refer to the relevant instructions for step 280.

[0224] 311. The network device sends a second message to the terminal device, which is used to indicate the augmentation method and / or filtering ratio corresponding to the SAS.

[0225] It should be understood that step 311 is an optional step when the first processing method is SAS. Specifically, when the first processing method is SAS, and one or all of the augmentation method and filtering ratio corresponding to SAS are not pre-configured to the terminal device, the network device indicates the augmentation method and / or filtering ratio to the terminal device so that the terminal device can augment and reconstruct one or more output data of the second coding model and one or more measurement information corresponding to the one or more output data based on the augmentation method and filtering ratio, and then use it for the adjustment of the third coding model.

[0226] 312. The network device sends a fourth message to the terminal device. The second message is used to indicate the weights associated with the data in the third dataset when adjusting the model in the method based on the core set.

[0227] It should be understood that step 312 is an optional step when the first processing method is based on the core set method, and the details can be found in the detailed explanation of step 270 above.

[0228] Optionally, one or more of the following may also be indicated in the first information: the first processing method in step 310 above, or the augmentation method and / or screening ratio corresponding to the SAS in step 311, or the weights associated with one or more measurement information corresponding to the second dataset in step 312. In other words, in addition to indicating the third dataset or one or more reference signals corresponding to the third dataset, the first information may also indicate one or more of the above.

[0229] 313. The terminal device adjusts the third encoding model based on the second and third datasets to obtain the fourth encoding model (such as encoder 4), thereby completing the pairing of the encoding model on the terminal side with the adjusted self-model on the network side.

[0230] Depending on the first processing method, the implementation in step 313 will also be different. For details, please refer to the description of step 250 above, which will not be repeated here.

[0231] In the method shown in Figure 9 above, some steps can be performed by the AI ​​entity serving the UE or base station. Please refer to the description of the corresponding steps in Figure 8. Figure 9 only shows the execution by the UE or base station as an example.

[0232] Option 2

[0233] Figure 10 is a schematic flowchart of the method 500 for adjusting a model provided in this application. Method 500 involves network devices and terminal devices, and method 200 can be implemented by the network device and terminal device respectively performing corresponding steps. Optionally, one or more steps in method 500 performed by communication devices (e.g., network devices or terminal devices) can also be replaced by a device for these communication devices (e.g., referred to as a first device). The first device can be a chip, processor, circuit, or AI entity serving the communication device, etc., applied to the communication device. The AI ​​entity can be deployed on or outside the communication device, as illustrated in Figure 8. The following embodiments use network devices and terminal devices as examples for description.

[0234] 510. The terminal device acquires the first dataset, which includes measurement information of the reference signal.

[0235] Regarding the reference signal, please refer to the explanation in step 210 above.

[0236] 520. The terminal device sends dataset #1, which is related to processing method #1 for the first dataset. Dataset #1 is used for adjusting the self-model on the network side. The self-model includes a first encoding model and a first decoding model.

[0237] The network device receives dataset #1.

[0238] Optionally, method 500 includes step 530.

[0239] 530. The terminal device processes the first dataset based on processing method #1 to obtain dataset #1.

[0240] In one implementation, the processing method #1 used on the terminal side includes Augmented Similarity Subclass Selection (SAS). In this case, dataset #1 includes: the dataset obtained after processing the first dataset based on processing method #1.

[0241] In another implementation, processing method #1 includes a data distillation method. In this case, the first dataset includes: a generated dataset of dataset #1, which is the output of the distillation method when applied to the first dataset.

[0242] It should be noted that, unlike the first processing method on the network side in Scheme 1, which includes meta-learning or reinforcement learning, the processing method #1 on the terminal side in Scheme 2 does not include meta-learning or reinforcement learning. This is because such processing methods require obtaining relevant information from both the network and terminal models, information that the network device can obtain but the terminal device cannot. Therefore, this type of processing method is not applicable on the terminal side.

[0243] Optionally, processing the first dataset based on processing method #1 to obtain dataset #1 can be performed by an AI entity serving the terminal device. When performed by the AI ​​entity, the terminal device sends the first dataset to the AI ​​entity and receives dataset #1 from the AI ​​entity. Optionally, the terminal device may also send information to the AI ​​entity indicating processing method #1, so that the AI ​​entity processes the first dataset based on processing method #1 to obtain dataset #1 and sends it to the terminal device.

[0244] Optionally, method 500 may also include one or more of steps 540 to 590.

[0245] 540. The terminal device sends third information to the network device, the third information being used to indicate processing method #1; or, processing method #1 is predefined.

[0246] Similar to Scheme 1, when processing method #1 is not pre-configured and the terminal side and network side do not reach a consensus on processing method #1, the terminal device indicates processing method #1 to the network device. Processing method #1 may include one or more of SAS or data distillation methods. In another possible implementation, third information is used to indicate: do not process the first dataset. In this case, dataset #1 is the first dataset itself.

[0247] 550. The terminal device sends a second message to the network device, the second message indicating the augmentation method and / or filtering ratio corresponding to the SAS, or the augmentation method and / or filtering ratio corresponding to the SAS is predefined.

[0248] When processing method #1 is SAS, and the augmentation method and / or filtering ratio corresponding to SAS are not pre-configured, the terminal device also indicates the augmentation ratio and filtering ratio corresponding to SAS to the network device. From the network side's perspective, based on this augmentation method and filtering ratio, the corresponding dataset can be augmented and reconstructed, and then the self-model can be adjusted based on the augmented and reconstructed dataset. Here, the "corresponding dataset" may include dataset #1 sent by the terminal device, and the output data corresponding to dataset #1 in the output data of the first encoding model.

[0249] 560. The network device adjusts the self-model on the network side based on dataset #1 to obtain the adjusted self-model. The adjusted self-model includes a second encoding model and a second decoding model.

[0250] As described above, when processing method #1 is SAS, dataset #1 includes a dataset obtained by filtering data from the first dataset using SAS. The network device adjusts its self-model based on dataset #1 to obtain the adjusted self-model. When processing method #1 is a data distillation method, dataset #1 includes a generated dataset from the first dataset. The network device adjusts its self-model based on the generated dataset to obtain the adjusted self-model. Furthermore, if processing method #1 includes not processing the first dataset, then dataset #1 is the first dataset, and the network device adjusts its self-model based on dataset #1 to obtain the adjusted self-model.

[0251] Alternatively, the adjustment of the self-model can also be performed by an AI entity serving the network device. This AI entity can be deployed on the network device or outside of the network device, without limitation. Here, we only use the example of the network device adjusting the self-model for illustration.

[0252] 570. Network devices obtain datasets #2.

[0253] As an example, the network device determines dataset #2 based on dataset #1 provided by the terminal device, or the network device obtains dataset #2 from the AI ​​entity. In the latter example, the network device and the AI ​​entity interact with the corresponding dataset, and optionally, may also interact with information related to determining dataset #2. For example, the network device sends information to the AI ​​entity indicating processing method #1, so that the AI ​​entity can process the corresponding dataset based on processing method #1 to obtain dataset #2. Then, the AI ​​entity returns dataset #2 to the network device. Here, dataset #2 is the output of the second encoding model when dataset #1 is used as input to the adjusted self-model.

[0254] The solution in step 570 is the same as the solution in step 230 of solution 1, and will not be described again here.

[0255] 580. Network device sends dataset #2.

[0256] In Scheme 2, dataset #2 is a dataset sent by the network device for adjusting the encoding model of the terminal device. Specifically, dataset #2 is used to adjust the third encoding model on the terminal side to obtain a fourth encoding model. The fourth encoding model matches the adjusted second decoding model. Dataset #2 is related to the first processing method.

[0257] The solution in step 580 is the same as the solution in step 220 of solution 1, and will not be described again here.

[0258] In one implementation, the first processing method includes a data distillation method, where dataset #1 is a generated dataset of the first dataset. In this case, dataset #2 includes the output data of the second encoding model when the generated dataset of the first dataset is used as input to the adjusted self-model. For the terminal side, after receiving dataset #2, the third encoding model is adjusted based on dataset #2 and dataset #1.

[0259] In another implementation, the first processing method includes SAS, and dataset #1 includes a dataset obtained by filtering the first dataset based on SAS. In this case, dataset #2 includes one of the following:

[0260] When dataset #1 is used as the input to the adjusted two-ended model, the output data of the second encoding model is used; or,

[0261] The measurement information obtained by augmenting and reconstructing the data in dataset #1 based on the augmentation method and screening ratio corresponding to SAS, and the measurement information obtained by augmenting and reconstructing the data, are used as the input of the adjusted self-model and the output data of the second coding model.

[0262] Taking AI-CSI as an example, in the first implementation, dataset #2 includes the output data of the second coding model before augmented reconstruction, i.e., the feedback CSI before augmented reconstruction. This feedback CSI before augmented reconstruction is the output of the second coding model when the measurement CSI contained in dataset #1 fed back by the terminal device is used as the input of the adjusted self-model. For the terminal side, one possible implementation includes: directly using dataset #2 for adjusting the third coding model, specifically, using dataset #1 and dataset #2 together for adjusting the third coding model; or, another possible implementation includes: the terminal device matches the feedback CSI before augmented reconstruction with the locally collected measurement CSI one by one, performs augmented reconstruction on the feedback CSI before augmented reconstruction and its corresponding measurement CSI, and then adjusts the third coding model based on the dataset of feedback CSI and measurement CSI obtained from the augmented reconstruction. In the latter possible implementation described above, dataset #2 includes: 1) the output data of the second coding model when dataset #1 is used as the input to the adjusted self-model before augmenting and reconstructing dataset #1; 2) the dataset obtained by augmenting and reconstructing dataset #1 (i.e., the measurement information of CSI-RS after augmentation and reconstruction), and the output data of the second coding model when the augmented and reconstructed dataset of dataset #1 is used as the input to the adjusted self-model after augmenting and reconstructing dataset #1. The terminal device can directly use the received dataset #2 for adjusting the third coding model.

[0263] 590. The terminal device adjusts the third encoding model based on dataset #1 and dataset #2 to obtain the fourth encoding model.

[0264] Specifically, the terminal device adjusts the third encoding model based on dataset #1 and dataset #2 to obtain a fourth encoding model. Dataset #1 is the model's input, and dataset #2 is the model's output. The fourth encoding model obtained after model adjustment matches the second decoding model in the adjusted self-model, thus completing the pairing of the terminal-side and network-side models. Optionally, step 590 can also be performed by an AI entity serving the terminal device. In this implementation, the terminal device provides dataset #1 and dataset #2 to the AI ​​entity.

[0265] Correspondingly, one or more of steps 240-280 in Scheme 1 can also be applied to this Scheme 2 and are related to dataset #2 in steps 570, 580, or 590, which will not be elaborated here.

[0266] As can be seen in the model adjustment process shown in Figure 10, after collecting the first dataset, the terminal device processes the first dataset according to processing method #1 to obtain dataset #1. The terminal device sends dataset #1 to the network device for adjusting the self-model on the network side. Then, the network side sends dataset #2 to the terminal device for adjusting the encoding model on the terminal side, to match the adjusted self-model on the network side. Specifically, the first dataset sent by the terminal device varies depending on the processing method #1; however, in some processing methods, such as data distillation, the data transmission overhead can be reduced compared to transmitting the entire dataset #1.

[0267] Figure 11 is a schematic diagram of the adjustment and matching of the dual-end model based on terminal-side data processing according to Scheme 2 of this application.

[0268] 601. The pairing of the third encoder on the terminal side (such as encoder 3) with the self-model on the network side, wherein the self-model includes the first encoding model and the first decoding model (such as encoder 1 and decoder 1).

[0269] 602. Optionally, the terminal device sends a request message to the network device, the request message being used to request the collection of measurement information of the reference signal, the measurement information of the reference signal being used for model adjustment.

[0270] As an example, the reference signal includes CSI-RS.

[0271] 603. The terminal device receives the configuration information from CSI-RS.

[0272] 604. The terminal device measures CSI-RS to obtain the first dataset, which includes the measurement information of CSI-RS.

[0273] 605. Optionally, the terminal device processes the first dataset based on processing method #1 to obtain dataset #1.

[0274] 606. Terminal device sends dataset #1.

[0275] In this context, dataset #1 is related to processing method #1. As shown in the embodiments above, when processing method #1 includes SAS, dataset #1 includes: a dataset obtained by processing the first dataset based on SAS; or, when processing method #1 is a data distillation method, dataset #1 includes: a generated dataset of the first dataset. Optionally, in a possible implementation, processing method #1 may also include: not processing the first dataset, in which case dataset #1 is the first dataset itself. In this implementation, method 600 may not include step 605.

[0276] Accordingly, the network device receives dataset #1.

[0277] 607. The network device adjusts its self-model based on dataset #1 to obtain the adjusted self-model.

[0278] The adjusted self-model includes a second encoding model and a second decoding model (such as encoder 2 and decoder 2).

[0279] 608. Optionally, the network device obtains dataset #2.

[0280] 609. The network device sends dataset #2 to the terminal device.

[0281] Steps 608 to 609 can refer to steps 570 to 580 above.

[0282] Optionally, method 600 may also include one or more of steps 610 to 612.

[0283] 610. The terminal device sends third information to the network device, which is used to indicate processing method #1.

[0284] 611. The terminal device sends a second message to the network device, the second message being used to indicate the augmentation method and / or filtering ratio corresponding to the SAS.

[0285] It should be understood that step 611 is an optional step when processing method #1 is SAS. Specifically, when processing method #1 is SAS, and one or both of the augmentation method and filtering ratio corresponding to SAS are unknown to the network device, the terminal device indicates the augmentation method and / or filtering ratio to the network device through the second information, so that the network device can perform augmentation reconstruction on the received dataset #1 based on the augmentation method and filtering ratio, or perform augmentation reconstruction on dataset #1 and the output data of the first encoder corresponding to dataset #1, and then adjust the self-model based on the dataset obtained from the augmentation reconstruction.

[0286] 612. The terminal device adjusts the third coding model based on dataset #2 to obtain the fourth coding model (such as encoder 4), thereby completing the pairing of the coding model on the terminal side and the adjusted self-model on the network side.

[0287] For details on step 612, please refer to the relevant explanation in step 590 above, which will not be repeated here.

[0288] Option 3

[0289] Overview: The terminal processes the collected first dataset (e.g., filtering or distilling) to obtain dataset #1, and provides dataset #1 to the network side. The network side adjusts the dual-end model based on dataset #1 provided by the terminal side. Then, the network device can process dataset #1 and determine dataset #3 based on the dataset obtained from processing dataset #1. Dataset #3 includes: the output data of the second encoding model when the dataset obtained from processing dataset #1 is used as input to the adjusted self-model. Finally, the network side sends dataset #3 to the terminal device for adjusting the encoding model on the terminal side. Optionally, the network side also sends information to the terminal side indicating reference signals corresponding to the data in dataset #3.

[0290] Given the detailed explanations of Scheme 1 and Scheme 2 above, those skilled in the art can understand how Scheme 3 is implemented based on the introductions of Scheme 1 and Scheme 2, as well as the general outline of Scheme 3. A brief introduction is given below.

[0291] Figure 12 is a schematic flowchart of the method 800 for adjusting the model provided in this application. The premise of this method is that the encoding model (i.e., encoder 3) on the terminal side matches the basic self-model on the network side, which includes encoding model 1 (i.e., encoder 1) and decoding model 1 (i.e. decoder 1).

[0292] 810. The terminal device acquires the first dataset, which includes measurement information of the reference signal.

[0293] In step 801, the terminal device can obtain a first dataset by measuring the reference signal on the network side. The first dataset contains the measurement information of the reference signal.

[0294] 820. Optionally, the terminal device processes the first dataset based on processing method #1 to obtain dataset #1.

[0295] Step 820 can be referred to the description of step 530 above. Here, dataset #1 is related to processing method #1. Processing method #1 includes one or more of SAS and data distillation methods.

[0296] 830. The terminal device sends dataset #1, and the network device receives dataset #1.

[0297] Optionally, method 800 includes one or more of steps 840-850. For example, one or more of steps 840-850 are performed by an AI entity of the serving network device.

[0298] 840. The network device adjusts the dual-end model based on dataset #1 to obtain the adjusted dual-end model. The adjusted self-model includes encoder 2 and decoder 2.

[0299] 850. The network device determines dataset #3 based on the first processing method and dataset #1.

[0300] In step 850, the network device processes dataset #1 using a first processing method to obtain dataset #4. This process can be referenced from the explanation in step 230 regarding "the network device processes the first dataset using a first processing method to obtain the second dataset." Dataset #1 is interpreted as the first dataset, and dataset #4 is interpreted as the second dataset. As described above, when the network side processes dataset #1, the first processing method can include one or more of the following: SAS, data distillation, meta-learning, reinforcement learning, or core set-based methods. Compared to the terminal-side processing method #1, the network side has more options for the first processing method.

[0301] As can be seen, in Scheme 3, the terminal side processes the first dataset obtained by the reference signal measurement based on processing method #1 to obtain dataset #1. The network side processes dataset #1 a second time based on the first processing method to obtain dataset #4.

[0302] Furthermore, the network device can determine dataset #3 based on dataset #4. Dataset #3 is related to the first processing method; specifically, dataset #3 includes the output data of the second encoding model when dataset #4 is used as the input to the adjusted self-model.

[0303] 860. The network device sends dataset #3. Furthermore, the network device can further instruct dataset #4.

[0304] Among them, datasets #3 and #4 are used to adjust encoder 3 on the terminal side to obtain encoder 4, which matches decoder 2 in the adjusted self-model.

[0305] Optionally, the network device indicating dataset #4 may include: directly sending dataset #4; or sending information #1, which indicates one or more reference signals corresponding to dataset #4. For example, information #1 may indicate the index of one or more reference signals corresponding to dataset #4, or information #1 may be a bitmap to indicate one or more reference signals corresponding to dataset #4. Based on information #1, the terminal side determines the data that corresponds one-to-one with the one or more reference signals from its local first dataset, which is dataset #4.

[0306] Optionally, step 870 may also be included.

[0307] 870. The terminal device adjusts the encoding model 3 based on datasets #3 and #4 to obtain the adjusted encoder 4. Encoder 4 is matched with decoder 2 in the adjusted self-model.

[0308] In this model, dataset #4 is used as the input to encoding model 3, and dataset #3 is used as the output to adjust encoding model 3.

[0309] It should be understood that some optional steps in Scheme 1 and / or Scheme 2 can also be applied to Scheme 3. For example, when the first processing method is SAS, the network side may send information #2 to the terminal side. Information #2 is used to indicate the augmentation method and / or filtering ratio corresponding to SAS, for the terminal side to augment and reconstruct datasets #3 and #4, and then for model adjustment. Another example is that the network side sends information #3 to the terminal side, where information #3 indicates the first processing method, etc. Since Scheme 2 already includes descriptions of these implementations, for the sake of brevity, they will not be repeated. Those skilled in the art will understand that some steps or implementation methods in Scheme 1 and Scheme 2 can be applied to Scheme 3, even if these steps or implementation methods are not shown one by one in the accompanying drawings corresponding to Scheme 3.

[0310] After the terminal side completes the adjustment of the coding model, the adjusted coding model on the terminal side is paired with the adjusted self-model on the network side.

[0311] It should be understood that Scheme 3 mainly describes the process related to model adjustment. Before model adjustment, the pairing process between the encoding model on the terminal side and the self-model on the network side can refer to the description in the foregoing embodiments. In addition, the alignment of the data processing methods (e.g., the first processing method or processing method #1) between the terminal side and the network side, and the interaction of some optional information when the first processing method or processing method #1 is certain processing methods (e.g., SAS), such as the interaction of augmentation methods and / or filtering ratios corresponding to SAS, can also refer to the description in the foregoing embodiments, and will not be repeated here.

[0312] In Scheme 3, the terminal sends dataset #1, obtained by processing the first dataset obtained from the measurement of the reference signal based on processing method #1, to the network side, which saves transmission overhead. Furthermore, when the network side sends the dataset for the terminal's encoding model after completing the self-model adjustment, it also sends dataset #4, obtained by further processing dataset #1 based on the first processing method, and the output of the second encoding model, i.e., dataset #3, indicating that dataset #4 is used as the input to the adjusted self-model. This also helps reduce the transmission overhead of the datasets. Therefore, Scheme 3 benefits from reducing the transmission overhead of the datasets from both the terminal-side and network-side data transmission.

[0313] The method for adjusting the model provided in this application has been described in detail above. The corresponding communication device is described below.

[0314] Figure 13 is a schematic block diagram of the communication device 1000 provided in this application. As shown in Figure 13, the communication device 1000 may include a processing module 1001 and a communication module 1002. The communication device 1000 may be a terminal device, or a communication device applied to or used in conjunction with a terminal device to achieve the corresponding functions of the terminal device, such as a processor, chip, circuit, or AI entity. Alternatively, the communication device 1000 may be a network device, or a communication device applied to or used in conjunction with a network device to achieve the corresponding functions of the network device, such as a processor, chip, circuit, or AI entity.

[0315] The communication module can also be called a transceiver module, transceiver, transceiver machine, or transceiver device, etc. The processing module can also be called a processor, processing board, processing unit, or processing device, etc. Optionally, the communication module is used to perform the sending and receiving operations on the terminal side or network device side in the above method. The device in the communication module that implements the receiving function can be regarded as a receiving unit, and the device in the communication module that implements the sending function can be regarded as a sending unit. That is, the communication module includes a receiving unit and a sending unit. When the communication device 1000 is applied to a network device or terminal device, the processing module 1001 can be used to implement the processing functions of the network device or terminal device in the embodiments of Figures 7 to 12, and the communication module 1002 can be used to implement the sending and receiving functions of the network device or terminal device. For example, when applied to the network side, the communication module 1002 can be used to receive a first dataset, send a second dataset, send first information, send second information, or send third information, etc.; the processing module 1001 can be used to: adjust the self-model based on the received first dataset, or determine the output data of the second encoding model in the adjusted self-model, etc. When the communication device 1000 is applied to a terminal device, the communication module 1002 can be used to: send a first dataset, receive a second dataset, receive first information, receive second information, or receive third information, etc.; the processing module 1001 can be used to: adjust the third encoding model, augment and reconstruct the received second dataset, etc.

[0316] Furthermore, it should be noted that the aforementioned communication module and / or processing module can be implemented through virtual modules. For example, the processing module can be implemented through software functional units or virtual devices, and the communication module can be implemented through software functions or virtual devices. Alternatively, the processing module or communication module can also be implemented through physical devices. For example, if the device is implemented using a chip / chip circuit, the communication module can be an input / output circuit and / or a communication interface, performing input operations (corresponding to the aforementioned receiving operation) and output operations (corresponding to the aforementioned sending operation); the processing module is an integrated processor, microprocessor, or integrated circuit.

[0317] The module division in this application is illustrative and represents only one logical functional division. In actual implementation, other division methods are possible. Furthermore, the functional modules in the various examples of this application can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware, as software functional modules, or a combination of hardware and software.

[0318] Figure 14 is a schematic block diagram of another communication device 1100 provided in this application. Optionally, the communication device 1100 may be a chip or a chip system. Optionally, in this application, the chip system may be composed of chips or may include chips and other discrete devices.

[0319] The communication device 1100 can be used to implement the functions of any of the network elements (e.g., network devices or terminal devices) described in the foregoing embodiments. The communication device 1100 may include at least one processor 1110. Optionally, the processor 1110 is coupled to a memory, which may be located within the communication device 1100, integrated with the processor, or located outside the communication device 1100. As an example, the communication device 1100 may also include at least one memory 1120. The memory 1120 stores the necessary computer programs (or computer instructions) and / or data for implementing the corresponding functions of any of the network elements in any of the above method embodiments; the processor 1110 may execute the computer programs stored in the memory 1120 to complete the methods implemented by any of the network elements in any of the above method embodiments.

[0320] The communication device 1100 may also include a communication interface 1130, through which the communication device 1100 can interact with other devices. For example, the communication interface 1130 may be a transceiver, circuit, bus, module, pin, or other type of communication interface. When the communication device 1100 is a chip-based device or circuit, the communication interface 1130 in the device 1100 may also be an input / output circuit, capable of inputting information (or receiving information) and outputting information (or sending information). The processor may be an integrated processor, a microprocessor, an integrated circuit, or a logic circuit, and the processor can determine the output information based on the input information.

[0321] The coupling in this application refers to indirect coupling or communication connection between devices, units, or modules, which can be electrical, mechanical, or other forms, used for information exchange between devices, units, or modules. The processor 1110 may operate in conjunction with the memory 1120 and the communication interface 1130. This application does not limit the specific connection medium between the processor 1110, the memory 1120, and the communication interface 1130.

[0322] Optionally, as shown in FIG14, the processor 1110, the memory 1120, and the communication interface 1130 are interconnected via a bus 1140. The bus 1140 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one line is used to represent the bus 1140 in FIG14, but this does not indicate that there is only one bus or one type of bus.

[0323] In one implementation, the communication device 1100 can be applied to the network side, such as the network device in the embodiments of this application, or a smart network element on the network side. Specifically, the communication device 1100 can be a network device, or it can be a device capable of supporting the network device to implement the corresponding functions of the network device in any of the above method embodiments. The memory 1120 stores computer programs (or computer instructions) and / or data that implement the corresponding functions of the network device. The processor 1110 can execute the computer programs or instructions stored in the memory 1120 to complete the methods executed by the network device in any of the above method embodiments. The communication interface in the communication device 1100 can be used to interact with terminal devices.

[0324] In another implementation, the communication device 1100 can be applied to the terminal side. For example, the communication device 1100 can be a terminal device, or a device capable of supporting the terminal device and implementing the corresponding functions of the terminal device in any of the above method embodiments, such as the host of an OTT system or a cloud server. The memory 1120 stores computer programs (or computer instructions) and / or data that implement the corresponding functions of the terminal device in any of the above method embodiments. The processor 1110 can execute the computer program stored in the memory 1120 to complete the method executed by the terminal device in any of the above method embodiments. The communication interface in the communication device 1100 can be used to interact with network devices (e.g., base stations), sending information to or receiving information from network devices.

[0325] Figure 15 is a schematic structural diagram of the chip provided in this application. Chip 30 includes a processing circuit 31 and a communication circuit 32. The processing circuit 31 can be a logic circuit, integrated circuit, etc., and the communication circuit 32 can be an input / output circuit, input / output interface, interface circuit, etc., which can input information (or receive information) or output information (or send information). Chip 30 can execute the methods executed by network devices or terminal devices in the various embodiments of this application. The processing circuit 31 can be one or more processors, or all or part of the circuitry in one or more processors used for control or processing.

[0326] Figure 16 is a schematic diagram of the system architecture of the communication device provided in this application. The input / output control module manages the input and output signals of the communication device (e.g., network device or terminal device). For example, the input / output control can be one or more forms such as a modem, keyboard, mouse, or touchscreen. The input / output control may also be part of the processor. The communication device establishes communication connections with other devices through the communication control module. The receiver / transmitter is used to communicate with other devices. The receiver / transmitter may include a modem for modulating information (transmitting device) or demodulating modulated information (receiving device). The antenna is used to transmit or receive signals. Storage can be used to store computer code, which can be executed by the processor to implement the corresponding functions of the communication device. The processor may include intelligent hardware devices such as a general-purpose processor, digital signal processor (DSP), central processing unit (CPU), field-programmable gate array (FPGA), graphics processing unit (GPU), neural processing unit (NPU), etc. The communication device shown in Figure 16 can be a network device or a terminal device in the embodiments of this application.

[0327] In addition, this application also provides a computer-readable storage medium storing computer instructions, which, when executed on a computer, cause operations and / or processes performed by a terminal device or network device in the various method embodiments of this application to be executed.

[0328] This application also provides a computer program product, which includes computer program code or instructions. When the computer program code or instructions are run on a computer, the operations and / or processes performed by a terminal device or network device in the various method embodiments of this application are executed.

[0329] This application also provides a chip including a processor, and a memory for storing a computer program, disposed independently of the chip. The processor executes the computer program stored in the memory, such that operations and / or processes performed by a terminal device or network device in any method embodiment are executed. Further, the chip may also include a communication interface. The communication interface may be an input / output interface or an interface circuit, etc. Further, the chip may also include a memory.

[0330] This application also provides a chip, which may include circuitry and an input / output interface. The circuitry may be logic circuitry, integrated circuits, etc., and exemplaryly, the circuitry may be one or more processors, or all or part of the circuitry in one or more processors used to implement one or more processing, control, or computing functions. The input / output interface may also be an input / output circuit, or an interface circuit, capable of inputting information (or receiving information) and / or outputting information (or sending information). The chip may include a chip system. Optionally, the chip system may be composed of chips or may include chips and other discrete devices. The chip can be used to execute the methods implemented by terminal devices or network devices in the various embodiments of this application. Optionally, the chip may be a baseband chip, also known as a modem.

[0331] Furthermore, this application provides a communication system, including the terminal device and network device in any embodiment of this application. This communication system can implement the adjustment model method provided in any embodiment of Figures 7 to 12.

[0332] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0333] The processor in this application embodiment has signal processing capabilities and can be a central processing unit (CPU), or a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component. It can implement or execute the methods, steps, and logic block diagrams disclosed in this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in this application can be directly embodied in the execution of the hardware processor, or executed by a combination of hardware and software modules within the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0334] In the embodiments of this application, memory is any other medium capable of carrying or storing desired program code in the form of instructions or data structures, and accessible by a computer, but is not limited thereto. The memory in this application can also be a circuit or any other means capable of implementing a storage function for storing computer programs and / or data; or, it can also be a circuit or any other means capable of implementing a storage function for storing computer programs and / or data. As an example, memory can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM). It should be noted that the memory used in the systems and methods described herein is intended to include, but is not limited to, the types described above or any other suitable types of memory.

[0335] The technical solutions provided in this application can be implemented in whole or in part through software, hardware, firmware, or any combination thereof. When implemented using software, they can be implemented in whole or in part as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a terminal device, an access network device, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., digital video discs (DVDs)), or semiconductor media, etc.

[0336] In the embodiments of this application, "at least one" refers to one or more items. "More than one" means two or more items. "And / or" is used to describe the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0337] The term "comprising" and any variations thereof used in the embodiments of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0338] In this application, examples may reference each other without logical contradiction. For example, methods and / or terms between method embodiments may reference each other, functions and / or terms between device embodiments may reference each other, and functions and / or terms between device examples and method examples may reference each other.

[0339] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0340] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0341] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0342] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0343] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0344] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for adjusting a model, characterized in that, include: Send a first dataset, which includes measurement information of a reference signal. The first dataset is used for adjusting a self-model, which includes a first encoding model and a first decoding model. A second dataset is received, which is related to a first processing method. The first processing method includes a processing method for the first dataset. The second dataset is used to adjust a third encoding model to obtain a fourth encoding model, wherein the fourth encoding model matches the second decoding model in the adjusted self-model.

2. The method according to claim 1, characterized in that, The second dataset includes at least one or more output data of the second encoding model, wherein the second encoding model is an encoding model obtained by adjusting the first encoding model.

3. The method according to claim 1 or 2, characterized in that, The method further includes: The third encoding model is adjusted based on the second dataset.

4. The method according to claim 2 or 3, characterized in that, The first processing method includes one or more of the following methods: Meta-learning, reinforcement learning, core set-based methods, data distillation methods, or augmented similarity subclassing are all options for SAS. One or more outputs of the second encoding model include: output data of the second encoding model when a third dataset is used as input to the adjusted self-model, wherein the third dataset is obtained by processing the data in the first dataset based on the first processing method.

5. The method according to claim 4, characterized in that, The method further includes: The third dataset is obtained by receiving first information, wherein the first information indicates one or more of the following: The third dataset, or one or more reference signals corresponding to the third dataset, or the first processing method, or parameters related to the first processing method; or, The first processing method is predefined, and the third dataset is obtained based on the first processing method and the first dataset; The second dataset and the third dataset are used together to adjust the third encoding model.

6. The method according to claim 5, characterized in that, The first processing method includes a data distillation method; The third dataset includes the generated dataset of the first dataset, and one or more output data of the second encoding model include the output of the second encoding model when the generated dataset is used as the input of the adjusted self-model. The second encoding model is the encoding model obtained by adjusting the first encoding model.

7. The method according to any one of claims 3-6, characterized in that, The adjustment of the third encoding model based on the second dataset includes: The third coding model is adjusted based on one or more output data of the second coding model contained in the second dataset, and one or more measurement information corresponding to the one or more output data, wherein the second coding model is the coding model obtained by adjusting the first coding model.

8. The method according to claim 5, characterized in that, The first processing method includes the SAS, and the parameters related to the first processing method include the augmentation method and / or screening ratio corresponding to the SAS.

9. The method according to claim 8, characterized in that, The method further includes: Based on the augmentation method and / or the filtering ratio, the second dataset and one or more measurement information corresponding to the second dataset are augmented and reconstructed to obtain augmented and reconstructed data. The adjustment of the third encoding model based on the second dataset includes: The third coding model is adjusted based on the augmented and reconstructed data.

10. The method according to claim 5, characterized in that, The first processing method includes a core set-based approach, and the parameters related to the first processing method include the weights associated with the data in the third dataset when adjusting the model.

11. The method according to claim 10, characterized in that, The adjustment of the third encoding model based on the second dataset includes: The third encoding model is adjusted based on the weights associated with the data in the second and third datasets.

12. The method according to any one of claims 5-11, characterized in that, The first information indicating the first processing method includes: the first information indicating that the first dataset should not be processed; The second dataset includes: the output data of the second encoding model when the first dataset is used as the input of the adjusted self-model, wherein the second encoding model is the encoding model obtained by adjusting the first encoding model.

13. A method for adjusting a model, characterized in that, include: Receive a first dataset, the first dataset including measurement information of a reference signal, the first dataset being used for adjusting a self-model, the self-model including a first encoding model and a first decoding model; as well as, A second dataset is sent, which relates to a first processing method for the first dataset. The second dataset is used to adjust a third encoding model to obtain a fourth encoding model, wherein the fourth encoding model matches a second decoding model in the adjusted self-model.

14. The method according to claim 13, characterized in that, The second dataset includes at least one or more output data of the second encoding model, wherein the second encoding model is an encoding model obtained by adjusting the first encoding model.

15. The method according to claim 13 or 14, characterized in that, The first processing method includes one or more of the following: meta-learning, reinforcement learning, core set-based methods, data distillation methods, or augmented similarity subclass selection (SAS); One or more output data of the second encoding model include: output data of the second encoding model when the third dataset is used as input to the adjusted self-model, wherein the third dataset is obtained by processing the data in the first dataset based on the first processing method.

16. The method according to claim 15, characterized in that, The method further includes: Send a first message, which indicates one or more of the following: The third dataset, or one or more reference signals corresponding to the third dataset, or the first processing method, or parameters related to the first processing method.

17. The method according to claim 16, characterized in that, The first processing method includes a data distillation method; The third dataset includes the generated dataset of the first dataset, and one or more output data of the second encoding model include the output of the second encoding model when the generated dataset is used as the input of the adjusted self-model. The second encoding model is the encoding model obtained by adjusting the first encoding model.

18. The method according to claim 16, characterized in that, The first processing method includes the SAS, and the parameters related to the first processing method include the augmentation method and / or screening ratio corresponding to the SAS.

19. The method according to claim 16, characterized in that, The first processing method includes the core set-based method, and the parameters related to the first processing method include: the weights associated with the data in the third dataset when adjusting the model.

20. A method for adjusting a model, characterized in that, include: Obtain a first dataset, which includes measurement information of a reference signal; Send dataset #1, which is related to processing method #1 for the first dataset. The dataset #1 is used for adjusting the self-model, which includes a first encoding model and a first decoding model.

21. The method according to claim 20, characterized in that, The method further includes: Based on processing method #1, the first dataset is processed to obtain dataset #1.

22. The method according to claim 20 or 21, characterized in that, The method further includes: Receive dataset #2, which is used to adjust the third encoding model to obtain a fourth encoding model. The fourth encoding model is matched with the second decoding model in the adjusted self-model. The dataset #2 is the output data of the second encoding model when the dataset #1 is used as the input of the adjusted self-model. The second encoding model is the encoding model obtained by adjusting the first encoding model.

23. The method according to claim 21 or 22, characterized in that, The processing method #1 is augmented similarity subclass selection (SAS), and the method further includes: Send a second message, which indicates the augmentation method and / or screening ratio corresponding to the SAS.

24. The method according to any one of claims 21 to 23, characterized in that, The method further includes: Send a third message, which indicates the processing method #1.

25. The method according to claim 20 or 21, characterized in that, The method further includes: Receive dataset #3, which is used to adjust the third encoding model to obtain a fourth encoding model. The fourth encoding model is matched with the second decoding model in the adjusted self-model. The dataset #3 is the output of the second encoding model when the dataset #4 is used as the input of the adjusted self-model. The dataset #4 is determined based on the dataset #1 and the first processing method.

26. The method according to claim 25, characterized in that, The first processing method includes one or more of the following: meta-learning, reinforcement learning, core set-based methods, data distillation methods, or augmented similarity subclass selection SAS.

27. The method according to claim 25 or 26, characterized in that, The method further includes: Receive information #1, which indicates one or more of the following: the dataset #4, or one or more reference signals corresponding to the dataset #4, or the first processing method, or parameters related to the first processing method.

28. The method according to any one of claims 25 to 27, characterized in that, The method further includes: The data in dataset #4 is received for the weights associated with each data point during model adjustment.

29. The method according to any one of claims 20 to 28, characterized in that, The processing method #1 includes a data distillation method; the dataset #1 includes: a generated dataset of the first dataset, wherein the generated dataset is the output of the distillation method when the distillation method is applied to the first dataset.

30. The method according to claim 22 or 23, characterized in that, The processing method #1 is a data distillation method, and the dataset #2 includes: the output data of the second encoding model when the generated dataset of the first dataset is used as the input of the adjusted self-model.

31. The method according to any one of claims 24 to 28, characterized in that, The third information indicates that the processing method #1 includes: The third information indicates that the first dataset is not processed; the dataset #1 includes the first dataset; and the dataset #2 includes the output data of the second encoding model when the dataset #1 is used as the input of the adjusted self-model, the second encoding model being an encoding model obtained by adjusting the first encoding model.

32. The method according to any one of claims 25 to 28, characterized in that, The first processing method includes the SAS, and the parameters related to the first processing method include the augmentation method and / or screening ratio corresponding to the SAS. Alternatively, the first processing method includes the core set-based method, and the parameters related to the first processing method include the weights associated with the data in dataset #4 when they are used for model adjustment.

33. A method for adjusting a model, characterized in that, include: Receive dataset #1, which is used for adjusting the self-model, the self-model including a first encoding model and a first decoding model, the dataset #1 being related to the processing method #1 for the first dataset, the first dataset including measurement information of the reference signal; as well as, Acquire dataset #2 or dataset #3, which is used to adjust the third encoding model to obtain a fourth encoding model. The fourth encoding model is matched with the second decoding model in the adjusted self-model. The dataset #2 is determined based on the dataset #1 and the adjusted self-model. The dataset #3 is related to the first processing method for the dataset #1.

34. The method according to claim 33, characterized in that, When the dataset #2 is obtained, the method further includes: Send the dataset #2.

35. The method according to claim 33 or 34, characterized in that, The processing method #1 includes augmented similarity subclass selection (SAS), and the method further includes: Receive second information, which is used to indicate the augmentation method and / or screening ratio corresponding to the SAS.

36. The method according to any one of claims 33 to 35, characterized in that, The method further includes: Receive third information, which indicates the processing method #1.

37. The method according to claim 33, characterized in that, When the dataset #3 is obtained, the method further includes: Send the dataset #3, which is the output of the second encoding model when the dataset #4 is used as the input of the adjusted self-model. The dataset #4 is determined based on the dataset #1 and the first processing method.

38. The method according to claim 37, characterized in that, The method further includes: Dataset #1 is processed based on the first processing method to obtain dataset #4. The first processing method includes one or more of the following: meta-learning, reinforcement learning, core set-based methods, data distillation methods, or augmented similarity subclass selection (SAS).

39. The method according to claim 37 or 38, characterized in that, The method further includes: Send information #1, which indicates one or more of the following: the dataset #4, or one or more reference signals corresponding to the dataset #4, or the first processing method, or parameters related to the first processing method.

40. The method according to claim 38 or 39, characterized in that, The first processing method includes the SAS, and the method further includes: Send message #2, which indicates the augmentation method and / or filtering ratio corresponding to the SAS.

41. The method according to any one of claims 38 to 40, characterized in that, The method further includes: The data in dataset #4 is sent for the respective associated weights during model adjustment.

42. The method according to any one of claims 38 to 40, characterized in that, The method further includes: Send message #3, which is used to indicate the first processing method.

43. The method according to any one of claims 33 to 42, characterized in that, The processing method #1 includes augmented similarity subclass selection (SAS); the dataset #1 includes: the dataset obtained after processing the first dataset based on the processing method #1.

44. The method according to any one of claims 33 to 43, characterized in that, The processing method #1 is to augment similarity subclass selection SAS, and the dataset #2 includes one of the following: When the dataset #1 is used as the input to the adjusted self-model, the output data of the second encoding model; or, The data obtained by augmenting and reconstructing the data in dataset #1 based on the augmentation method and filtering ratio corresponding to SAS, and the data obtained by augmenting and reconstructing the data as input to the adjusted self-model, are the output data of the second encoder.

45. A communication device, characterized in that, It includes modules or units for performing the methods as described in any one of claims 1 to 12, 20 to 32, or modules or units for performing the methods as described in any one of claims 13 to 19, 33 to 44.

46. ​​A communication device, characterized in that, The device includes a processor coupled to a memory, the processor being configured to execute a computer program or instructions stored in the memory to cause the communication device to perform the method as described in any one of claims 1 to 12, 20 to 32, or to perform the method as described in any one of claims 13 to 19, 33 to 44.

47. The communication device according to claim 46, characterized in that, It also includes a memory for storing the computer program or instructions.

48. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed on a computer, implement the method as described in any one of claims 1 to 12, 20 to 32, or the method as described in any one of claims 13 to 19, 33 to 44.

49. A computer program product, characterized in that, Includes a computer program or instructions for performing the method as described in any one of claims 1 to 12, 20 to 32, or includes instructions for performing the method as described in any one of claims 13 to 19, 33 to 44.

50. A communication system, characterized in that, It includes one or more of the following means: means for implementing the method as described in any one of claims 1 to 12, or 20 to 32, or means for implementing the method as described in any one of claims 13 to 19, 33 to 44.

Images