Communication method and communication device

Abstract

Provided are a communication method and a communication device. The method comprises: when a first condition is satisfied, a terminal device determines to use a first model for inference in a first cell, the first condition being associated with one or more of the following: training data of the first model; first configuration information, used for indicating a network-device-side condition for an inference phase of a model and / or a first association identifier, the first association identifier being used for indicating an additional network-device-side condition for the inference phase; and second configuration information, used for indicating a network-device-side condition for a training phase of a model and / or a second association identifier, the second association identifier being used for indicating an additional network-device-side condition for the training phase.

Description

Communication methods and communication equipment Technical Field

[0001] This application relates to the field of communication technology, and more specifically, to a communication method and a communication device. Background Technology

[0002] In some communication systems, terminal devices can perform inference using models (such as artificial intelligence (AI) models or machine learning (ML) models) within a cell. Ensuring the performance of terminal devices performing inference using these models within a cell is a technical problem that needs to be solved. Summary of the Invention

[0003] This application provides a communication method and a communication device. The various aspects covered by this application are described below.

[0004] In a first aspect, a communication method is provided, the method comprising: upon the satisfaction of the first condition, a terminal device determining to perform inference using a first model in a first cell; the first condition being associated with one or more of the following: training data of the first model; first configuration information for indicating network device-side conditions and / or a first association identifier for the inference phase of the model, the first association identifier indicating additional network device-side conditions for the inference phase; and second configuration information for indicating network device-side conditions and / or a second association identifier for the training phase of the model, the second association identifier indicating additional network device-side conditions for the training phase.

[0005] In a second aspect, a communication method is provided, the method comprising: a network device sending configuration information of a first model association to a terminal device, the configuration information of the first model association including one or more of the following: first configuration information for indicating network device-side conditions and / or a first association identifier during the inference phase of the model, the first association identifier being used to indicate additional network device-side conditions during the inference phase; and second configuration information for indicating network device-side conditions and / or a second association identifier during the training phase of the model, the second association identifier being used to indicate additional network device-side conditions during the training phase.

[0006] Thirdly, a communication device is provided, the device being a terminal device. The device includes: a determining unit, configured to determine, when a first condition is met, to use a first model for inference in a first cell; the first condition being associated with one or more of the following: training data of the first model; first configuration information for indicating network device-side conditions and / or a first association identifier for the inference phase of the model, the first association identifier indicating additional network device-side conditions for the inference phase; and second configuration information for indicating network device-side conditions and / or a second association identifier for the training phase of the model, the second association identifier indicating additional network device-side conditions for the training phase.

[0007] Fourthly, a communication device is provided, the device being a network device. The device includes: a first transmitting unit, configured to transmit configuration information for a first model association to a terminal device, the configuration information for the first model association including one or more of the following: first configuration information, used to indicate network device-side conditions and / or a first association identifier during the inference phase of the model, the first association identifier indicating additional network device-side conditions during the inference phase; and second configuration information, used to indicate network device-side conditions and / or a second association identifier during the training phase of the model, the second association identifier indicating additional network device-side conditions during the training phase.

[0008] Fifthly, a communication device is provided, including a memory and a processor, the memory being used to store a program, and the processor being used to invoke the program in the memory to cause the communication device to perform the method of the first aspect or the second aspect.

[0009] In a sixth aspect, an apparatus is provided, including a processor configured to invoke a program from a memory to cause the apparatus to perform the methods of the first or second aspect.

[0010] In a seventh aspect, a chip is provided, including a processor for calling a program from memory to cause a device having the chip mounted to perform the methods of the first or second aspect.

[0011] Eighthly, a computer-readable storage medium is provided, the computer-readable storage medium storing a program that causes a computer to perform the method of the first or second aspect.

[0012] Ninth aspect, a computer program product is provided, the computer program product comprising a program that causes a computer to perform the method of the first or second aspect.

[0013] In a tenth aspect, a computer program is provided that causes a computer to perform the methods of the first or second aspect.

[0014] This application introduces a first condition, which is associated with one or more of the following: training data of the first model, first configuration information, and second configuration information. The first condition can assist the terminal device in determining whether it can use the first model for inference within the first cell, thereby ensuring the performance of the terminal device in using the first model for inference within the first cell. Attached Figure Description

[0015] Figure 1 is a schematic diagram of the communication system used in the embodiments of this application.

[0016] Figure 2 is a schematic diagram of the input and output relationship of the transmit beam prediction model provided in the embodiments of this application.

[0017] Figure 3 is a schematic diagram of the input and output relationship of the optimal beam quality prediction model provided in the embodiments of this application.

[0018] Figure 4 is a schematic diagram of the measurement window and prediction window in time-domain beam prediction provided in the embodiments of this application.

[0019] Figure 5 is a schematic diagram of channel state information (CSI) prediction provided in an embodiment of this application.

[0020] Figure 6 is a schematic flowchart of the communication method provided in an embodiment of this application.

[0021] Figure 7 is a schematic structural diagram of a communication device provided in one embodiment of this application.

[0022] Figure 8 is a schematic structural diagram of a communication device provided in another embodiment of this application.

[0023] Figure 9 is a schematic structural diagram of the communication device provided in an embodiment of this application. Detailed Implementation

[0024] The technical solutions in this application will now be described with reference to the accompanying drawings.

[0025] Communication system

[0026] Figure 1 is a system architecture example diagram of a wireless communication system 100 applicable to embodiments of this application. The wireless communication system 100 may include a network device 110 and a terminal device 120. The network device 110 may be a device that communicates with the terminal device 120. The network device 110 can provide network coverage for a specific geographical area and can communicate with the terminal device 120 located within that coverage area. The terminal device 120 can access a network (such as a wireless network) through the network device 110. Optionally, the wireless communication system 100 may also include other network entities such as a network controller and a mobility management entity; this embodiment of the application does not limit this.

[0027] It should be understood that the technical solutions of the embodiments of this application can be applied to various communication systems, such as 5G systems or new radio (NR), long term evolution (LTE) systems, LTE frequency division duplex (FDD) systems, LTE time division duplex (TDD) systems, etc. The technical solutions provided in this application can also be applied to future communication systems, such as sixth-generation mobile communication systems, satellite communication systems, and so on.

[0028] The terminal device in this application embodiment can also be referred to as user equipment (UE), access terminal, user unit, user station, mobile station, mobile station (MS), mobile terminal (MT), remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent, or user device. The terminal device in this application embodiment can be a device that provides voice and / or data connectivity to a user, and can be used to connect people, objects, and machines, such as a handheld device with wireless connectivity, vehicle-mounted device, etc. The terminal devices in the embodiments of this application can be mobile phones, tablets, laptops, PDAs, mobile internet devices (MIDs), wearable devices, virtual reality (VR) devices, augmented reality (AR) devices, wireless terminals in industrial control, self-driving, remote medical surgery, smart grids, transportation safety, smart cities, and smart homes, etc. Optionally, the terminal device can act as a base station. For example, the terminal device can act as a scheduling entity, providing sidelink signals between terminal devices in vehicle-to-everything (V2X) or device-to-device (D2D) systems. For instance, cellular phones and cars communicate with each other using sidelink signals. Cellular phones and smart home devices communicate without relaying communication signals through base stations.

[0029] The network device in this application embodiment can be a device for communicating with terminal devices. This network device can be, for example, an access network device or a wireless access network device. For instance, the network device can be a base station. The term "base station" can broadly encompass various names, or be replaced by, the following: NodeB, evolved NodeB (eNB), next-generation NodeB (gNB), relay station, access point, transmitting and receiving point (TRP), transmitting point (TP), 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), positioning node, etc. A base station can be a macro base station, micro base station, relay node, donor node, or the like, or a combination thereof.

[0030] This application describes AI and ML technologies in wireless communication. To facilitate understanding of the embodiments of this application, the AI ​​and ML technologies in wireless communication involved in the embodiments of this application are described below with reference to the accompanying drawings.

[0031] AI / ML technology in wireless communication

[0032] In recent years, with the development of different types of neural networks and machine learning algorithms, AI and ML technologies have been widely applied in various fields such as image, speech, and video processing. Typical neural network architectures include fully connected networks, convolutional neural networks (CNNs), recurrent neural networks (RNNs), and transformer structures with self-attention mechanisms. These neural network architectures can accomplish different tasks.

[0033] The rapid development of AI and ML technologies has led to widespread interest in the integration of AI with wireless communication technologies, attracting significant attention from both academia and industry. In 3GPP Releases 18 and 19, extensive research, evaluation, and standardization work has been conducted on AI / ML-based CSI feedback, beam management, and positioning technologies. Furthermore, the integration of AI / ML technologies into future wireless communication systems may generate even more wireless AI / ML use cases. Examples include AI / ML-based channel estimation methods, AI / ML-based modulation and demodulation techniques, and AI / ML-based integrated receiver designs. These use cases all demonstrate performance gains compared to traditional non-AI / ML algorithms. Therefore, the design of future 6th generation (6G) wireless communication systems may incorporate more AI / ML modules to enhance overall system performance.

[0034] AI / ML-based beam management

[0035] In the discussions of 3GPP Release 18 (SI) phase, AI / ML-based beam management (BM) received widespread attention as a major use case of the Release 18 AI project. Although many details regarding how to implement AI / ML-based beam management remain unresolved, 3GPP RAN1 defines two typical use cases: spatial-domain and temporal downlink (DL) beam prediction. Spatial-domain downlink beam prediction is referred to below as BM-Use Case 1. Temporal-domain downlink beam prediction is referred to below as BM-Use Case 2. See below for protocol details.

[0036] For AI / ML-based beam management, BM-use case 1 and BM-use case 2 are supported for characterization and baseline performance evaluation.

[0037] BM-Use Case 1: Perform spatial domain DL beam prediction on beams in set A based on the measurement results of beams in set B.

[0038] BM-Use Case 2: Perform time-domain DL beam prediction on beams in set A based on historical measurement results of beams in set B.

[0039] FFS: Details of BM-Use Case 1 and BM-Use Case 2

[0040] FFS: Other Sub-use Cases

[0041] Note: For BM-use case 1 and BM-use case 2, the beams in set A and set B can be in the same frequency range.

[0042] The following sections will provide a detailed description of BM-use case 1 and BM-use case 2 in conjunction with Figures 2 and 3.

[0043] Beam prediction in the spatial domain (BM - Use Case 1)

[0044] Spatial domain beam prediction can be understood as: predicting the spatial domain of the downlink transmit beams in set A by measuring the downlink transmit beams in set B. The relationship between set B and set A can be varied.

[0045] Set B can be a subset of set A. Set B can be understood as a partial subset of the downlink transmit beams. Set A can be understood as the complete set of downlink transmit beams. The optimal beam in the complete set can be predicted by measuring the subset, thereby reducing overhead.

[0046] Set B and set A can also be two different beam sets. For example, set B can be a set of synchronization signal block (SSB) resources with a small number of beams (each SSB resource has a large spatial coverage area), while set A can be a set of channel state information-reference signal (CSI-RS) resources with a large number of beams (each CSI-RS resource has a small spatial coverage area).

[0047] In some embodiments, beam prediction in the spatial domain can be implemented using different sub-models.

[0048] Referring to Figure 2, which illustrates the relationship between the input and output of the transmit beam prediction model, this model can be considered to solve a multi-class classification problem. The relationship between the model's input and output is that of a subset of the L1-RSRP input to the optimal K transmit beams, where the subset of transmit beam measurements (a portion of the total L1-RSRP measurements) serves as the model's input. The model's output is the optimal K transmit beam indices selected from the entire set, i.e., the K transmit beams with the highest L1-RSRP. The labels used by the model are the optimal (i.e., highest L1-RSRP) K transmit beam indices measured in the entire set.

[0049] Figure 3 illustrates the optimal beam quality prediction model. This model can be considered to solve a linear regression problem. The relationship between the model's input and output is the relationship from a subset of input L1-RSRPs to the optimal L1-RSRPs of the K transmit beams. The input to the optimal beam quality prediction model in Figure 3 is the same as the input to the transmit beam prediction model in Figure 2, the difference being that the output of the optimal beam quality prediction model is K (K>=1) optimal L1-RSRPs. The labels used in the optimal beam quality prediction model are the optimal K L1-RSRPs measured in the full set, and the corresponding K transmit beam indices.

[0050] Beam prediction in the time domain (BM - Use Case 2)

[0051] Time-domain beam prediction can be understood as: predicting the optimal beam (selected from set A) for one or more future moments by measuring the transmitted beams in set B at one or more historical moments.

[0052] For use cases of purely time-domain beam prediction, set B and set A can be the same set. Of course, for use cases that combine time-domain and spatial-domain beam prediction, set B can be a subset of set A. Alternatively, set B can be a different set than set A.

[0053] Figure 4 illustrates the measurement window and prediction window in beam prediction in the time domain. Figure 4 shows the downlink historical measurement window with a time interval of 100 ms, and the prediction window, also with a time interval of 100 ms, used to predict the optimal future beam.

[0054] In some embodiments, a long short-term memory (LSTM) network model can be used for time-domain beam prediction.

[0055] In the above description, "beam" may include or be replaced by: a space transmission filter. A downlink reference signal can be used to associate the transmit beam. The downlink reference signal may be, for example, CSI-RS or SSB. During beam indication, a transmission configuration indicator (TCI) status can be used to refer to the transmit beam. The TCI status may include the downlink reference signal.

[0056] Use cases for other terminal device-side models in New Radio (NR).

[0057] Aside from beam prediction, 3GPP conducted research on channel state information (CSI) prediction in Releases 18 and 19. The purpose of CSI prediction is to address the problem of CSI obsolescence caused by dynamic changes in the wireless environment. Traditional CSI feedback methods require frequent reporting of the current channel state, which incurs significant overhead in high-mobility scenarios. By predicting future CSI, the system can maintain effective control over the channel state without requiring frequent reporting.

[0058] The models used for prediction on terminal devices can be commonly used deep learning models. These deep learning models include, but are not limited to, recurrent neural networks (RNNs), long short-term memory networks (LSTMs), and gated recurrent units (GRUs). These models are suitable for processing time-series data and can predict future CSI states based on historical CSI measurements. Furthermore, some studies have shown that convolutional neural networks (CNNs) can also effectively capture correlations between the time, spatial, and frequency domains, thereby improving prediction performance.

[0059] 3GPP specifies the use of a single-sided model on the terminal equipment side for CSI prediction.

[0060] When a terminal device uses a model for CSI prediction, the model's input is historical CSI measurements. That is, the model's input is typically a series of past CSI measurements. These measurements reflect how the channel state changes over time. See Figure 5, which illustrates a schematic diagram of CSI prediction. As shown in Figure 5, the model's input can be CSI data from multiple (four in Figure 5) historical measurement times. Here, the CSI measurements can be the H matrix of the equivalent channel. Alternatively, the CSI measurements can be the singular value vector or eigenvector obtained by performing singular value or eigenvalue decomposition on the channel matrix.

[0061] When a terminal device uses a model to predict CSI, the model's output is the predicted CSI value for the future. In other words, the model's output is a prediction of the CSI at one or more future times. Similarly, the predicted CSI value here can be the H matrix of the equivalent channel. Alternatively, the predicted CSI value can also be the eigenvector obtained by eigenvalue decomposition of the channel matrix.

[0062] Model consistency between the training and inference phases

[0063] The training and inference phases of a model may involve network-side conditions and network-side additional conditions. In other words, during the training phase, the model is trained or data is collected under certain network-side and additional conditions. During the inference phase, the model is used for inference under those same conditions.

[0064] NW-side conditions may include, but are not limited to: the time-frequency domain location of the measurement resources corresponding to the measurement set, and the number of resources (i.e., the number of beams) included in the prediction set. Network devices can configure NW-side conditions via radio resource control (RRC) signaling.

[0065] Additional conditions on the NW side can include, but are not limited to: the beam pattern of the measurement set, the beam pattern of the prediction set, and the downtilt angle of the base station antenna. Here, beam pattern can be understood as whether a wide beam or a narrow beam is used. NW-side additional conditions pertain to the specific implementation of the NW-side equipment. Network equipment operators and vendors tend to avoid exposing the specific implementation details of the NW-side wireless environment. In other words, NW-side additional conditions are generally not directly configured on the NW side. Therefore, network equipment typically does not directly use RRC signaling to configure NW-side additional conditions.

[0066] To ensure the performance of terminal devices in inference using the model, it is necessary to consider the consistency of NW-side conditions and additional NW-side conditions during the training and inference phases. Only when the NW-side conditions during training and inference are consistent, and both the additional NW-side conditions during training and inference are consistent, can the performance of terminal devices in inference using the model trained during the training phase be guaranteed.

[0067] As described above regarding NW-side conditions and additional NW-side conditions, network devices can directly configure NW-side conditions, thus ensuring consistency between the training and inference phases. However, network devices typically do not directly configure additional NW-side conditions, making it difficult to guarantee consistency between the training and inference phases. Therefore, ensuring consistent NW-side conditions between the training and inference phases is a technical problem that needs to be addressed.

[0068] To address the aforementioned technical issues, associated identification (Associated ID) has been introduced in related technologies. 3GPP protocols specify that Associated IDs can be used to replace additional conditions that cannot be directly configured on the NW side (i.e., NW-side additional conditions). The Associated ID serves as an indicator, informing the terminal device of the consistency of NW-side additional conditions during the model training and inference phases. For example, during the data collection phase (for model training), the network device configures an Associated ID of 2, and the terminal device trains a model for beam prediction and deploys the model on its side. When the terminal device enters the inference phase, the network device configures at least one of the measurement and prediction sets required by the model (corresponding to NW-side conditions), along with an Associated ID of 2. The terminal device can then consider that the NW-side additional conditions of the model during the inference phase are consistent with those during the training phase, and therefore determines that the model can be used for beam prediction. See below for protocol details.

[0069] Further research is needed on the consistency of NW-side additional conditions during the training and inference phases of the UE-side models used in BM-Case 1 and BM-Case 2, where NW-side additional conditions may at least affect the UE's assumptions about beamforming for set A or set B:

[0070] Option 1: Based on associated ID

[0071] For future research: What assumptions can the UE make when the training and inference phases have the same association ID?

[0072] For future research: How to introduce associated IDs, for example, within or outside the CSI framework.

[0073] Option 2: Based on performance monitoring

[0074] For future research: details

[0075] Other options are not excluded.

[0076] When the same association ID is configured under the CSI framework, the UE can assume that the downlink beam set or list has similar properties during the training and inference phases. See below for protocol details.

[0077] For the UE-side model in beam management, it supports associated ID.

[0078] [Work Assumptions]

[0079] The associated ID can be configured at least within the CSI framework.

[0080] For future research: Detailed information

[0081] For future research: on whether / how to configure / indicate associated IDs via other signals and / or in other processes / frameworks.

[0082] The UE may assume that DL Rx beams or beam sets / lists associated with the same association ID have similar properties.

[0083] For future research: Whether / how to define the "similar properties" of DL Rx beams or beam sets / lists.

[0084] Furthermore, the protocol stipulates that associated IDs can be used within the same cell to ensure the consistency of additional conditions on the NW side during the training and inference phases. See below for the protocol details.

[0085] Confirm the following working assumptions

[0086] Working assumptions

[0087] Regarding the associated ID of Rel-19, the UE assumes that the additional conditions on the NW side with the same associated ID are consistent at least within the cell.

[0088] For future research: Whether / how the UE's assumptions apply to multiple cells (including feasibility studies)

[0089] As can be seen from the above description, the association ID is only used within a single cell to ensure the consistency of the NW-side additional conditions during the training and inference phases. In other words, the association ID is not uniformly encoded across multiple cells.

[0090] For example, the association ID #1 in cell A and the association ID #1 in cell B may represent different NW-side additional conditions, such as different beam patterns, different antenna panel structures, and different antenna downtilt angles. When the terminal device moves from cell A (used for model training and configured with association ID #1) to cell B (used for model inference and configured with another association ID #1), the fact that the association ID in the inference phase is the same as the association ID in the training phase does not guarantee the consistency of the NW-side additional conditions between the training and inference phases.

[0091] Therefore, when using the associated ID proposed in related technologies, it is difficult to guarantee the consistency of the NW-side additional conditions during the inference and training phases when the terminal device is trained and inferred in different cells. This may reduce the performance of model inference and limit the generalization ability of the model.

[0092] Furthermore, the relevant technologies only address how to determine the consistency of the NW-side additional conditions during the training and inference phases when association IDs are configured in both phases. For other scenarios, such as when association IDs are not configured during training, inference, or neither phase, the relevant technologies do not explicitly address how to determine the consistency of the NW-side additional conditions in these cases.

[0093] To address the aforementioned technical problems, embodiments of this application provide a communication method. This communication method introduces a first condition, which is associated with one or more of the following: training data of a first model, first configuration information, and second configuration information. The first condition assists the terminal device in determining whether to use the first model for inference within a first cell, thereby ensuring the performance of the terminal device in using the first model for inference within the first cell.

[0094] The communication method provided in this application can be applied to a single-sided model on the terminal device side. A "single-sided model on the terminal device side" can be understood as a model trained by the terminal device and used by the terminal device for inference. For example, the single-sided model on the terminal device side is the beam prediction model or CSI prediction model mentioned above. Of course, the single-sided model on the terminal device side can also be other models, such as a model used for positioning (hereinafter referred to as a positioning model). This positioning model can infer the location information of the terminal device by analyzing wireless channel characteristics (such as channel impulse response or power delay profile).

[0095] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0096] Figure 5 is a schematic flowchart of the communication method provided in an embodiment of this application. Referring to Figure 5, the communication method provided in an embodiment of this application may include the following step S510.

[0097] In step S510, if the first condition is met, the terminal device determines to use the first model for inference in the first cell.

[0098] The first cell here can be understood as the cell currently used for inference. The first model can be deployed on the terminal device side. When the terminal device is a UE, the first model can also be called the UE-side model. This application embodiment does not specifically limit the type of the first model. The first model can be an AI model or an LM model. For example, the first model can be any of the following: RNN, LSTM, GRU, CNN, etc. The first model here can have or perform certain functions. For ease of description, the function performed by the first model is referred to as the first function below. The first function can include, but is not limited to, the following functions: beam prediction function, CSI prediction function, etc. When the first model can perform the beam prediction function, the first model can also be called a beam prediction model. When the first model can perform the CSI prediction function, the first model can also be called a CSI prediction model.

[0099] Terminal devices can obtain the first model in various ways. In some embodiments, the terminal device can download the first model from an OTT (over-the-top) server. For example, the UE downloads a beam prediction model from an OTT server. In other embodiments, the terminal device can obtain the first model through a model delivery / transfer function from the network device to the terminal device. For example, the UE obtains the beam prediction model through a model delivery / transfer function (from the network device to the UE). In still other embodiments, the terminal device can obtain the first model through training.

[0100] If the first condition is met, the terminal device may consider the first model suitable for inference in the first cell. In other words, the terminal device determines that the first model can be used for inference in the first cell. When the first model is used to perform prediction functions, "using the first model for inference" can include or be replaced by: using the first model for prediction. It should be understood that "if the first condition is met, the terminal device determines that it will use the first model for inference in the first cell" does not mean that the terminal device will use the first model for inference in the first cell if the first condition is met. Whether the terminal device uses the first model for inference in the first cell if the first condition is met depends on actual needs. The role of the first condition is to assist the terminal device in determining whether the first model is suitable for inference in the first cell, not to trigger the terminal device to use the first model for inference in the first cell.

[0101] In some implementations, the first condition can be associated with the training data of the first model. The training data of the first model can be understood as the data used in the process of training the first model. That is, the first model is trained based on this training data. In some cases, the first model can be trained based on data from a single cell. That is, the training data of the first model can include data from one cell. In other cases, the first model can be trained based on data from multiple cells. That is, the training data of the first model can include data from multiple cells. When the first model is trained based on data from a single cell, it can be trained based on data from the first cell. Alternatively, the first model can also be trained based on data from cells other than the first cell. When the first model is trained based on data from multiple cells, these multiple cells can include the first cell. Alternatively, these multiple cells may not include the first cell. The training data can be data collected by the terminal device. The terminal device can collect the training data of the first model in various ways. Optionally, in some embodiments, the terminal device can collect data from one or more cells as training data for the first model. Optionally, in other embodiments, the terminal device can download a dataset from a network device or an OTT server as training data for the first model. This dataset can include data collected from one or more cells.

[0102] In some implementations, the first condition can be associated with first configuration information. This first configuration information can be sent from the network device to the terminal device. Optionally, in some embodiments, the first configuration information can be used to indicate network device-side conditions during the model's inference phase. Optionally, in other embodiments, the first configuration information can be used to indicate a first association identifier. Optionally, in other embodiments, the first configuration information can be used to indicate both the network device-side conditions and the first association identifier during the model's inference phase. For a description of network device-side conditions, please refer to the relevant sections above; they will not be repeated here. The first association identifier can be used to indicate additional network device-side conditions during the inference phase. The first association identifier can, for example, be the association ID mentioned above.

[0103] In other implementations, the first condition can be associated with second configuration information. This second configuration information can be sent from the network device to the terminal device. Optionally, in some embodiments, the second configuration information can be used to indicate network device-side conditions during the model's training phase. Optionally, in other embodiments, the second configuration information can be used to indicate a second association identifier. Optionally, in other embodiments, the second configuration information can be used to indicate both network device-side conditions and a second association identifier during the model's training phase. The second association identifier can be used to indicate additional network device-side conditions during the training phase. The second association identifier can, for example, be the association ID mentioned above.

[0104] As described in step S510, this embodiment introduces a first condition for inference by the terminal device in the first cell. The first condition is associated with one or more of the following: training data of the first model, first configuration information, and second configuration information. The first condition can assist the terminal device in determining whether to use the first model for inference in the first cell, thereby ensuring the performance of the terminal device in inference using the first model in the first cell.

[0105] As mentioned earlier, terminal devices can perform training and inference within the same cell or across different cells. In the scenario where the terminal device performs training and inference within the same cell, to ensure performance when using the first model for inference in the first cell, intra-cell consistency needs to be considered. In the scenario where the terminal device performs training and inference across different cells, to ensure performance when using the first model for inference in the first cell, inter-cell consistency needs to be considered. The methods provided in this application's embodiments are described below for both scenarios: training and inference within the same cell and training and inference across different cells.

[0106] Scenario 1: Terminal devices perform training and inference in different cells.

[0107] Scenario 1 can include the following three situations.

[0108] Scenario 1: The first model is trained based on data from multiple cells.

[0109] Scenario 2: The first model is trained based on data from other cells (hereinafter referred to as the second cell) that are different from the first cell.

[0110] Scenario 3: Configure association identifiers uniformly for multiple communities (hereinafter referred to as the first community group).

[0111] The following sections will provide detailed explanations for each of these three scenarios.

[0112] Scenario 1: The first model is trained based on data from multiple cells.

[0113] In this case, the first condition may include one or more of the following: the data of multiple cells includes the data of the first cell; the association identifier corresponding to the data of the first cell is the same as the first association identifier.

[0114] "Data from multiple cells" can be understood as a dataset from multiple cells, or a mixed dataset from multiple cells. Optionally, in some embodiments, the terminal device can collect data in multiple different cells to obtain data from multiple cells. Alternatively, the terminal device can undergo a data collection process from multiple different cells to obtain data from multiple cells. The terminal device can collect data from multiple cells during the model training phase to facilitate training based on this data to obtain a first model. Optionally, in other embodiments, the terminal device can download a dataset from a network device or an OTT server, which may include data from multiple cells. The terminal device can train based on this data to obtain a first model. Although the NW-side attachment conditions of multiple cells may be inconsistent, the model trained based on data from multiple cells has better generalization ability and helps avoid the problem of inconsistent NW-side attachment conditions between the training and inference phases.

[0115] Optionally, in some embodiments, the data from multiple cells can be represented in the following form: {Cell ID#1, Associated ID#A, measurement results of measurement set (Set B), measurement results of prediction set (Set A)}; {Cell ID#2, Associated ID#B, measurement results of measurement set (Set B), measurement results of prediction set (Set A)}; ...; {Cell ID#N, Associated ID#X, measurement results of measurement set (Set B), measurement results of prediction set (Set A)}. The measurement results of the measurement set can be used as input to the model, and the measurement results of the prediction set can be used as labels for model training, which belongs to the supervised learning method.

[0116] The presence or absence of data from the first cell can be determined based on the cell IDs. Each of the multiple cells can have a cell ID. If the cell IDs of the multiple cells include the cell ID of the first cell, then the data from the multiple cells can be considered to include the data from the first cell. These cell IDs can be included in the dataset of the multiple cells. The cell ID can be included or replaced with a cell ID. The cell ID can be configured via system messages when the terminal device accesses the cell. For example, suppose the cell ID of the first cell is Cell ID#2, the first model is trained based on data from multiple cells, and the cell IDs corresponding to these multiple cells are Cell ID#1, Cell ID#2, ..., Cell ID#N, where N is an integer greater than or equal to 2. Since the cell IDs of the multiple cells include the cell ID of the first cell, i.e., Cell ID#2, it can be considered that the data from these multiple cells includes the data from the first cell.

[0117] The cell number mentioned in this application embodiment can be a local cell number, or a local cell index. Alternatively, the cell number mentioned in this application embodiment can also be a global cell number, or a global cell index. A local cell number can be, for example, a physical cell identifier (PCI). The PCI value can range from 0 to 1007, representing a total of 1008 IDs. The PCI can be carried by the synchronization signal block (SSB) of each cell. A global cell number can be, for example, the cell global identifier (CGI) in NR, i.e., the NR cell global identifier (NCGI).

[0118] The first association identifier mentioned here can be used to indicate the NW-side additional conditions during the inference phase of the model. The first association identifier can be configured to the terminal device along with the measurement resources required for inference (i.e., NW-side conditions). The association identifier corresponding to the data of the first cell can be used to indicate the NW-side additional conditions used by the terminal device when training the model using the data of the first cell. The association identifier corresponding to the data of the first cell can be included in the datasets of multiple cells. It can be determined whether the first association identifier is the same as the association identifier corresponding to the data of the first cell based on the datasets of multiple cells. If the first association identifier is the same as the association identifier corresponding to the data of the first cell, it indicates that the NW-side additional conditions used by the terminal device when training the model using the data of the first cell may be the same as the NW-side additional conditions used by the terminal device when using the first model for inference in the first cell.

[0119] In some implementations, the first condition may include: the data from multiple cells includes the data from the first cell. That is, when the data from multiple cells includes the data from the first cell, the terminal device can assume that the first model is consistent during the training and inference phases, and thus determine that the first model should be used for inference in the first cell. For example, during the training phase, the network device configures the same NW-side additional conditions for the data from multiple cells, and the terminal device trains the first model based on the data from these multiple cells and the NW-side additional conditions. During the inference phase, the network device configures the same NW-side additional conditions for the terminal device's inference in the first cell. In this case, if the cell index of the first cell is included in the cell indices of the multiple cells, the terminal device can assume that the first model is consistent during the training and inference phases, and thus determine that the first model should be used for inference in the first cell.

[0120] In some implementations, the first condition may include: data from multiple cells includes data from the first cell, and the association identifier corresponding to the data from the first cell is the same as the first association identifier. That is, if data from multiple cells includes data from the first cell, and the association identifier corresponding to the data from the first cell is the same as the first association identifier, the terminal device can consider the first model to be consistent during the training and inference phases, thus determining that the first model is used for inference in the first cell. For example, the mixed dataset of multiple cells used to train the first model may be {Cell ID#1, Associated ID#A, measurement results of measurement set (Set B), measurement results of prediction set (Set A)}; {Cell ID#2, Associated ID#B, measurement results of measurement set (Set B), measurement results of prediction set (Set A)}; ...; {Cell ID#N, Associated ID#X, measurement results of measurement set (Set B), measurement results of prediction set (Set A)}. If the cell identifier of the first cell is Cell ID#2, and the association identifier configured by the network device during the inference phase is Associated ID#B. Since the mixed dataset includes a combination of Cell ID#2 and Associated ID#B, the terminal device can assume that the first model is consistent in both the training and inference phases, and thus determine that the first model is used for inference in the first cell.

[0121] Scenario 2: The first model is trained based on data from the second cell.

[0122] In this case, the first condition may include: the association identifier corresponding to the data of the second cell has a mapping relationship with the first association identifier.

[0123] The "association identifier corresponding to the data of the second cell" can be used to indicate the additional conditions on the NW side used by the terminal device when training the model based on the data of the second cell.

[0124] The first association identifier here can be understood as the association identifier corresponding to the first cell. This first association identifier can be used to indicate the NW-side additional conditions corresponding to the first cell during the inference phase. Alternatively, this first association identifier can be used to indicate the NW-side additional conditions used by the terminal device when performing inference in the first cell.

[0125] For a specific cell, association identifiers can be used to encode the additional conditions on the cell's NW side. For example, association identifiers can range from #A to #Z, corresponding to 26 different NW side additional conditions.

[0126] As mentioned earlier, the association identifier within one cell cannot directly (or by default) correspond to the same association identifier in another cell. In other words, even if two different cells have the same association identifier, it does not mean that the NW-side additional conditions for those two cells are consistent. For example, if the association identifier for data in the second cell is the same as that in the first cell, it does not mean that the NW-side additional conditions for the first cell are consistent with those for the second cell.

[0127] Based on the above understanding, a mapping relationship can be established between the associated identifiers of different cells. This mapping relationship can be indicated by mapping relationship information. This mapping relationship information can be configured via RRC signaling. Alternatively, this mapping relationship information can also be configured via activation or deactivation signaling of the media access control-control element (MAC CE). Optionally, in some embodiments, the mapping relationship indicated by the mapping relationship information can be a one-to-one mapping relationship. That is, one associated identifier of one cell can correspond to one associated identifier of another cell. Or, one associated identifier of one cell can have a mapping relationship with one associated identifier of another cell. Optionally, in other embodiments, the mapping relationship indicated by the mapping relationship information can be a one-to-many mapping relationship. That is, one associated identifier of one cell can correspond to multiple associated identifiers of multiple other cells. Or, one associated identifier of one cell can have a mapping relationship with multiple associated identifiers of multiple other cells.

[0128] The terminal device can determine whether to use the first model for inference in the first cell based on the mapping relationship information mentioned above. If the mapping relationship information determines that the association identifier corresponding to the data in the second cell has a mapping relationship with the first association identifier, then the terminal device can determine that the NW-side additional conditions used when using the data in the second cell for model training are consistent with the NW-side additional conditions used when using the first model for inference in the first cell. Therefore, the terminal device can determine to use the first model for inference in the first cell.

[0129] The mapping relationship information here can include one mapping relationship. Alternatively, it can include multiple mapping relationship information. Each mapping relationship can be used to indicate a mapping relationship. Each mapping relationship can include the cell identifiers corresponding to multiple cells that have mapping relationships with each other, as well as the association identifiers corresponding to each of these cells. Based on the cell identifiers and association identifiers in the mapping relationship information, it can be determined which cells correspond to which association identifiers have mapping relationships. That is, based on the cell identifiers and association identifiers in the mapping relationship information, it can be determined which cells correspond to which association identifiers indicate NW-side additional conditions that are consistent. For a detailed introduction to cell identifiers, please refer to the relevant sections above; it will not be repeated here.

[0130] As a concrete example, two mapping relationships can be configured via RRC configuration signaling. The first mapping relationship is {Cell ID#1, Associated ID#A}, {Cell ID#2, Associated ID#B}, {Cell ID#5, Associated ID#Q}. The second mapping relationship is {Cell ID#4, Associated ID#D}, {Cell ID#5, Associated ID#E}, {Cell ID#6, Associated ID#H}, {Cell ID#7, Associated ID#B}. Based on the first mapping relationship, the terminal device can determine that there are mapping relationships between the associated identifiers: Associated ID#A (cell with Cell ID#1), Associated ID#B (cell with Cell ID#2), and Associated ID#Q (cell with Cell ID#5). In other words, the NW-side additional conditions indicated by the associated identifier Associated ID#A for cell with Cell ID#1, the NW-side additional conditions indicated by the associated identifier Associated ID#B for cell with Cell ID#2, and the NW-side additional conditions indicated by the associated identifier Associated ID#Q for cell with Cell ID#5 are consistent. Based on the second mapping relationship information, it can be determined that there are mapping relationships between the associated identifiers Associated ID#D for cell with Cell ID#4, Associated ID#E for cell with Cell ID#5, Associated ID#H for cell with Cell ID#6, and Associated ID#B for cell with Cell ID#7. In other words, the NW-side additional conditions indicated by the associated ID#D corresponding to the cell with cell ID#4, the NW-side additional conditions indicated by the associated ID#E corresponding to the cell with cell ID#5, the NW-side additional conditions indicated by the associated ID#H corresponding to the cell with cell ID#6, and the NW-side additional conditions indicated by the associated ID#B corresponding to the cell with cell ID#7 are consistent.During the inference phase, the terminal device trains a first model based on data from the cell with cell ID#1 and the associated identifier Associated ID#A. During the inference phase, the terminal device performs inference in the cell with cell ID#2, and the network device is configured with the associated identifier Associated ID#B. Based on the aforementioned first mapping relationship, the terminal device can determine to use the first model for inference in the cell with cell ID#2.

[0131] Scenario 3: Configure a unified association identifier for the first cell group.

[0132] In scenarios one and two described above, the association identifier was configured individually for each cell. In scenario three, the network device can uniformly configure or allocate association identifiers for certain cells. "Network device uniformly configures association identifiers for certain cells" can be understood as: the network device uniformly configures association identifiers for these cells during both the training and inference phases. In other words, for these cells, the same association identifier represents the same NW-side additional conditions. For ease of description, these cells will be referred to as the first cell group below.

[0133] In the third case, the first condition may include: the first association identifier corresponds to the first cell group, and the first cell group includes the first cell. That is, if the first association identifier corresponds to the first cell group and the first cell group includes the first cell, the terminal device can determine that the first model is used for inference in the first cell.

[0134] As mentioned earlier, the first association identifier can be used to indicate additional NW-side conditions during the inference phase. In the third case, since the network equipment uniformly configures the association identifier for the first cell group, the first association identifier can be used to indicate additional NW-side conditions for the first cell group. That is to say, the first association identifier corresponds to the first cell group.

[0135] The cells included in the first cell group can be determined based on the network device configuration information. In other words, the grouping method of the cells can be determined based on the network device configuration information, that is, which cells are assigned to the same cell group. For example, the network device can group multiple cells with similar wireless environment deployments into the same cell group, allowing these cells to share the same association identifier for the same NW-side additional conditions.

[0136] Whether a first cell group includes a first cell can be determined based on the cell identifier. If multiple cell identifiers corresponding to the first cell group include the cell identifier of the first cell, then the first cell group can be considered to include the first cell. If multiple cell identifiers corresponding to the first cell group do not include the cell identifier of the first cell, then the first cell group can be considered to not include the first cell. The multiple cell identifiers corresponding to the first cell group can be indicated by the network device through a configuration list. That is, the network device can indicate the cell identifiers of multiple cells belonging to the same cell group through a configuration list. As a specific example, the list configured by the network device is {Cell ID#1, Cell ID#2, Cell ID#5}, meaning the first cell group includes the cell with cell ID#1, the cell with cell ID#2, and the cell with cell ID#5. If the cell identifier of the first cell is Cell ID#5, then it can be determined that the first cell group includes the first cell. If the cell identifier of the first cell is Cell ID#7, then it can be determined that the first cell group does not include the first cell.

[0137] For multiple cells in the first cell group, the same association identifier indicates the same NW-side additional conditions. Taking the network device configuration list as {Cell ID#1, Cell ID#2, Cell ID#5} as an example, the same association identifier, such as Associated ID#3, represents the same NW-side additional conditions. That is, the NW-side additional conditions indicated by the association identifier Associated ID#3 for the cell with cell ID#1, the cell with cell ID#2, and the cell with cell ID#5 are the same. For cells not in the above list, such as {Cell ID#7, Cell ID#8, Cell ID#9}, the same association identifier, such as Associated ID#3, cannot represent the same NW-side additional conditions. In other words, the NW-side additional conditions indicated by the associated ID#3 corresponding to the cell with cell ID#7, the NW-side additional conditions indicated by the associated ID#3 corresponding to the cell with cell ID#8, and the NW-side additional conditions indicated by the associated ID#3 corresponding to the cell with cell ID#9 may be different.

[0138] When a first associated identifier corresponds to a first cell group and the first cell group includes the first cell, it can be shown that the first model is consistent between the training and inference phases, thus allowing the terminal device to determine whether to use the first model for inference in the first cell. For example, the network device is configured with a cell list of {Cell ID#1, Cell ID#2, Cell ID#5}. During the training phase, the terminal device trains the first model based on data from cells with cell ID#1 or Cell ID#2. During the inference phase, the terminal device performs inference in the cell with cell ID#5. Since Cell ID#5 is included in the cell list configured by the network device, the terminal device considers the first model consistent between the training and inference phases, and therefore determines to use the first model for inference in the cell with cell ID#5.

[0139] The method provided by the embodiments of this application has been described above for scenario one. The method provided by the embodiments of this application will be described below for scenario two.

[0140] Scenario 2: Terminal devices perform training and inference within the same cell.

[0141] In Scenario 2, since the model's training and inference take place in the same cell (hereinafter referred to as the first cell), it is not necessary to determine whether the first condition is met based on the cell identifier mentioned in Scenario 1. Because the association identifier may be an optional configuration in both the model's training and inference phases, Scenario 2 can include the following four cases.

[0142] Scenario 2 can include the following four situations.

[0143] Scenario 1: The first configuration information indicates the first association identifier, and the second configuration information indicates the second association identifier.

[0144] Scenario 2: The first configuration information does not indicate the first association identifier, but the second configuration information indicates the second association identifier.

[0145] Scenario 3: The first configuration information indicates the first association identifier, but the second configuration information does not indicate the second association identifier.

[0146] Scenario 4: The first configuration information does not indicate the first association identifier, and the second configuration information does not indicate the second association identifier.

[0147] The following sections will provide detailed explanations of the four scenarios in Scenario 2.

[0148] Scenario 1: The first configuration information indicates the first association identifier, and the second configuration information indicates the second association identifier.

[0149] In Scenario 1, the network device configures an association identifier during both the training and inference phases. In Scenario 1, the first condition can include: the first association identifier and the second association identifier are the same. That is, if the first association identifier and the second association identifier are the same, the terminal device can determine that the first model is used for inference in the first cell. Here, the first association identifier indicates the additional conditions on the network device side during the inference phase. Here, the second association identifier indicates the additional conditions on the network device side during the training phase. Since training and inference are performed within the same cell, the fact that the first and second association identifiers are the same indicates that the additional conditions on the network device side during the inference phase are the same as those during the training phase. In other words, the fact that the first and second association identifiers are the same indicates that the additional conditions on the network device side during the inference phase are consistent with those during the training phase. Therefore, if the first and second association identifiers are the same, the terminal device can determine that the first model is used for inference in the first cell.

[0150] For example, during the training phase (data collection phase), the network device configures the measurement resources (i.e., NW-side conditions) for the training dataset and an association identifier (to characterize the additional NW-side conditions). The terminal device trains a model based on the dataset resources and the association identifier. This model can be used for beam prediction, CSI prediction, etc. During the inference phase, the network device similarly configures the measurement resources (i.e., NW-side conditions) and an association identifier (to characterize the additional NW-side conditions) for inference. If the association identifier configured by the network device during the inference phase is the same as the association identifier configured during the training phase, the terminal device determines that it can use the model obtained during the training phase for inference.

[0151] For example, during the training phase (data collection phase), the network device configures measurement resources (i.e., NW-side conditions) for the training dataset and configures multiple association identifiers (to characterize additional NW-side conditions). The terminal device trains multiple models based on the dataset resources and the multiple association identifiers. This model can be used for beam prediction, CSI prediction, etc. During the inference phase, the network device similarly configures measurement resources (i.e., NW-side conditions) and multiple association identifiers (to characterize additional NW-side conditions) for inference. If the multiple association identifiers configured by the network device during the inference phase and the multiple association identifiers configured by the network device during the training phase both include a certain association identifier (e.g., association identifier #1), then the terminal device can determine to use the model corresponding to association identifier #1 from the multiple models obtained during the training phase for inference. The model corresponding to association identifier #1 is the model trained based on association identifier #1.

[0152] Scenario 2: The first configuration information does not indicate the first association identifier, but the second configuration information indicates the second association identifier.

[0153] In scenario two, the network device configures association identifiers during the training phase but not during the inference phase. For example, during training, the network device configures the measurement resources (i.e., NW-side conditions) for the dataset and configures one or more association identifiers (representing additional NW-side conditions). The terminal device trains one or more models based on the dataset resources and one or more association identifiers. During inference, the network device configures the measurement resources (i.e., NW-side conditions) for inference but does not configure association identifiers (representing additional NW-side conditions).

[0154] Scenario 3: The first configuration information indicates the first association identifier, but the second configuration information does not indicate the second association identifier.

[0155] In scenario three, the network device did not configure association identifiers during the training phase but configured them during the inference phase. For example, during training, the network device configured one or more sets of measurement resources for the dataset (i.e., NW-side conditions) but did not configure association identifiers (representing additional NW-side conditions). The terminal device trained one or more models based on the measurement resources of one or more datasets. During inference, the network device configured the measurement resources (i.e., NW-side conditions) for inference and also configured association identifiers (representing additional NW-side conditions).

[0156] In scenario three, because the association identifier configured during the inference phase cannot correspond to the association identifier configured during the training phase, the terminal device cannot recognize the association identifier, and therefore the association identifier configured during the inference phase is ineffective. Scenario three can be considered equivalent to not configuring an association identifier during the inference phase.

[0157] Scenario 4: The first configuration information does not indicate the first association identifier, and the second configuration information does not indicate the second association identifier.

[0158] In scenario four, the network device does not configure association identifiers during either the training or inference phases. For example, during training, the network device configures the measurement resources (i.e., NW-side conditions) for the dataset, but does not configure association identifiers (representing additional NW-side conditions). The terminal device trains one or more models based on the measurement resources of one or more datasets. During inference, the network device configures the measurement resources (i.e., NW-side conditions) for inference, but does not configure association identifiers (representing additional NW-side conditions).

[0159] For scenarios two, three, and four, the first condition can be associated with the number of models trained on the terminal device to perform the first function.

[0160] The first function here can be understood as the function performed by the first model. For example, the first function could be a beam prediction function or a CSI prediction function. The terminal device can train one model to perform the first function. Alternatively, the terminal device can train multiple models to perform the first function.

[0161] In some implementations, the first condition may include: the number of models trained by the terminal device to perform the first function is 1.

[0162] Optionally, in some embodiments, if the number of models used to perform the first function is 1, the terminal device may assume that the first model has consistent NW-side additional conditions during the training and inference phases. Therefore, when the number of models used to perform the first function is 1, the terminal device can determine that the first model is used for inference in the first cell. This embodiment can be applied to case two of scenario two.

[0163] For example, during the training phase, the network device is configured with an association identifier, and the terminal device trains a beam prediction model based on the data from the first cell and the association identifier configured by the network device. During the inference phase, even if the network device is not configured with an association identifier, the terminal device can assume that the beam prediction model has consistent NW-side conditions between the training and inference phases. Therefore, the terminal device can use this beam prediction model for beam prediction within the first cell.

[0164] Optionally, in other embodiments, if the number of models used to perform the first function is 1, the terminal device can further determine whether the NW-side conditions of the model during the training phase match the NW-side conditions indicated by the first configuration information. The NW-side conditions of the model during the training phase may include the model's training dataset resources. The NW-side conditions indicated by the first configuration information may include the inference resources configured by the network device. The NW-side conditions of the model during the training phase can be determined based on the model's training dataset. If the NW-side conditions of the model during the training phase match the NW-side conditions indicated by the first configuration information, it indicates that the NW-side conditions of the model are consistent between the training and inference phases. Therefore, if the NW-side conditions of the model during the training phase match the NW-side conditions indicated by the first configuration information, the terminal device can use the model for inference in the first cell. This embodiment can be applied to scenarios three and four of scenario two.

[0165] For example, during the training phase, the network device is configured with the following training dataset: a measurement set (Set B) containing 8 resources and a prediction set (Set A) containing 64 resources. The terminal device trains a beam prediction model based on this training dataset. During the inference phase, the network device is configured with the following inference resources: a measurement set (Set B) containing 8 resources and a prediction set (Set A) containing 64 resources. Since the configured inference resources are the same as the training dataset resources of the beam prediction model, the terminal device can use the beam prediction model to perform beam prediction in the first cell.

[0166] The above discussion covered how to determine if the first condition is met when there is only one model used to perform the first function. The following discussion covers how to determine if the first condition is met when there are multiple models used to perform the first function.

[0167] When there are multiple models used to perform the first function, the NW-side conditions of each model during the training phase can be determined based on multiple configuration information. In other words, multiple models can correspond to multiple configuration information. The NW-side conditions of the first model during the training phase can be determined based on a third configuration information.

[0168] In some implementations, the first condition may include: the third configuration information matches the first configuration information. As mentioned earlier, when there are multiple models used to perform the first function, the NW-side conditions of the multiple models during the training phase can be determined based on multiple configuration information of the models. The NW-side conditions of the first model during the training phase can be determined based on the third configuration information. That is, the third configuration information can be used to indicate the NW-side conditions of the first model during the training phase. And the first configuration information can be used to indicate the NW-side conditions of the first model during the inference phase. Therefore, "the third configuration information matches the first configuration information" can indicate that the NW-side conditions of the first model during the inference phase are consistent with the NW-side conditions of the first model during the training phase. Therefore, when the third configuration information matches the first configuration information, the terminal device can determine that the first model is used for inference in the first cell. That is, when there are multiple models used to perform the first function, the terminal device can select the model corresponding to the third configuration information from multiple models as the first model. Or, the terminal device can select the model corresponding to the NW-side conditions that match the NW-side conditions during the inference phase from multiple models as the first model. This implementation can be applied to cases two, three, and four of scenario two.

[0169] Optionally, in some embodiments, multiple configuration information corresponding to multiple models can be determined based on the configuration of the training datasets of multiple models. For example, consider a terminal device training two beam prediction models based on two training datasets. The training dataset configuration for the first beam prediction model is: measurement set (Set B) with 8 resources, prediction set (Set A) with 64 resources, and association ID 1. The training dataset configuration for the second beam prediction model is: measurement set (Set B) with 16 resources, prediction set (Set A) with 128 resources, and association ID 2. In this example, the NW-side condition for the first beam prediction model is: measurement set with 8 resources, prediction set with 64 resources. That is, the size of the measurement set for the first beam prediction model is 8, and the size of the prediction set is 64. The NW-side condition for the second beam prediction model is: measurement set with 16 resources, prediction set with 128 resources. That is, the size of the measurement set for the second beam prediction model is 16, and the size of the prediction set is 128. If the network device is configured with a measurement set of 8 resources and a prediction set of 64 resources during the inference phase, the terminal device can use the first beam prediction model for beam prediction. If the network device is configured with a measurement set of 16 resources and a prediction set of 128 resources during the inference phase, the terminal device can use the second beam prediction model for beam prediction.

[0170] In some cases, the NW-side conditions of multiple models during the training phase may be determined based on the same configuration information. That is, multiple models may have identical NW-side conditions during training. In this situation, the terminal device cannot determine the first model for inference from among the multiple models based on third-party configuration information. In other words, the terminal device cannot select the first model for inference from among the multiple models based on their NW-side conditions. For example, during the training phase, the training data configuration for multiple models is: 8 resources for the measurement set and 64 resources for the prediction set. That is, the input size and output size of the multiple models are the same. Therefore, during the inference phase, the terminal device cannot select the model for inference from among the multiple models based on the model's input and output sizes. In this case, the terminal device can select one model from among the multiple models as the model for inference based on one or more of the following: the model's index, the model's performance, and network device indication information.

[0171] Therefore, in some cases, the first condition can also be associated with one or more of the following: the index of the first model, the performance of the first model, and the indication information of the network device.

[0172] In some implementations, the first condition can be associated with the index of the first model. The index of the first model can include or be replaced by: the number of the first model. The number of the first model can be a function-based number. For example, a function ID. Alternatively, the number of the first model can be a number based on the model implementation. For example, a model ID. Alternatively, the number of the first model can be a number based on the dataset. Here, the dataset can be understood as the dataset used to train the first model. For example, a dataset ID.

[0173] Optionally, in some embodiments, the first condition may include: the first model is the model with the highest or lowest index among multiple models. That is, the terminal device can select the model with the highest or lowest index from multiple models used to perform the first function as the first model, and the terminal device can use the first model to perform inference in the first cell.

[0174] The implementation method of associating the first condition with the index of the first model, as well as the corresponding implementation method, can be applied to cases two, three, and four in scenario two.

[0175] In some implementations, the first condition may be associated with indication information from the network device. In this implementation, the first condition may include: the first model is the model indicated by the indication information from a plurality of models. The network device may send indication information to the terminal device, which may indicate one or more models as the first model from a plurality of models used to perform the first function.

[0176] Optionally, in some embodiments, the network device may send the indication information to the terminal device in response to a request from the terminal device. For example, the terminal device may send a request to the network device via signaling to inquire which model among multiple models can be used as the first model in the inference phase. This signaling may, for example, be user assistance information (UAI) signaling.

[0177] The implementation method of associating the first condition with the indication information of the network device, and the corresponding implementation method, can be applied to cases two, three, and four in scenario two.

[0178] In some implementations, the first condition can be correlated with the performance of the first model.

[0179] Optionally, in some embodiments, the terminal device may randomly select one model from multiple models for inference and determine whether the selected model can be used as the first model for inference in the first cell based on the inference performance of the selected model. For example, the terminal device may randomly select one model from multiple models for inference and monitor the inference performance of the selected model according to a certain metric through model performance monitoring, thereby determining whether to use the selected model as the first model for inference in the first cell.

[0180] Optionally, in other embodiments, the terminal device may select some or all of the multiple models for inference, and determine the model with the best inference performance among these models as the first model used for inference in the first cell. For example, the terminal device may use all of the multiple models for inference and perform performance monitoring. Based on the performance monitoring indicators, the terminal device selects the model with the best performance as the first model used for inference in the first cell.

[0181] The implementation method of associating the performance of the first condition with the first model, as well as the corresponding implementation method, can be applied to cases two, three, and four in scenario two.

[0182] To facilitate understanding of the communication method provided in the embodiments of this application, the communication method provided in the embodiments of this application will be described in more detail below with reference to specific examples.

[0183] Example 1

[0184] The UE collects datasets from N cells and trains a beam prediction model based on these datasets. The datasets for the N cells are: {Cell ID#1, Associated ID#A, measurement results from measurement set (set B), measurement results from prediction set (set A)}, {Cell ID#2, Associated ID#B, measurement results from measurement set (set B), measurement results from prediction set (set A)}, ..., {Cell ID#N, Associated ID#X, measurement results from measurement set (set B), measurement results from prediction set (set A)}. The measurement results from the measurement set serve as the model input, and the measurement results from the prediction set serve as the labels for model training, which is a supervised learning method. Assume the cell identifier currently used for inference is Cell ID#2, and the associated identifier corresponding to the data from the cell currently used for inference is Associated ID#B. Since Cell ID#2 and Associated ID#B have appeared in the mixed dataset, the UE can determine that the beam prediction model trained by the UE is consistent between the training and inference phases. The UE can use this beam prediction model to perform beam prediction in the current cell.

[0185] Example 2

[0186] The network device configures two mapping relationships via RRC signaling to indicate the mapping relationships between the associated identifiers of multiple cells. The first mapping relationship is {Cell ID#1, Associated ID#A}, {Cell ID#2, Associated ID#B}, {Cell ID#5, Associated ID#Q}, and the second mapping relationship is {Cell ID#4, Associated ID#D}, {Cell ID#5, Associated ID#E}, {Cell ID#6, Associated ID#H}, {Cell ID#7, Associated ID#B}. During the training phase, the UE trains the model based on the data of the cell with cell ID#5. Specifically, the UE trains based on the data of the corresponding associated identifier Associated ID#Q to generate beam prediction model 1, and trains based on the data of the corresponding associated identifier Associated ID#E to generate beam prediction model 2. The UE may perform any of the following operations: perform beam prediction using beam prediction model 1 in the cell with cell ID#1, perform beam prediction using beam prediction model 1 in the cell with cell ID#2, perform beam prediction using beam prediction model 2 in the cell with cell ID#4, perform beam prediction using beam prediction model 2 in the cell with cell ID#6, and perform beam prediction using beam prediction model 2 in the cell with cell ID#7.

[0187] Example 3

[0188] During the training phase, the network device configures a cell list, namely {Cell ID#1, Cell ID#2, Cell ID#5}, and an associated identifier, namely Associated ID#3. During the training phase, the UE trains its beam prediction model based on data from cells with cell ID#1 or Cell ID#2, and with additional NW-side conditions corresponding to Associated ID#3. During the inference phase, the network device configures the associated identifier as Associated ID#3. If the UE performs beam prediction within a cell with cell ID#5, the UE can use the beam prediction model obtained during the training phase.

[0189] Example 4

[0190] The UE performs training and inference within the same cell. The network device configures association identifiers during both the training and inference phases. During the data collection phase, the network device configures the measurement resources (i.e., NW-side conditions) for the dataset and configures one or more association identifiers (representing additional NW-side conditions). The UE trains one or more beam prediction models based on the dataset resources and one or more association identifiers. During the inference phase, the network device similarly configures the measurement resources (i.e., NW-side conditions) and one or more association identifiers (representing additional NW-side conditions) for inference. If one of the one or more beam prediction models trained by the UE has the same association identifier in both the training and inference phases, the UE can use that beam prediction model for beam prediction within that cell.

[0191] Example 5

[0192] The UE performs training and inference within the same cell. The network device configures an association identifier during the training phase but not during the inference phase. If the UE trains a first beam prediction model with the following configuration: measurement set (set B) of 8 resources, prediction set (set A) of 64 resources, and association identifier 1, and trains a second beam prediction model with the following configuration: measurement set (set B) of 16 resources, prediction set (set A) of 128 resources, and association identifier 2, then during the inference phase, the UE can choose to use either the first or second beam prediction model based on the NW-side conditions, i.e., the configured sizes of the measurement and prediction sets (either a combination of 8 and 64, or a combination of 16 and 128). If the NW-side conditions during the inference phase have measurement and prediction set sizes of 8 and 64, the UE can choose to use the first beam prediction model. If the NW-side conditions during the inference phase have measurement and prediction set sizes of 16 and 128, the UE can choose to use the second beam prediction model.

[0193] If a UE trains multiple beam prediction models with the following configuration: a measurement set (set B) of 8 resources, a prediction set (set A) of 64 resources, and an association identifier of 1, then because the NW-side conditions corresponding to these multiple beam prediction models are the same (the input and output dimensions of the multiple beam prediction models are the same, all being a combination of a measurement set of 8 transmit filters and a prediction set of 64 transmit beams), the UE cannot select the corresponding model based on the configuration information. In this case, the UE can choose the model with the lowest or highest model index for beam prediction.

[0194] Example 6

[0195] The UE performs training and inference within the same cell. The network device does not configure an association identifier during the training phase but configures one during the inference phase. The UE trains one or more beam prediction models during the training phase. In this case, because the association identifier configured during the inference phase does not correspond to the association identifier configured during the training phase, the UE cannot recognize the association identifier, and therefore the association identifier configured during the inference phase is ineffective. This embodiment is equivalent to not configuring an association identifier during the inference phase.

[0196] The UE first determines the number of trained beam prediction models. If the UE has only trained one beam prediction model, it also needs to check whether the configured inference resources match the beam prediction model (i.e., the input and output of the beam prediction model). If they match, the UE can assume that the beam prediction model has consistency with the NW-side additional conditions during the training and inference phases, and can use the beam prediction model for beam prediction.

[0197] If the UE has trained multiple beam prediction models, it needs to select one that matches the configured inference resources, using the same selection method as in Example 5. If the UE can find a beam prediction model that matches the configured inference resources, it can assume that the model has consistency with the NW-side additional conditions during both the training and inference phases, and can use it for beam prediction. If the UE cannot find a beam prediction model that matches the configured inference resources, it can perform any of the following actions: Action 1: The UE selects a beam prediction model for beam prediction according to a default rule (e.g., based on the lowest or highest model index). Action 2: The UE queries the network device via signaling (e.g., UAI signaling) which beam prediction model is suitable for the association identifier during the inference phase. Action 3: The UE randomly selects a beam prediction model for beam prediction. The prediction performance of the selected beam prediction model is monitored using model performance monitoring according to a certain metric to determine whether the selected model has consistency. Action 4: The UE selects all models from multiple beam prediction models for beam prediction and performs performance monitoring. The UE determines the beam prediction model with the best performance based on the performance monitoring indicators and uses the best-performing beam prediction model for beam prediction.

[0198] Example 7

[0199] The UE performs training and inference within the same cell. The network device does not configure an association identifier during either the training or inference phases. In this embodiment, it is impossible to determine the consistency of the NW-side additional conditions between the training and inference phases based on the association identifier. This embodiment can employ the same approach as in Embodiment 6.

[0200] The embodiments of this application have been described in detail above with specific examples. It should be noted that the examples described above are merely to help those skilled in the art understand the embodiments of this application, and are not intended to limit the embodiments of this application to the specific numerical values ​​or scenarios illustrated. Those skilled in the art can obviously make various equivalent modifications or variations based on the given examples, and such modifications or variations also fall within the scope of the embodiments of this application.

[0201] The method embodiments of this application have been described in detail above with reference to Figure 6. The apparatus embodiments of this application will be described in detail below with reference to Figures 7 to 9. It should be understood that the descriptions of the method embodiments correspond to the descriptions of the apparatus embodiments; therefore, any parts not described in detail can be referred to the preceding method embodiments.

[0202] Figure 7 is a schematic diagram of the structure of a communication device 700 provided in an embodiment of this application. The communication device 700 shown in Figure 7 is a terminal device. The communication device 700 includes a determining unit 710. The determining unit 710 is used to determine, when a first condition is met, to use a first model for inference in a first cell; the first condition is associated with one or more of the following: training data of the first model; first configuration information for indicating network device-side conditions and / or a first association identifier for the inference phase of the model, wherein the first association identifier is used to indicate additional network device-side conditions for the inference phase; second configuration information for indicating network device-side conditions and / or a second association identifier for the training phase of the model, wherein the second association identifier is used to indicate additional network device-side conditions for the training phase.

[0203] In some implementations, the first model is trained based on data from multiple cells, and the first condition includes one or more of the following: the data from the multiple cells includes the data from the first cell; the association identifier corresponding to the data from the first cell is the same as the first association identifier.

[0204] In some implementations, the first model is trained based on data from the second cell, the first association identifier is the association identifier corresponding to the first cell, and the first condition includes a mapping relationship between the association identifier corresponding to the data of the second cell and the first association identifier.

[0205] In some implementations, the first condition includes the first association identifier corresponding to a first cell group, where the first cell group includes the first cell.

[0206] In some implementations, the first condition is associated with the first configuration information and / or the second configuration information, including: when the first configuration information does not indicate the first association identifier and / or the second configuration information does not indicate the second association identifier, the first condition is associated with the number of models trained by the terminal device for performing a first function; wherein the first function is a function performed by the first model.

[0207] In some implementations, the first condition includes that the number of models trained by the terminal device to perform the first function is 1.

[0208] In some implementations, the first condition is associated with the number of models trained by the terminal device to perform the first function, including: when the terminal device trains multiple models to perform the first function, the first condition is associated with one or more of the following: the index of the first model; the performance of the first model; and indication information of the network device.

[0209] In some implementations, the network device-side conditions of the multiple models during the training phase are determined based on multiple configuration information, including third configuration information corresponding to the first model, and the first condition includes the matching of the third configuration information with the first configuration information.

[0210] In some implementations, the first condition includes the first model being the model with the lowest or highest index among the plurality of models.

[0211] In some implementations, the first condition includes the first model being the model indicated by the indication information from the plurality of models.

[0212] Figure 8 is a schematic diagram of the structure of a communication device 800 provided in an embodiment of this application. The communication device 800 shown in Figure 8 is a network device. The communication device 800 includes a first sending unit 810. The first sending unit 810 is used to send configuration information of a first model association to a terminal device. The configuration information of the first model association includes one or more of the following: first configuration information, used to indicate network device-side conditions and / or a first association identifier during the inference phase of the model, wherein the first association identifier is used to indicate additional network device-side conditions during the inference phase; second configuration information, used to indicate network device-side conditions and / or a second association identifier during the training phase of the model, wherein the second association identifier is used to indicate additional network device-side conditions during the training phase.

[0213] In some implementations, whether the first model is allowed to be used in the first cell is determined based on a first condition, which is associated with one or more of the following: the training data of the first model; the first configuration information; and the second configuration information.

[0214] In some implementations, the first model is trained based on data from multiple cells, and the first condition includes one or more of the following: the data from the multiple cells includes the data from the first cell; the association identifier corresponding to the data from the first cell is the same as the first association identifier.

[0215] In some implementations, the first model is trained based on data from the second cell, the first association identifier is the association identifier corresponding to the first cell, and the first condition includes a mapping relationship between the association identifier corresponding to the data of the second cell and the first association identifier.

[0216] In some implementations, the first condition includes the first association identifier corresponding to a first cell group, where the first cell group includes the first cell.

[0217] In some implementations, the first condition is associated with the first configuration information and / or the second configuration information, including: when the first configuration information does not indicate the first association identifier and / or the second configuration information does not indicate the second association identifier, the first condition is associated with the number of models trained by the terminal device for performing a first function; wherein the first function is a function performed by the first model.

[0218] In some implementations, the first condition includes that the number of models trained by the terminal device to perform the first function is 1.

[0219] In some implementations, the first condition is associated with the number of models trained by the terminal device to perform the first function, including: when the terminal device trains multiple models to perform the first function, the first condition is associated with one or more of the following: the index of the first model; the performance of the first model; and indication information of the network device.

[0220] In some implementations, the network device-side conditions of the multiple models during the training phase are determined based on multiple configuration information, including third configuration information corresponding to the first model, and the first condition includes the matching of the third configuration information with the first configuration information.

[0221] In some implementations, the first condition includes the first model being the model with the lowest or highest index among the plurality of models.

[0222] In some implementations, the first condition includes the first model being the model indicated by the indication information from the plurality of models.

[0223] Figure 9 is a schematic diagram of the structure of a communication device applicable to embodiments of this application. The dashed lines in Figure 9 indicate that the unit or module is optional. This device 900 can be used to implement the methods described in the above method embodiments. Device 900 can be a chip, a terminal device, or a network device.

[0224] The apparatus 900 may include one or more processors 910. The processor 910 may support the apparatus 900 in implementing the methods described in the preceding method embodiments. The processor 910 may be a general-purpose processor or a special-purpose processor. For example, the processor may be a central processing unit (CPU). Alternatively, the processor may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0225] The apparatus 900 may further include one or more memories 920. The memories 920 store a program that can be executed by the processor 910, causing the processor 910 to perform the methods described in the preceding method embodiments. The memories 920 may be independent of the processor 910 or integrated within the processor 910.

[0226] The device 900 may also include a transceiver 930. The processor 910 can communicate with other devices or chips via the transceiver 930. For example, the processor 910 can send and receive data with other devices or chips via the transceiver 930.

[0227] This application also provides a computer-readable storage medium for storing a program. This computer-readable storage medium can be applied to the communication device provided in this application, and the program causes a computer to execute the methods performed by the communication device in various embodiments of this application.

[0228] This application also provides a computer program product. The computer program product includes a program. The computer program product can be applied to the communication device provided in this application embodiment, and the program causes a computer to execute the methods performed by the communication device in various embodiments of this application.

[0229] This application also provides a computer program. This computer program can be applied to the communication device provided in this application, and causes the computer to execute the methods performed by the communication device in various embodiments of this application.

[0230] It should be understood that the terms "system" and "network" in this application can be used interchangeably. Furthermore, the terminology used in this application is only for explaining specific embodiments of the application and is not intended to limit the application. The terms "first," "second," "third," and "fourth," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. In addition, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0231] In the embodiments of this application, the term "instruction" can be a direct instruction, an indirect instruction, or an indication of a relationship. For example, A instructing B can mean that A directly instructs B, such as B being able to obtain information through A; it can also mean that A indirectly instructs B, such as A instructing C, so B can obtain information through C; or it can mean that there is a relationship between A and B.

[0232] In the embodiments of this application, "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean that B is determined solely based on A; B can also be determined based on A and / or other information.

[0233] In the embodiments of this application, the term "correspondence" can indicate a direct or indirect correspondence between two things, or an association between two things, or a relationship such as instruction and being instructed, configuration and being configured.

[0234] In this application embodiment, "predefined" or "preconfigured" can be implemented by pre-storing corresponding codes, tables, or other means that can be used to indicate relevant information in the device (e.g., including terminal devices and network devices). This application does not limit the specific implementation method. For example, predefined can refer to what is defined in the protocol.

[0235] In this application embodiment, the "protocol" may refer to a standard protocol in the field of communication, such as the LTE protocol, the NR protocol, and related protocols applied to future communication systems. This application does not limit this.

[0236] In the embodiments of this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0237] In the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0238] 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.

[0239] 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.

[0240] 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.

[0241] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of 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 the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, 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, fiber optic, 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 read 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 (e.g., solid-state disks, SSDs), etc.

[0242] 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 communication method, characterized in that, include: If the first condition is met, the terminal device determines to use the first model for inference in the first cell; The first condition is associated with one or more of the following: The training data of the first model; First configuration information is used to indicate network device-side conditions and / or a first association identifier for the inference phase of the model, wherein the first association identifier is used to indicate additional network device-side conditions for the inference phase. The second configuration information is used to indicate the network device-side conditions and / or the second association identifier during the training phase of the model, wherein the second association identifier is used to indicate the additional network device-side conditions during the training phase.

2. The method according to claim 1, characterized in that, When the first condition is associated with the training data of the first model and the first configuration information, the first model is trained based on data from multiple cells, and the first condition includes one or more of the following: The data from the multiple cells includes the data from the first cell; The association identifier corresponding to the data of the first cell is the same as the first association identifier.

3. The method according to claim 1, characterized in that, When the first condition is associated with the training data of the first model and the first configuration information, the first model is trained based on the data of the second cell, the first association identifier is the association identifier corresponding to the first cell, and the first condition includes a mapping relationship between the association identifier corresponding to the data of the second cell and the first association identifier.

4. The method according to claim 1, characterized in that, When the first condition is associated with the first configuration information, the first condition includes the first association identifier corresponding to the first cell group, and the first cell group includes the first cell.

5. The method according to claim 1, characterized in that, The first condition is associated with the first configuration information and / or the second configuration information, including: If the first configuration information does not indicate the first associated identifier and / or the second configuration information does not indicate the second associated identifier, the first condition is associated with the number of models trained by the terminal device for performing the first function; wherein, the first function is the function performed by the first model.

6. The method according to claim 5, characterized in that, The first condition includes that the number of models trained by the terminal device to perform the first function is 1.

7. The method according to claim 5, characterized in that, The first condition is associated with the number of models trained by the terminal device to perform the first function, including: When the model trained on the terminal device to perform the first function comprises multiple models, the first condition is associated with one or more of the following: The index of the first model; The performance of the first model; Instructions for network devices.

8. The method according to claim 7, characterized in that, The network device-side conditions for the multiple models during the training phase are determined based on multiple configuration information, including third configuration information corresponding to the first model, and the first condition includes the matching of the third configuration information with the first configuration information.

9. The method according to claim 7, characterized in that, The first condition includes the first model being the model with the lowest or highest index among the plurality of models.

10. The method according to claim 7, characterized in that, The first condition includes the first model being the model indicated by the indication information from the plurality of models.

11. The method according to any one of claims 1 to 10, characterized in that: The training data for the first model is the data collected by the terminal device; or, The training data for the first model is the data received by the terminal device from the network device.

12. The method according to any one of claims 1 to 11, characterized in that, The first configuration information and / or the second configuration information are configuration information sent by the network device to the terminal device.

13. A communication method, characterized in that, include: The network device sends configuration information associated with a first model to the terminal device, wherein the configuration information associated with the first model includes one or more of the following: First configuration information is used to indicate network device-side conditions and / or a first association identifier for the inference phase of the model, wherein the first association identifier is used to indicate additional network device-side conditions for the inference phase. The second configuration information is used to indicate the network device-side conditions and / or the second association identifier during the training phase of the model, wherein the second association identifier is used to indicate the additional network device-side conditions during the training phase.

14. The method according to claim 13, characterized in that, Whether the first model is allowed to be used in the first cell is determined based on a first condition, which is associated with one or more of the following: The training data of the first model; The first configuration information; The second configuration information.

15. The method according to claim 14, characterized in that, When the first condition is associated with the training data of the first model and the first configuration information, the first model is trained based on data from multiple cells, and the first condition includes one or more of the following: The data from the multiple cells includes the data from the first cell; The association identifier corresponding to the data of the first cell is the same as the first association identifier.

16. The method according to claim 14, characterized in that, When the first condition is associated with the training data of the first model and the first configuration information, the first model is trained based on the data of the second cell, the first association identifier is the association identifier corresponding to the first cell, and the first condition includes a mapping relationship between the association identifier corresponding to the data of the second cell and the first association identifier.

17. The method according to claim 14, characterized in that, When the first condition is associated with the first configuration information, the first condition includes the first association identifier corresponding to the first cell group, and the first cell group includes the first cell.

18. The method according to claim 14, characterized in that, The first condition is associated with the first configuration information and / or the second configuration information, including: If the first configuration information does not indicate the first associated identifier and / or the second configuration information does not indicate the second associated identifier, the first condition is associated with the number of models trained by the terminal device for performing the first function; wherein, the first function is the function performed by the first model.

19. The method according to claim 18, characterized in that, The first condition includes that the number of models trained by the terminal device to perform the first function is 1.

20. The method according to claim 18, characterized in that, The first condition is associated with the number of models trained by the terminal device to perform the first function, including: When the model trained on the terminal device to perform the first function comprises multiple models, the first condition is associated with one or more of the following: The index of the first model; The performance of the first model; Instructions for network devices.

21. The method according to claim 20, characterized in that, The network device-side conditions for the multiple models during the training phase are determined based on multiple configuration information, including third configuration information corresponding to the first model, and the first condition includes the matching of the third configuration information with the first configuration information.

22. The method according to claim 20, characterized in that, The first condition includes the first model being the model with the lowest or highest index among the plurality of models.

23. The method according to claim 20, characterized in that, The first condition includes the first model being the model indicated by the indication information from the plurality of models.

24. The method according to any one of claims 14 to 23, characterized in that, The method further includes: The network device sends the training data of the first model to the terminal device.

25. A communication device, characterized in that, The communication device is a terminal device, and the device includes: The determining unit is used to determine, when the first condition is met, to use the first model for inference in the first cell; The first condition is associated with one or more of the following: The training data of the first model; First configuration information is used to indicate network device-side conditions and / or a first association identifier for the inference phase of the model, wherein the first association identifier is used to indicate additional network device-side conditions for the inference phase. The second configuration information is used to indicate the network device-side conditions and / or the second association identifier during the training phase of the model, wherein the second association identifier is used to indicate the additional network device-side conditions during the training phase.

26. The device according to claim 25, characterized in that, When the first condition is associated with the training data of the first model and the first configuration information, the first model is trained based on data from multiple cells, and the first condition includes one or more of the following: The data from the multiple cells includes the data from the first cell; The association identifier corresponding to the data of the first cell is the same as the first association identifier.

27. The device according to claim 25, characterized in that, When the first condition is associated with the training data of the first model and the first configuration information, the first model is trained based on the data of the second cell, the first association identifier is the association identifier corresponding to the first cell, and the first condition includes a mapping relationship between the association identifier corresponding to the data of the second cell and the first association identifier.

28. The device according to claim 25, characterized in that, When the first condition is associated with the first configuration information, the first condition includes the first association identifier corresponding to the first cell group, and the first cell group includes the first cell.

29. The device according to claim 25, characterized in that, The first condition is associated with the first configuration information and / or the second configuration information, including: If the first configuration information does not indicate the first associated identifier and / or the second configuration information does not indicate the second associated identifier, the first condition is associated with the number of models trained by the terminal device for performing the first function; wherein, the first function is the function performed by the first model.

30. The device according to claim 29, characterized in that, The first condition includes that the number of models trained by the terminal device to perform the first function is 1.

31. The device according to claim 29, characterized in that, The first condition is associated with the number of models trained by the terminal device to perform the first function, including: When the model trained on the terminal device to perform the first function comprises multiple models, the first condition is associated with one or more of the following: The index of the first model; The performance of the first model; Instructions for network devices.

32. The device according to claim 31, characterized in that, The network device-side conditions for the multiple models during the training phase are determined based on multiple configuration information, including third configuration information corresponding to the first model, and the first condition includes the matching of the third configuration information with the first configuration information.

33. The device according to claim 31, characterized in that, The first condition includes the first model being the model with the lowest or highest index among the plurality of models.

34. The device according to claim 31, characterized in that, The first condition includes the first model being the model indicated by the indication information from the plurality of models.

35. The method according to any one of claims 25 to 34, characterized in that: The training data for the first model is the data collected by the terminal device; or, The training data for the first model is the data received by the terminal device from the network device.

36. The method according to any one of claims 25 to 35, characterized in that, The first configuration information and / or the second configuration information are configuration information sent by the network device to the terminal device.

37. A communication device, characterized in that, The communication device is a network device, and the device includes: The first sending unit is configured to send configuration information associated with a first model to the terminal device, wherein the configuration information associated with the first model includes one or more of the following: First configuration information is used to indicate network device-side conditions and / or a first association identifier for the inference phase of the model, wherein the first association identifier is used to indicate additional network device-side conditions for the inference phase. The second configuration information is used to indicate the network device-side conditions and / or the second association identifier during the training phase of the model, wherein the second association identifier is used to indicate the additional network device-side conditions during the training phase.

38. The device according to claim 37, characterized in that, Whether the first model is allowed to be used in the first cell is determined based on a first condition, which is associated with one or more of the following: The training data of the first model; The first configuration information; The second configuration information.

39. The device according to claim 38, characterized in that, When the first condition is associated with the training data of the first model and the first configuration information, the first model is trained based on data from multiple cells, and the first condition includes one or more of the following: The data from the multiple cells includes the data from the first cell; The association identifier corresponding to the data of the first cell is the same as the first association identifier.

40. The device according to claim 38, characterized in that, When the first condition is associated with the training data of the first model and the first configuration information, the first model is trained based on the data of the second cell, the first association identifier is the association identifier corresponding to the first cell, and the first condition includes a mapping relationship between the association identifier corresponding to the data of the second cell and the first association identifier.

41. The device according to claim 38, characterized in that, When the first condition is associated with the first configuration information, the first condition includes the first association identifier corresponding to the first cell group, and the first cell group includes the first cell.

42. The device according to claim 38, characterized in that, The first condition is associated with the first configuration information and / or the second configuration information, including: If the first configuration information does not indicate the first associated identifier and / or the second configuration information does not indicate the second associated identifier, the first condition is associated with the number of models trained by the terminal device for performing the first function; wherein, the first function is the function performed by the first model.

43. The device according to claim 42, characterized in that, The first condition includes that the number of models trained by the terminal device to perform the first function is 1.

44. The device according to claim 42, characterized in that, The first condition is associated with the number of models trained by the terminal device to perform the first function, including: When the model trained on the terminal device to perform the first function comprises multiple models, the first condition is associated with one or more of the following: The index of the first model; The performance of the first model; Instructions for network devices.

45. The device according to claim 44, characterized in that, The network device-side conditions for the multiple models during the training phase are determined based on multiple configuration information, including third configuration information corresponding to the first model, and the first condition includes the matching of the third configuration information with the first configuration information.

46. ​​The device according to claim 44, characterized in that, The first condition includes the first model being the model with the lowest or highest index among the plurality of models.

47. The device according to claim 44, characterized in that, The first condition includes the first model being the model indicated by the indication information from the plurality of models.

48. The device according to any one of claims 38 to 47, characterized in that, The device also includes: The second sending unit is used to send the training data of the first model to the terminal device.

49. A communication device, characterized in that, The device includes a transceiver, a memory, and a processor. The memory stores a program, and the processor invokes the program in the memory and controls the transceiver to receive or transmit signals so that the communication device performs the method as claimed in any one of claims 1-12 or any one of claims 13-24.

50. An apparatus, characterized in that, Includes a processor for calling a program from memory to cause the device to perform the method as claimed in any one of claims 1-12 or any one of claims 13-24.

51. A chip, characterized in that, Includes a processor for calling a program from memory, causing a device on which the chip is mounted to perform the method as claimed in any one of claims 1-12 or any one of claims 13-24.

52. A computer-readable storage medium, characterized in that, It contains a program that causes a computer to perform the method as claimed in any one of claims 1-12 or any one of claims 13-24.

53. A computer program product, characterized in that, Includes a program that causes a computer to perform the method as claimed in any one of claims 1-12 or any one of claims 13-24.

54. A computer program, characterized in that, The computer program causes the computer to perform the method as claimed in any one of claims 1-12 or any one of claims 13-24.

Images