Communication method and communication apparatus

Abstract

The present application provides a communication method and a communication apparatus. In the communication method, a first network device can send first information to a terminal, wherein the first information is used for indicating a first target node, the first target node is used for acquiring second information, the second information is used for obtaining a first AI model, the first AI model is used for the terminal, and the first AI model and a second AI model used for the first network device jointly achieve a first AI function; and the terminal obtains the first AI model on the basis of the first information. According to the method, the terminal can obtain, among dual-end AI models, a terminal-side AI model that jointly achieves the first AI function with a network-side AI model.

Description

Communication methods and communication devices

[0001] This application claims priority to Chinese Patent Application No. 202510498796.6, filed on April 17, 2025, entitled "Communication Method and Communication Apparatus", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of communications, and more particularly to a communication method and a communication device. Background Technology

[0003] With the development of mobile communication technology, the application of artificial intelligence (AI) / machine learning (ML) technologies in mobile communication networks has been proposed to improve communication performance. The 3rd generation partnership project (3GPP) is exploring several use cases for optimization using AI / ML technologies, such as beam management, positioning, or channel state information (CSI) feedback.

[0004] A dual-end AI model refers to AI inference that is jointly performed by an AI model on the terminal side and an AI model on the network side. In other words, a dual-end AI model means that there is an AI model on the terminal side and an AI model on the network side, and these two models are used together to complete AI inference.

[0005] However, how to obtain the AI ​​model on the terminal side to jointly complete AI inference with the AI ​​model on the network side has become a pressing technical problem that needs to be solved. Summary of the Invention

[0006] This application provides a communication method and a communication device that enables a terminal to obtain the terminal-side AI model, which together with the network-side AI model, completes the first AI function in a dual-end AI model.

[0007] Firstly, this application provides a communication method that can be applied to a terminal device, such as a terminal or a communication module within a terminal, or a circuit or chip (such as a modem chip, also known as a baseband chip, or a system-on-chip (SoC) chip containing a modem core, or a system-in-package (SIP) chip) responsible for communication functions within the terminal. In this application, a terminal is used as an example for description.

[0008] The communication method includes: receiving first information from a first network device, the first information being used to instruct a first target node, the first target node being used to obtain second information, wherein the second information is used to obtain a first AI model, the first AI model is used by a terminal, and the first AI model and a second AI model used by the first network device jointly complete a first AI function; and obtaining the first AI model based on the first information.

[0009] The aforementioned first AI model and the second AI model used in the first network device jointly complete the first AI function, which can also be understood as: both the first AI model and the second AI model are associated with / correspond to / are used for the first AI function. That is, the aforementioned first AI model can also be considered as the terminal-side AI model in the dual-end AI model used to complete the first AI function, and the second AI model can be considered as the network-side AI model in the dual-end AI model used to complete the first AI function.

[0010] In the above method, the first target node is the node used to obtain the second information. The second information is used to train the first AI model; therefore, the second information can also be called training information or information related to the first AI model. Specifically, the second information is the information needed during the training of the first AI model. For example, the second information includes at least one of the following for obtaining the first AI model: dataset, model structure, and / or model parameters. Based on the communication method provided in the first aspect, the first network device indicates the first target node for obtaining the second information of the first AI model to the terminal. Thus, the terminal can obtain the second information from the first target node to obtain the first AI model, thereby obtaining the first AI model; or, the terminal can indicate the first target node for obtaining the second information to the terminal-side server, thereby enabling the terminal-side server to obtain the second information from the first target node, train the first AI model based on the second information, and indicate the first AI model to the terminal, thereby enabling the terminal to obtain the first AI model that, together with the second AI model used by the first network device, completes / implements the first AI function. Furthermore, based on this technical solution, the first network device does not need to send the second information to the terminal, thus saving air interface resources.

[0011] In conjunction with the first aspect, in one possible implementation, the method further includes: receiving at least one of the following information from the first network device: information indicating that the first network device has a first AI function, information indicating a first mode, information indicating the model structure of a second AI model, or an identifier of the second information.

[0012] Information used to indicate that the first network device has the first AI function can also be understood as information used to indicate that the first network device supports the first AI function.

[0013] The information used to indicate the first method can also be described as: information indicating the first method associated with / corresponding to the first AI mode that matches the second AI model. The first method can be based on a first dataset to obtain the first AI model, or it can be based on a first model structure and / or first model parameters to obtain the first AI model. That is, it can be understood as the first method that the first network device should / needs to use when instructing the terminal to obtain a first AI model that matches the second AI model. Alternatively, it can be understood as the first method supported by the first network device when instructing the terminal to obtain the first AI model.

[0014] Information used to indicate the model structure of the second AI model can also be understood as information used to indicate to the terminal the model structure of the second AI model, which is a first network device and works together with the first AI model to perform the first AI function.

[0015] The identifier of the second information can also be described as the version number of the second information. Different version numbers of the second information are different; therefore, one version number corresponds to one version of the second information. There may be scenarios where the first target node updates the second information. In this scenario, the first network device sends the identifier of the second information to the terminal to indicate to the terminal the version of the second information associated with the first AI model that matches the second AI model.

[0016] Based on this implementation, the terminal can obtain at least one of the identifiers of the first method and the second information corresponding to the first AI model that matches the second AI model. This helps to improve the matching degree between the first AI model and the second AI model obtained by the terminal, thereby improving the inference performance of the dual-end AI model including the first AI model and the second AI model.

[0017] In one implementation, the method further includes sending third information to a first network device, the third information being used to request a first AI model. Correspondingly, the first network device can send first information to the terminal after receiving the third information.

[0018] The third information is used to request the first AI model, which can also be understood as: the third information is used to request the second information needed to train the first AI model.

[0019] For example, in one implementation, if the terminal supports a first AI function and / or the terminal supports obtaining a first AI model based on a first method, the terminal sends third information to the first network device to request the second information required to obtain the first AI model, thereby obtaining the first AI model and further enabling it to work with the second AI model to complete the first AI function.

[0020] In conjunction with the first aspect, in one possible implementation, the above method further includes: receiving fourth information, the fourth information being used to indicate a first set of network devices or a first set of cells associated with the second information, the first set of network devices including at least one network device, the at least one network device including a first network device, the first set of cells including at least one cell, and the at least one cell including a first cell currently accessed by the terminal.

[0021] In other words, the fourth information can be considered as indicating to the terminal the geographical range to which the first AI model determined based on the second information can be applied. Thus, when the terminal moves within the coverage area indicated by the fourth information, the terminal only needs to obtain the address of the first target node once, without having to request the address of the first target node from each network device in the first network device set, which can save the terminal's signaling overhead.

[0022] In conjunction with the first aspect, in one possible implementation, the above method further includes: sending fifth information to the first network device, the fifth information being used to indicate that the terminal has a third AI model, the third AI model and the fourth AI model used by the first network device jointly performing the second AI function.

[0023] For example, the fifth information includes at least one of the following: the identifier of the sixth information, information for indicating the second method, or information for indicating the second target node; the sixth information includes at least one of the following for obtaining the third AI model: the second dataset, the second model structure, and / or the second model parameters; the second method is any one of the following: the third AI model is obtained based on the second dataset, or the third AI model is obtained based on the second model structure and / or the second model parameters; the second target node is used to obtain the sixth information.

[0024] In one implementation, after the terminal reports its third AI model to the first network device, the first network device can determine whether the third AI model matches the fourth AI model. If they do not match, the first network device can also send a seventh message to the terminal. The seventh message received from the first network device indicates whether the third AI model and the fourth AI model are mismatched or the reason for the mismatch. The reason for the mismatch includes one or more of the following: mismatch in the identifier of the sixth message, or mismatch in the second method. Based on this implementation, after receiving the seventh message, the terminal can re-obtain the third AI model that matches the fourth AI model, i.e., update the third AI model, thereby improving the inference performance of the dual-end AI model including the first AI model and the second AI model. Understandably, re-obtaining the third AI model here can be achieved by the terminal or the terminal-side server retraining it, or by the terminal or the terminal-side server re-obtaining the information used to obtain the third AI model from the first target node, and then re-obtaining the third AI model based on the re-obtained information associated with the third AI model.

[0025] Secondly, this application provides a communication method that can be applied to a first network device, such as a first network equipment or a communication module within the first network device, or a circuit or chip (such as a modem chip, also known as a baseband chip, or a SoC chip or system-in-package (SIP) chip containing a modem core) responsible for communication functions within the first network device. In this application, a first network device is used as an example for description.

[0026] The communication method includes: sending first information to a terminal, the first information being used to instruct a first target node, the first target node being used to obtain second information, wherein the second information is used to obtain a first AI model, the first AI model being used by the terminal, and the first AI model and a second AI model used by the first network device jointly performing a first AI function.

[0027] Based on the communication method provided in the second aspect, the first network device indicates a first target node for obtaining second information of the first AI model to the terminal. In this way, the terminal can indicate the first target node for obtaining the second information of the first AI model to the terminal-side server, thereby causing the terminal-side server to request the second information from the first target node, train the first AI model based on the second information, and then indicate the first AI model to the terminal. This allows the terminal to obtain a first AI model that, together with the second AI model used by the first network device, completes / implements the first AI function. Furthermore, based on this technical solution, the first network device does not need to send the second information to the terminal, thus saving air interface resources.

[0028] In conjunction with the second aspect, in one possible implementation, the second information includes at least one of the following for obtaining the first AI model: a first dataset, a first model structure, and / or first model parameters.

[0029] In conjunction with the second aspect, in one possible implementation, the method further includes: sending at least one of the following information to the terminal: information indicating that the first network device has a first AI function, information indicating a first mode, information indicating the model structure of the second AI model, or an identifier of the second information; the first mode is any one of the following: obtaining the first AI model based on the first dataset, or obtaining the first AI model based on the first model structure and / or the first model parameters.

[0030] In conjunction with the second aspect, in one possible implementation, the method further includes: receiving third information from the terminal, the third information being used to request the first AI model.

[0031] In conjunction with the second aspect, in one possible implementation, the method further includes: sending fourth information to the terminal, the fourth information being used to indicate a first set of network devices or a first set of cells associated with the second information, the first set of network devices including at least one network device, the at least one network device including a first network device, the first set of cells including at least one cell, and the at least one cell including a first cell currently accessed by the terminal.

[0032] In conjunction with the second aspect, in one possible implementation, the method further includes: receiving fifth information from the terminal, the fifth information indicating that the terminal has a third AI model, the third AI model and a fourth AI model for the first network device jointly performing the second AI function.

[0033] In conjunction with the second aspect, in one possible implementation, the fifth information includes at least one of the following: an identifier for the sixth information, information indicating the second method, or information indicating the second target node; the sixth information includes at least one of the following for obtaining the third AI model: a second dataset, a second model structure, and / or second model parameters; the second method is any one of the following: the third AI model is obtained based on the second dataset, or the third AI model is obtained based on the second model structure and / or the second model parameters; the second target node is used to obtain the sixth information.

[0034] In conjunction with the second aspect, in one possible implementation, the above method further includes: sending a seventh message to the terminal, the seventh message being used to indicate the mismatch between the third AI model and the fourth AI model or the reason for the mismatch; the reason for the mismatch includes one or more of the following: the identifier of the sixth message is mismatched, or the second method is mismatched.

[0035] The beneficial effects in the second aspect can be referenced from those in the first aspect, and will not be elaborated further.

[0036] Thirdly, this application provides a communication method that can be applied to a terminal device, such as a terminal or a communication module within a terminal, or a circuit or chip within a terminal responsible for communication functions (such as a modem chip, also known as a baseband chip, or a SoC chip containing a modem core, or a system-in-package (SIP) chip). In this application, a terminal is used as an example for description.

[0037] The communication method includes: sending fifth information to a first network device, the fifth information indicating that the terminal has a third AI model, the third AI model and a fourth AI model used by the first network device jointly performing a second AI function; receiving seventh information from the first network device, the seventh information indicating that the third AI model and the fourth AI model do not match or the reason for the mismatch; the reason for the mismatch includes one or more of the following: the identifier of the sixth information does not match, the second method does not match.

[0038] Based on the method provided in the third aspect, the terminal reports its third AI model to the first network device. Correspondingly, the first network device determines whether the third AI model matches the fourth AI model. The third and fourth AI models then jointly complete the second AI function. If a mismatch is determined, the device will indicate to the terminal the mismatch or the reason for the mismatch. Based on this implementation, after receiving the seventh information, the terminal can request the terminal-side server to retrain a third AI model that matches the fourth AI model, thereby improving the inference performance of the dual-end AI model, which includes both the first and second AI models.

[0039] In one possible implementation, the fifth information includes at least one of the following: an identifier for the sixth information, information indicating the second method, or information indicating the second target node; the sixth information includes at least one of the following for obtaining the third AI model: a second dataset, a second model structure, and / or second model parameters; the second method is any one of the following: the third AI model is obtained based on the second dataset, or the third AI model is obtained based on the second model structure and / or the second model parameters; the second target node is used to obtain the sixth information.

[0040] Fourthly, this application provides a communication method that can be applied to a first network device, such as a first network equipment or a communication module within the first network device, or a circuit or chip (such as a modem chip, also known as a baseband chip, or a SoC chip or system-in-package (SIP) chip containing a modem core) responsible for communication functions within the first network device. In this application, a first network device is used as an example for description.

[0041] The communication method includes: receiving fifth information from a terminal, the fifth information indicating that the terminal has a third AI model, the third AI model and a fourth AI model for a first network device jointly performing a second AI function; sending seventh information to the terminal, the seventh information indicating that the third AI model and the fourth AI model do not match or the reason for the mismatch; the reason for the mismatch includes one or more of the following: the identifier of the sixth information does not match, the second method does not match.

[0042] The beneficial effects of the fourth aspect can be referred to the description of the beneficial effects in the third aspect, and will not be repeated here.

[0043] Fifthly, this application provides a communication method that can be applied to a terminal device, such as a terminal or a communication module within a terminal, or a circuit or chip (such as a modem chip, also known as a baseband chip, or a system-on-chip (SoC) chip containing a modem core, or a system-in-package (SIP) chip) responsible for communication functions within the terminal. In this application, a terminal is used as an example for description.

[0044] The communication method includes: sending a first request to a first network device, the first request being used to request model parameters of a first AI model whose model structure is a target model structure, the first AI model being used by the terminal, the first AI model and a second AI model used by the first network device jointly performing a first AI function; and receiving model parameters from the first AI model of the first network device.

[0045] The first request is used to request the model parameters of the first AI model whose model structure is the target model structure. It can also be understood as: the first request is used to request the model parameters of the first AI model when the first AI model adopts the target AI model structure.

[0046] Based on this communication method, when the terminal requests the model parameters of the first AI model from the first network device, it also indicates the target model structure of the first AI model to be used. This allows the first network device to instruct the network-side server to train the second AI model and the first AI model to complete the first AI function. The model structure of the trained first AI model is the target model structure, thereby increasing the probability that the first AI model trained by the network side can be applied to the terminal.

[0047] In conjunction with the fifth aspect, in one possible implementation, the first request is further used to indicate a first parameter quantity threshold; wherein the number of parameters of the model parameters of the first AI model is less than or equal to the first parameter quantity threshold. In this implementation, after receiving the first parameter quantity threshold, the first network device, when sending the model parameters of the first AI model to the terminal, sends model parameters of the first AI model whose parameter quantity is less than or equal to the first parameter quantity threshold, thereby ensuring that the total size of the model parameters of the first AI model sent by the first network device does not exceed the first parameter quantity threshold of the model parameters of the first AI model requested by the terminal in this instance.

[0048] In conjunction with the fifth aspect, in one possible implementation, the above method further includes: sending first capability information of the terminal, the first capability information being used to indicate at least one of the following: the terminal supports a first AI function, whether the terminal supports a first network device providing model parameters of the first AI model to the terminal, at least one model structure supported by the terminal for the first AI model, and a maximum parameter threshold for the model parameters of the first AI model supported by the terminal.

[0049] In conjunction with the fifth aspect, in one possible implementation, at least one of the following information is received from the first network device: information indicating that the first network device has a first AI function, information indicating the model structure associated with the model parameters of the first AI model that the first network device can provide, an identifier indicating the model parameters of the first AI model provided by the first network device, and a set of network devices or a set of cells associated with the model parameters of the first AI model.

[0050] Information used to indicate the AI ​​model structure associated with the model parameters of the first AI model that the first network device can provide can also be understood as the first network device indicating to the terminal which AI model structure's model parameters it can provide.

[0051] In conjunction with the fifth aspect, in one possible implementation, the above method further includes: sending a first instruction message to a first network device, the first instruction message being used to instruct a third AI model possessed by the terminal, the third AI model and a fourth AI model used by the first network device jointly performing the second AI function.

[0052] For example, the first indication information includes at least one of the following: an identifier for indicating the model structure and / or model parameters of the third AI model possessed by the terminal.

[0053] In one implementation where the terminal sends a first instruction message to the first network device, the method may further include: receiving a second instruction message sent by the first network device, wherein the second instruction message is used to indicate the mismatch between the third AI model and the fourth AI model or the reason for the mismatch; the reason for the mismatch includes one or more of the following: the identifiers of the model parameters of the third AI model are mismatched, or the model structure of the third AI model is mismatched.

[0054] Sixthly, this application provides a communication method that can be applied to a first network device, such as a first network equipment or a communication module within the first network device, or a circuit or chip (such as a modem chip, also known as a baseband chip, or a SoC chip or system-in-package (SIP) chip containing a modem core) responsible for communication functions within the first network device. In this application, a first network device is used as an example for description.

[0055] The communication method includes: receiving a first request from a terminal, the first request being used to request model parameters of a first AI model whose model structure is a target model structure, the first AI model being used by the terminal, the first AI model and a second AI model used by a first network device jointly performing a first AI function; and sending the model parameters of the first AI model whose model structure is a target model structure.

[0056] Based on this communication method, when the terminal requests the model parameters of the first AI model from the first network device, it also indicates the target model structure of the first AI model to be used. This allows the first network device to instruct the network-side server to train the second AI model and the first AI model to complete the first AI function. The model structure of the trained first AI model is the target model structure, thereby increasing the probability that the first AI model trained by the network side can be applied to the terminal.

[0057] In conjunction with the sixth aspect, in one possible implementation, the first request is further used to indicate a first parameter quantity threshold; wherein the parameter quantity of the model parameters of the first AI model is less than or equal to the first parameter quantity threshold.

[0058] In conjunction with the sixth aspect, in one possible implementation, the above method further includes: receiving first capability information from the terminal, the first capability information indicating at least one of the following: the terminal supports a first AI function, whether the terminal supports a first network device providing model parameters of the first AI model to the terminal, at least one model structure supported by the terminal for the first AI model, and a maximum parameter threshold for the model parameters of the first AI model supported by the terminal.

[0059] In conjunction with the sixth aspect, in one possible implementation, at least one of the following information is sent: information indicating that the first network device has a first AI function; information indicating the AI ​​model structure associated with the model parameters of the first AI model that the first network device can provide; an identifier indicating the model parameters of the first AI model provided by the first network device; and a set of network devices or a set of cells associated with the model parameters of the first AI model.

[0060] In conjunction with the sixth aspect, in one possible implementation, the above method further includes: receiving first indication information from a terminal, the first indication information indicating a third AI model possessed by the terminal, the third AI model and a fourth AI model used by the first network device jointly performing a second AI function. For example, the first indication information includes at least one of the following: an identifier indicating the model structure and / or model parameters of the third AI model possessed by the terminal.

[0061] In one implementation, the method may further include: sending a second indication message, which indicates whether the third AI model and the fourth AI model are mismatched or the reason for the mismatch; the reason for the mismatch includes one or more of the following: the identifiers of the model parameters of the third AI model are mismatched, or the model structure of the third AI model is mismatched. Thus, if the terminal receives the second indication message, it can re-determine the third AI model that matches the fourth AI model, thereby improving the inference performance when performing the second AI function based on the third AI model and the fourth AI model.

[0062] Seventhly, this application provides a communication method that can be applied to a terminal device, such as a terminal or a communication module within a terminal, or a circuit or chip (such as a modem chip, also known as a baseband chip, or a system-on-chip (SoC) chip containing a modem core, or a system-in-package (SIP) chip) responsible for communication functions within the terminal. In this application, a terminal is used as an example for description.

[0063] The communication method includes: sending a first request to a first network device, the first request being used to request model parameters of a first AI model whose parameter quantity is less than a first parameter quantity threshold, the first AI model being used by the terminal, the first AI model and a second AI model used by the first network device jointly performing a first AI function; and receiving model parameters from the first AI model of the first network device.

[0064] Based on this communication method, when the terminal requests model parameters of the first AI model from the first network device, it also indicates to the first network device that the number of requested parameters is less than a first parameter threshold. This allows the first network device, upon receiving the first request, to instruct the network-side server to train a second AI model for completing the first AI function. When training the first AI model, the number of model parameters of the trained first AI model is less than or equal to the first parameter threshold, thereby increasing the probability that the first AI model trained by the network side can be applied to the terminal.

[0065] In conjunction with the seventh aspect, in one possible implementation, the first request is further used to indicate the target model structure of the first AI model. In this implementation, after receiving the first request, the first network device can instruct the network-side server to train a second AI model to perform the first AI function. When the first AI model is trained, the model structure of the trained first AI model is the target model structure, thereby increasing the probability that the first AI model trained on the network side can be applied to the terminal.

[0066] In conjunction with the seventh aspect, in one possible implementation, the above method further includes: sending first capability information of the terminal, the first capability information being used to indicate at least one of the following: the terminal supports a first AI function, whether the terminal supports a first network device providing model parameters of the first AI model to the terminal, at least one model structure supported by the terminal for the first AI model, and a maximum parameter threshold for the model parameters of the first AI model supported by the terminal.

[0067] In conjunction with the seventh aspect, in one possible implementation, at least one of the following information is received from the first network device: information indicating that the first network device has a first AI function, information indicating the AI ​​model structure associated with the model parameters of the first AI model that the first network device can provide, an identifier indicating the model parameters of the first AI model provided by the first network device, and a set of network devices or a set of cells associated with the model parameters of the first AI model.

[0068] In conjunction with the seventh aspect, in one possible implementation, the above method further includes: sending a first instruction message to a first network device, the first instruction message being used to instruct a third AI model possessed by the terminal, the third AI model and a fourth AI model used by the first network device jointly performing the second AI function.

[0069] For example, the first indication information includes at least one of the following: an identifier for indicating the model structure and / or model parameters of the third AI model possessed by the terminal.

[0070] In one implementation where the terminal sends a first instruction message to the first network device, the method may further include: receiving a second instruction message sent by the first network device, wherein the second instruction message is used to indicate the mismatch between the third AI model and the fourth AI model or the reason for the mismatch; the reason for the mismatch includes one or more of the following: the identifiers of the model parameters of the third AI model are mismatched, or the model structure of the third AI model is mismatched.

[0071] Eighthly, this application provides a communication method that can be applied to a terminal device, such as a terminal or a communication module within a terminal, or a circuit or chip (such as a modem chip, also known as a baseband chip, or a system-on-chip (SoC) chip containing a modem core, or a system-in-package (SIP) chip) responsible for communication functions within the terminal. In this application, a terminal is used as an example for description.

[0072] The communication method includes: receiving a first request from a terminal, the first request being for requesting model parameters of a first AI model whose parameter quantity is less than a first parameter quantity threshold, the first AI model being used by the terminal, the first AI model and a second AI model used by the first network device jointly performing a first AI function; and sending the model parameters of the first AI model.

[0073] Based on this communication method, when the terminal requests model parameters of the first AI model from the first network device, it also indicates to the first network device that the number of requested parameters is less than a first parameter threshold. This allows the first network device, upon receiving the first request, to instruct the network-side server to train a second AI model for completing the first AI function. When training the first AI model, the number of model parameters of the trained first AI model is less than or equal to the first parameter threshold, thereby increasing the probability that the first AI model trained by the network side can be applied to the terminal.

[0074] In conjunction with the eighth aspect, in one possible implementation, the first request is also used to indicate the target model structure of the first AI model.

[0075] In conjunction with the eighth aspect, in one possible implementation, the above method further includes: receiving first capability information from the terminal, the first capability information indicating at least one of the following: the terminal supports a first AI function, whether the terminal supports a first network device providing model parameters of the first AI model to the terminal, at least one model structure supported by the terminal for the first AI model, and a maximum parameter threshold for the model parameters of the first AI model supported by the terminal.

[0076] In conjunction with the eighth aspect, in one possible implementation, at least one of the following information is sent: information indicating that the first network device has a first AI function; information indicating the AI ​​model structure associated with the model parameters of the first AI model that the first network device can provide; an identifier indicating the model parameters of the first AI model provided by the first network device; and a set of network devices or a set of cells associated with the model parameters of the first AI model.

[0077] In conjunction with the eighth aspect, in one possible implementation, the above method further includes: receiving first instruction information from the terminal, the first instruction information being used to indicate a third AI model possessed by the terminal, the third AI model and a fourth AI model for the first network device jointly performing the second AI function.

[0078] For example, the first indication information includes at least one of the following: an identifier for indicating the model structure and / or model parameters of the third AI model possessed by the terminal.

[0079] In one implementation where the terminal sends a first instruction message to the first network device, the method may further include: sending a second instruction message, the second instruction message being used to indicate the mismatch between the third AI model and the fourth AI model or the reason for the mismatch; the reason for the mismatch includes one or more of the following: the identifiers of the model parameters of the third AI model are mismatched, the model structure of the third AI model is mismatched.

[0080] Ninthly, this application provides a communication device including modules or units for implementing the methods in any aspect and any possible implementation of any aspect. It should be understood that each module or unit can implement its corresponding function by executing a computer program.

[0081] In a tenth aspect, this application provides a communication device including a processor for performing the methods described in any of the foregoing aspects and any possible implementations of any of the foregoing aspects.

[0082] In one possible implementation, the processor is configured to, by executing a computer program or instructions, and / or, by logic circuitry, cause the communication device to perform the method described in any of the above aspects and any possible implementations of any of the aspects.

[0083] The apparatus may further include a memory for storing instructions and / or data. The memory is coupled to the processor, which, when executing the instructions stored in the memory, can implement the methods described in the foregoing aspects. Optionally, the memory and the processor are integrated together.

[0084] The device may also include a communication interface for communicating with other devices. For example, the communication interface may be a transceiver, circuit, bus, module or other type of communication interface.

[0085] In one aspect, this application provides a chip system including at least one processor for supporting the functions involved in any of the above aspects and any possible implementations of any aspect.

[0086] In one possible design, the chip system also includes a memory for storing program instructions and data, which may be located within or outside the processor.

[0087] The chip system can consist of chips or include chips and other discrete components.

[0088] In a twelfth aspect, this application provides a computer-readable storage medium including a computer program that, when run on a computer, causes the computer to implement the methods in any of the foregoing aspects and any possible implementations of any of the foregoing aspects.

[0089] In a thirteenth aspect, this application provides a computer program product comprising: a computer program (also referred to as code or instructions) that, when run, causes a computer to perform the methods of any of the foregoing aspects and any possible implementation thereof. Attached Figure Description

[0090] Figure 1 is a schematic diagram of the architecture of a communication system provided in an embodiment of this application;

[0091] Figure 2 is a schematic diagram of a communication system based on a split architecture applicable to embodiments of this application;

[0092] Figure 3 shows the network element function division and protocol layer structure of an open access network device;

[0093] Figure 4 illustrates a wireless air interface AI / ML framework;

[0094] Figure 5 is a flowchart illustrating a communication method provided in one embodiment of this application;

[0095] Figure 6 is a flowchart illustrating a communication method provided in one embodiment of this application;

[0096] Figure 7 is a flowchart illustrating a communication method provided in another embodiment of this application;

[0097] Figure 8 is a flowchart illustrating a communication method provided in yet another embodiment of this application;

[0098] Figure 9 is a structural schematic diagram of a communication device provided in an embodiment of this application;

[0099] Figure 10 is a structural schematic diagram of a communication device provided in another embodiment of this application. Detailed Implementation

[0100] Figure 1 is a schematic diagram of the architecture of a communication system 10 provided in an embodiment of this application. It is understood that the system architecture described in this embodiment is for the purpose of more clearly illustrating the technical solutions of this application and does not constitute a limitation on the technical solutions provided in this embodiment.

[0101] As shown in Figure 1, the communication system 10 includes a radio access network (RAN) 100 and a core network (CN) 200. RAN 100 includes at least one RAN node (110a and 110b in Figure 1, collectively referred to as 110) and at least one terminal (120a-120j in Figure 1, collectively referred to as 120). RAN 100 may also include other RAN nodes, such as wireless relay equipment and / or wireless backhaul equipment (not shown in Figure 1). Terminal 120 is wirelessly connected to RAN node 110. RAN node 110 is wirelessly or wired connected to core network 200. The core network equipment in core network 200 and RAN node 110 in RAN 100 can be different physical devices, or they can be the same physical device integrating core network logical functions and radio access network logical functions.

[0102] RAN 100 can be a cellular system related to the 3rd Generation Partnership Project (3GPP), such as 4G, 5G mobile communication systems, or future-oriented evolution systems. RAN 100 can also be an open RAN (O-RAN or ORAN), a cloud radio access network (CRAN), or a wireless fidelity (WiFi) system. RAN 100 can also be a communication system that integrates two or more of the above systems.

[0103] RAN node 110, sometimes also referred to as access network equipment, RAN entity, or access node, constitutes part of the communication system and is used to help terminals achieve wireless access. Multiple RAN nodes 110 in communication system 10 can be of the same type or different types. In some scenarios, the roles of RAN node 110 and terminal 120 are relative. For example, network element 120i in Figure 1 can be a helicopter or drone, which can be configured as a mobile base station. For terminals 120j accessing RAN 100 through network element 120i, network element 120i is a base station; but for base station 110a, network element 120i is a terminal. RAN node 110 and terminal 120 are sometimes both referred to as communication devices. For example, network elements 110a and 110b in Figure 1 can be understood as communication devices with base station functions, and network elements 120a-120j can be understood as communication devices with terminal functions.

[0104] In one possible scenario, the RAN node can be a base station, an evolved NodeB (eNodeB), an access point (AP), a transmission reception point (TRP), a next-generation NodeB (gNB), a base station in a future mobile communication system, or an access node in a WiFi system. The RAN node can be a macro base station (as shown in Figure 1, 110a), a micro base station or indoor station (as shown in Figure 1, 110b), a relay node or donor node, or a radio controller in a CRAN scenario. Optionally, the RAN node can also be a server, wearable device, vehicle, or in-vehicle equipment. For example, the access network equipment in vehicle-to-everything (V2X) technology can be a roadside unit (RSU). All or part of the functions of the RAN node in this application can also be implemented through software functions running on hardware, or through virtualization functions instantiated on a platform (e.g., a cloud platform). The RAN node can also be equipped with communication modules, circuits, or chips that perform corresponding communication functions. The RAN node can also be configured with program instructions for performing corresponding communication functions, as well as corresponding program instructions. The RAN node in this application can also be a logical node, logical module, or software capable of implementing all or part of the RAN node's functions.

[0105] In another possible scenario, multiple RAN nodes collaborate to assist the terminal in achieving wireless access, with each RAN node performing a portion of the base station's functions. For example, RAN nodes can be central units (CUs), distributed units (DUs), CU-control plane (CPs), CU-user plane (UPs), or radio units (RUs), etc. CUs and DUs can be separate entities or included in the same network element, such as a baseband unit (BBU). RUs can be included in radio frequency equipment or radio frequency units, such as remote radio units (RRUs), active antenna units (AAUs), or remote radio heads (RRHs).

[0106] In different systems, CU (or CU-CP and CU-UP), DU, or RU may have different names, but those skilled in the art will understand their meaning. For example, in an ORAN system, CU can also be called O-CU (open CU), DU can also be called O-DU, CU-CP can also be called O-CU-CP, CU-UP can also be called O-CU-UP, and RU can also be called O-RU. For ease of description, this application uses CU, CU-CP, CU-UP, DU, and RU as examples. Any of the units among CU (or CU-CP, CU-UP), DU, and RU in this application can be implemented through software modules, hardware modules, or a combination of software and hardware modules.

[0107] The terminal involved in the embodiments of this application can be a device or module that accesses the aforementioned communication system and has corresponding communication functions. The terminal can also be called a terminal device, user equipment (UE), mobile station, mobile terminal, etc. Terminals can be widely used in various scenarios, such as device-to-device (D2D), vehicle-to-everything (V2X) communication, machine-type communication (MTC), Internet of Things (IoT), virtual reality, augmented reality, industrial control, autonomous driving, telemedicine, smart grids, smart furniture, smart offices, smart wearables, smart transportation, smart cities, etc. Terminals can be mobile phones, tablets, computers with wireless transceiver capabilities, wearable devices, vehicles, drones, helicopters, airplanes, ships, robots, robotic arms, smart home devices, transportation vehicles with wireless communication capabilities, communication modules, etc. The embodiments of this application do not limit the device form of the terminal. The terminal typically contains a communication module, circuit, or chip that performs the corresponding communication functions. The terminal can also be configured with program instructions for performing the corresponding communication functions.

[0108] Referring to Figure 2, which is a schematic diagram of a communication system based on a split architecture applicable to embodiments of this application, as shown in Figure 2, the access network device communicates with the core network (CN) device via a backhaul link and with the terminal via an air interface. Specifically, the BBU in the access network device communicates with the core network device via a backhaul link, the RU in the access network device communicates with the terminal via an air interface, and the BBU communicates with at least one RU via a fronthaul link (FH). The BBU and RU may be co-located or not. The BBU may include at least one CU and at least one DU, and the CU and DU can communicate with each other via a midhaul link.

[0109] Referring to Figure 3, which illustrates the network element function division and protocol layer structure of an O-RAN device, the access network device can be divided into CU, DU, and RU.

[0110] In some examples, the CU is a logical node carrying the radio resource control (RRC) layer, service data adaptation protocol (SDAP) layer, packet data convergence protocol (PDCP) layer, and other control functions of the access network equipment. The CU connects to network nodes such as the core network through interfaces, which can be interfaces such as E2 interfaces. Optionally, the CU may have some core network functions. The CU (e.g., PDCP layer and higher layers) connects to the DU (e.g., RLC layer and lower layers) through interfaces, which can be interfaces such as F1 interfaces. In some examples, these interfaces (e.g., F1 interfaces) can provide control plane (C-Plane) and user plane (U-Plane) functions (e.g., interface management, system information management, UE context management, RRC message transmission, etc.). F1AP is the application protocol of the F1 interface, defining the F1 signaling procedures in some examples. The F1 interface supports control plane F1-C and user plane F1-U.

[0111] In some examples, the CU can be split into CU-CP (control unit-control plane) and CU-UP (control unit-user plane). CU-CP is a logical node carrying the RRC layer and PDCP-C (control plane part of PDCP) layer, used to implement the CU's control plane functions. CU-CP can interact with network elements in the core network used to implement control plane functions. These network elements in the core network can be access and mobility function (AMF) network elements, such as the access and mobility management function (AMF) in a 5G system. The AMF network element is responsible for mobility management in the mobile network, such as terminal device location updates, terminal device registration with the network, and terminal device handover. CU-UP is a logical node carrying the SDAP layer and PDCP-U (user plane part of PDCP) layer, used to implement the CU's user plane functions. CU-UP can interact with network elements in the core network used to implement user plane functions. These network elements in the core network, such as the UPF (user plane function) in a 5G system, are responsible for data forwarding and receiving in terminal devices. The above CU and DU configurations are merely examples; the functions of the CU and DU can be configured as needed. For instance, the CU or DU can be configured to have more protocol layer functions, or only some protocol layer processing functions. For example, some RLC layer functions and protocol layer functions above the RLC layer can be placed in the CU, while the remaining RLC layer functions and protocol layer functions below the RLC layer can be placed in the DU. Furthermore, the functions of the CU or DU can be divided according to service type or other system requirements, such as by latency. Functions that require low latency can be placed in the DU, while functions that do not require low latency can be placed in the CU.

[0112] In some examples, a DU is a logical node carrying the radio link control (RLC) layer, medium access control (MAC) layer, higher physical layer (PHY) layer, and other functions. In some examples, a DU can control at least one RU. The DU connects to the RU through interfaces, which can be fronthaul interfaces. In some examples, the Higher PHY layer includes the PHY layer processing, such as forward error correction (FEC) encoding and decoding, scrambling, modulation, and demodulation.

[0113] In some examples, the RU is a logical node carrying both lower physical layer (PHY) and radio frequency (RF) processing. In some examples, the RU can be a 3GPP transmission reception point (TRP), a remote radio head (RRH), or other similar entities. In some examples, the Low-PHY includes PHY processing functions such as Fast Fourier Transform (FFT), Inverse Fast Fourier Transform (IFFT), digital beamforming, and filtering. The RU communicates with one or more UEs via a radio link.

[0114] The DU and RU can be co-located or not. The DU and RU exchange control plane and user plane information via a lower-layer split-control, user, and synchronization (LLS-CUS) interface through a fronthaul link. LLS-CUS may include LLS-C and LLS-U interfaces that provide the control plane (C-Plane) and user plane (U-Plane), respectively. In some examples, the control plane (C-Plane) refers to real-time control between the DU and RU. The DU and RU exchange management information via an LLS-M interface on the fronthaul link; the management plane (M-Plane) refers to non-real-time management operations between the DU and RU.

[0115] DU and RU can cooperate to implement the functions of the PHY layer. A DU can be connected to one or more RUs. The functions of DU and RU can be configured in various ways depending on the design. For example, a DU can be configured to implement baseband functions, and an RU can be configured to implement mid-RF functions. Another example is that a DU can be configured to implement higher-level functions in the PHY layer, and an RU can be configured to implement lower-level functions in the PHY layer, or to implement both lower-level and RF functions. Higher-level functions in the physical layer can include a portion of the physical layer's functions that are closer to the MAC layer, while lower-level functions in the physical layer can include another portion of the physical layer's functions that are closer to the mid-RF side.

[0116] To better understand the technical solutions of the embodiments of this application, some concepts used in the embodiments of this application will be introduced first.

[0117] 1. Machine Learning

[0118] Machine learning (ML) is an important technological approach to achieving artificial intelligence (AI). AI enables machines to possess human-like intelligence, using computer hardware and software to simulate certain intelligent behaviors of humans, including machine learning and other methods. Machine learning refers to learning models or rules from raw data, such as neural networks, decision trees, and support vector machines. Machine learning can be divided into supervised learning, unsupervised learning, and reinforcement learning.

[0119] Supervised learning, based on collected sample values ​​(or samples) and sample labels, uses machine learning algorithms to learn the mapping relationship between sample values ​​and sample labels, and expresses this learned mapping relationship using a machine learning model. The process of training the machine learning model is the process of learning this mapping relationship. For example, in signal detection, the noisy received signal is the sample, and the corresponding real constellation point is the label. Machine learning aims to learn the mapping relationship between samples and labels through training, that is, to enable the machine learning model to learn a signal detector. During training, the model parameters are optimized by calculating the error between the model's predicted values ​​and the real labels. Once the mapping relationship is learned, it can be used to predict the sample label of each new sample. The mapping relationship learned in supervised learning can include linear mappings and nonlinear mappings. Based on the type of label, the learning task can be divided into classification tasks and regression tasks.

[0120] Unsupervised learning relies solely on collected sample values, using algorithms to discover inherent patterns within the samples. One type of unsupervised learning algorithm uses the samples themselves as supervisory signals; that is, the model learns the mapping relationship from sample to sample, which is called self-supervised learning. During training, model parameters are optimized by calculating the error between the model's predictions and the samples themselves. Self-supervised learning can be used for signal compression and decompression recovery applications; common algorithms include autoencoders and generative adversarial networks.

[0121] Reinforcement learning, unlike supervised learning, is a type of algorithm that learns problem-solving strategies through interaction with the environment. Unlike supervised and unsupervised learning, reinforcement learning problems do not have explicit "correct" action labels. The algorithm needs to interact with the environment to obtain reward signals from the environment, and then adjust its decision actions to obtain a larger reward signal value. For example, in downlink power control, the reinforcement learning model adjusts the downlink transmission power of each terminal device based on the total system throughput feedback from the wireless network, aiming to achieve a higher system throughput. The goal of reinforcement learning is also to learn the mapping relationship between the environment state and the optimal decision action. However, because the label of the "correct action" cannot be obtained in advance, the network cannot be optimized by calculating the error between the action and the "correct action." Reinforcement learning training is achieved through iterative interaction with the environment.

[0122] Deep neural networks (DNNs) are a specific implementation of machine learning. According to the general approximation theorem, neural networks can theoretically approximate any continuous function, thus enabling them to learn arbitrary mappings. Traditional communication systems rely on extensive expert knowledge to design communication modules, while DNN-based deep learning communication systems can automatically discover hidden pattern structures from large datasets, establish mapping relationships between data, and achieve performance superior to traditional modeling methods.

[0123] Based on their construction method, DNNs can be divided into feedforward neural networks (FNNs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs). FNNs can be neural networks where neurons in adjacent layers are completely connected pairwise, which makes FNNs typically require a large amount of storage space and have high computational complexity.

[0124] CNNs are neural networks specifically designed to process data with a grid-like structure. For example, time-series data (discrete sampling along the time axis) and image data (two-dimensional discrete sampling) can both be considered grid-like data. CNNs do not use all the input information at once for computation; instead, they use a fixed-size window to extract a portion of the information for convolution operations, which significantly reduces the computational cost of model parameters. Furthermore, depending on the type of information extracted by the window (such as people and objects in an image representing different types of information), each window can use different convolution kernels, allowing CNNs to better extract features from the input data.

[0125] Recurrent Neural Networks (RNNs) are a type of distributed neural network (DNN) that utilizes feedback time-series information. Their input includes the current input value and their own output value from the previous time step. RNNs are well-suited for acquiring temporally correlated sequence features, and are particularly applicable to applications such as speech recognition and channel coding / decoding.

[0126] 2. AI Model

[0127] AI models refer to function models that map inputs of a certain dimension to outputs of a certain dimension, and the parameters of these models can be obtained through machine learning training. For example, f(X) = aX² + b is a quadratic function model, which can be viewed as an AI model. a and b correspond to the parameters of this model and can be obtained through machine learning training.

[0128] In machine learning, the data used for model training, validation, and / or testing can form a dataset or training dataset. The quantity and / or quality of data in the dataset or training dataset will affect the performance of the machine learning. Model training involves selecting an appropriate loss function (which measures the difference between the model's predictions and the true values) and using optimization algorithms to train the model parameters to minimize the loss function value. Model testing involves evaluating the model's performance using test data after training. Model application involves using the trained model to solve real-world problems.

[0129] A neural network, or artificial neural network, is a mathematical model that mimics the behavioral characteristics of animal neural networks to perform distributed parallel information processing. It is a special form of AI model.

[0130] 2. Model Training

[0131] Model training involves selecting an appropriate function (such as a loss function) and using optimization algorithms to train the model parameters so that the difference between the model's predicted values ​​and the ground truth (or target values, labels) tends to be minimized.

[0132] For example, model training methods include, but are not limited to, supervised learning, self-supervised learning, and knowledge distillation.

[0133] 3. Model files and model parameters

[0134] Model files and / or model parameters can be used to determine the model.

[0135] The model file can be used to indicate the model structure, which may include, but is not limited to, Transformer neural networks, DNN, CNN, FNN, RNN, multilayer perceptron (MLP), etc.

[0136] Model parameters can refer to the parameters in a neural network model. Model parameters include, but are not limited to, the number of layers in the neural network, the type and weight of neurons in each layer of the neural network, etc.

[0137] Take DNN as an example. The idea behind DNN comes from the neuronal structure of the brain. Each neuron can perform a weighted summation operation on its inputs and then use the result of the weighted summation operation to generate the output through a non-linear function. For example, the input of a neuron is x = [x0, x1, ..., x...]. N-1 The weights corresponding to the inputs are w = [w0, w1, ..., w] N-1 The bias of the weighted summation is b. The nonlinear function f() can take many forms; for example, the nonlinear function f() can be the maximum value function max{0, x}. Then the effect of a neuron's execution is... Where N is a positive integer, and n is a positive integer greater than or equal to 0 and less than or equal to (N-1). The weights of the weighted summation operation of neurons in a neural network and the nonlinear function are called the parameters of the neural network. The parameters of all neurons in a neural network constitute the parameters of the neural network.

[0138] A typical DNN has multiple neural network layers, including an input layer, one or more hidden layers, and an output layer. Generally, the first layer is the input layer, the last layer is the output layer, and the layers in between are hidden layers. Each layer contains multiple neurons. Layers are fully connected; that is, any neuron in the i-th layer is connected to any neuron in the (i+1)-th layer. The input layer processes the received values ​​(i.e., the DNN's input) through neurons and then passes them to the hidden layers. Similarly, the hidden layers pass the computation results to the final output layer, producing the DNN's output.

[0139] 4. Model Reasoning

[0140] Model inference, also known as AI model inference or AI reasoning, refers to the process of obtaining output data from a trained model by inputting data into the model. In other words, model inference is the process of obtaining output data based on a trained AI model by inputting data into the trained AI model.

[0141] With the development of mobile communication, the application of AI / ML technology in mobile communication networks has been proposed to improve communication performance. 3GPP is researching several use cases for optimization using AI / ML technology, such as beam management, positioning, or channel state information (CSI) feedback. 3GPP Release R 18 also proposed the radio interface AI / ML framework shown in Figure 4. Referring to Figure 4, the AI / ML system includes a data collection unit, a model training unit, a management unit, an inference unit, and a model storage unit. The data collected by the data collection unit is used in different AI / ML stages, such as the large amount of diverse data required for the model training stage (training data), the real-time data required for the model inference stage (inference using the trained model), and the near real-time large amount of data required for the model monitoring stage (monitoring data). The model training unit trains the model based on the collected training data and transmits the trained model to the model storage unit for storage. The management unit monitors the model's performance based on the collected monitoring data and performs functions such as model activation, deactivation, switching, selection, and rollback based on the model's performance. The inference unit performs inference based on the collected inference data and the model in the storage unit (i.e., the trained model).

[0142] Currently, 3GPP has proposed a dual-end AI model. A dual-end AI model refers to AI inference being performed jointly by an AI model on the terminal side and an AI model on the network side. In other words, a dual-end AI model can be considered as consisting of an AI model on the terminal side and an AI model on the network side, used together to complete AI inference. The AI ​​model on the terminal side is also called the edge AI model, and the AI ​​model on the network side can be simply referred to as the network-side AI model.

[0143] The working principle of a dual-end AI model is illustrated using CSI feedback as an example. In a communication system, the terminal can feed back CSI to the access network equipment, which then adjusts the beam direction, determines a suitable modulation and coding scheme, and rationally allocates and schedules resources. In traditional CSI feedback methods, the terminal compresses the channel matrix to obtain the CSI based on mathematical formulas (e.g., singular value decomposition of the channel matrix) and codebook matching, and then feeds back the CSI to the access network equipment. The CSI includes, for example, a precoding matrix indicator (PMI) and a channel quality indicator (CQI). Correspondingly, the access network equipment can recover the channel matrix based on the CSI fed back by the terminal and perform beamforming and other operations on the data or information sent to the terminal equipment based on the channel matrix. In the AI-based CSI feedback approach (which can be simply called AI CSI feedback), the terminal has an AI model (also known as an AI encoder), and the access network device has an AI model (also known as an AI decoder). Specifically, the AI ​​encoder is used by the terminal to compress the channel matrix into low-dimensional feature vectors (also known as compressed features), and the AI ​​decoder is used by the access network device to recover the channel matrix based on the low-dimensional feature vectors output by the AI ​​encoder. The compression and recovery of the channel matrix are achieved through the combined use of these two AI models; that is, the reasoning for channel matrix compression and recovery is completed based on the combined use of these two AI models.

[0144] In other words, the AI ​​CSI feedback model consists of two parts (the terminal's AI encoder and the access network device's AI decoder). First, the terminal performs an inference process based on the compressed features output by the AI ​​encoder. Then, the access network device performs an inference process based on the compressed features output by the AI ​​decoder to recover the channel matrix. Compared to traditional CSI feedback methods, AI-based channel matrix compression can retain more details, thereby improving the accuracy of CSI feedback.

[0145] In this application, an AI function can be understood as an AI feature, or a configuration that enables an AI feature, such as radio resource control (RRC) configuration. Different AI use cases (e.g., AI CSI feedback, AI beam management, AI positioning, etc.) can be considered as different AI features or different AI functions. An AI model is a specific implementation of an AI function; one AI function may correspond to one or more AI models. For example, taking the aforementioned AI CSI feedback as an example, the AI ​​model corresponding to the AI ​​CSI feedback use case (which can also be considered as the AI ​​CSI feedback function) includes the AI ​​encoder on the terminal side and the AI ​​decoder on the network side. That is, the AI ​​encoder on the terminal side and the AI ​​decoder on the network side are associated with the AI ​​CSI feedback use case (AI CSI feedback function).

[0146] For dual-end AI models, the AI ​​model on the terminal side and the AI ​​model on the network side need to be matched to achieve good AI performance. For example, in the AI ​​CSI feedback use case, to obtain an accurate channel matrix, the low-dimensional feature vector output by the AI ​​encoder on the terminal side should be input into the AI ​​decoder on the network side, and the AI ​​decoder should be able to recover the channel matrix well based on this low-dimensional feature vector. To achieve matching between the terminal-side AI model and the network-side AI model in dual-end AI models, there are currently two methods for achieving matching between dual-end models:

[0147] Implementation method 1): The network side and the terminal side train their respective AI models.

[0148] Specifically, the network side (e.g., a network server) first performs end-to-end model training to obtain a trained dual-end AI model. Then, it sends the AI ​​model used on the network side from the trained dual-end AI model to the network device, and sends information about the endpoint AI model (e.g., dataset, AI model structure, and / or model parameters) to the terminal side. Correspondingly, the terminal side trains based on the information sent by the network side to obtain an AI model equivalent to the endpoint AI model trained on the network side, thus achieving matching between the endpoint AI model and the network-side AI model.

[0149] In implementation method 1), the terminal needs to train a model based on the information sent by the network side to obtain an edge AI model that matches the network-side AI model. Since the terminal's local computing power is limited, the process of training the edge AI model typically occurs on the terminal's server (UE-side server, or OTT server). Specifically, after receiving the information used in the training process from the network side, the terminal sends this information to the terminal's server. Correspondingly, the terminal's server trains the edge AI model based on this information and then sends the trained edge AI model back to the terminal.

[0150] Implementation method 2):

[0151] The network side (e.g., a network-side server) first performs end-to-end model training to obtain a trained dual-end AI model. Then, the AI ​​model used on the network side from the trained dual-end AI model is sent to the network device, and the trained end-side AI model is instructed to the terminal, thereby achieving the matching between the end-side AI model and the network-side AI model.

[0152] However, for implementation method 1), there is a problem of high air interface resource overhead. As for implementation method 2), the AI ​​models at both ends are developed entirely by one end. In this case, there is a possibility that the AI ​​model on the terminal side instructed by the network side may not be applicable to the terminal side.

[0153] In view of this, this application provides a communication method and a communication device that enables a terminal to obtain the terminal-side AI model in a dual-end model used to complete the first AI function, so as to jointly complete the first AI function with the network-side AI model in the dual-end model.

[0154] The technical solution provided in this application will now be described with reference to the accompanying drawings.

[0155] Referring to Figure 5, which is a flowchart illustrating a communication method 500 provided in an embodiment of this application, the various steps of method 500 are described in detail below.

[0156] S510, the first network device sends first information; correspondingly, the terminal receives the first information; the first information is used to instruct the first target node, the first target node is used to obtain second information, wherein the second information is used to obtain the first AI model, the first AI model is used by the terminal, and the first AI model and the second AI model used by the first network device jointly complete the first AI function.

[0157] The first network device refers to the network device currently accessed by the terminal. This network device can also be replaced by access network equipment, base stations, etc. In this method, the cell currently accessed by the terminal is referred to as the first cell. Understandably, the first cell belongs to the first network device; that is, the first network device supports / manages / controls the first cell.

[0158] The first network device can indicate a first target node to the terminal through first information. For example, the first information includes the address of the first target node. The address of the first target node is, for example, its Internet Protocol (IP) address. In this application, the first target node is used to obtain second information, and the second information is used to obtain the first AI model.

[0159] Specifically, the aforementioned first AI model refers to an AI model used on a terminal that, together with a second AI model used on a first network device, performs the first AI function. In other words, the AI ​​model used to perform the first AI function can be considered to be divided into two parts: a first AI model for the terminal and a second AI model for the first network device. That is, the AI ​​model used to perform the first AI function can be considered to consist of a first AI model and a second AI model; the first AI model is a part of the AI ​​models used to perform the first AI function, and the second AI model is the other part. For example, the AI ​​model composed of the first AI model and the second AI model can also be called a first dual-end AI model. A dual-end AI model can also be called a bilateral AI model, etc.

[0160] The joint completion of the first AI function by the first AI model and the second AI model can also be understood as: both the first AI model and the second AI model are related to / correspond to the first AI function. AI functions can also be replaced with AI features / scenarios / use cases, etc., and the first AI function can also be replaced with the first AI feature / scenarios / use cases, etc. For example, AI CSI feedback, AI beam management, and AI localization can be considered different AI functions.

[0161] Specifically, in method 500, the first AI model is obtained through training. The aforementioned second information is the information required to train the first AI model; therefore, in this application, the second information can also be referred to as training information.

[0162] In implementation, the first target node can, for example, obtain a first dual-end AI model through model training. Then, it instructs the first network device on a second AI model derived from the trained first dual-end AI model and stores second information used to obtain the first AI model. For example, the second information may include at least one of the following for obtaining the first AI model: a first dataset, a first model structure, and / or first model parameters. In other words, the first target node is a node used to train or store data related to the first dual-end AI model. For example, the first target node can be a server or a network-side AI functional entity. Understandably, when the first target node is updated, the first network device indicates the updated address of the first target node to the terminal.

[0163] The first target node can be a server deployed on the network side, or a network element or logical functional entity used for data collection or model training on the network side. For example, it can be a data collection coordination function (DCCF) or a data collection function (DCF), or a network element managed by a network equipment vendor or operator.

[0164] Optionally, method 500 may further include S501: the first network device sends at least one of the following information to the terminal: information indicating that the first network device has a first AI function (which may be referred to as information 1), information indicating a first mode (which may be referred to as information 2), information indicating the model structure of the second AI model (which may be referred to as information 3), and an identifier of the second information (which may be referred to as information 4).

[0165] Indicating that the first network device has the first AI function can also be understood as indicating that the first network device supports the first AI function.

[0166] The information used to indicate the first method can also be described as: information indicating the first method associated with / corresponding to the first AI mode that matches the second AI model. Therefore, the first method can also be described as a first pairing method, a first docking method, or a first matching method.

[0167] For the first approach, for example, the first AI model can be obtained based on a first dataset, or it can be obtained based on a first model structure and / or first model parameters. Understandably, the information used to indicate the first approach can also be understood as information indicating the first approach supported by the first network device. When the first approach indicated by the first network device is obtaining the first AI model based on the first model parameters, the model structure of the first AI model can be predefined by the protocol, or it can be indicated by the first network device to the terminal.

[0168] The information used to indicate the model structure of the second AI model can also be understood as information used to indicate the model structure of the second AI model when the first network device performs the first AI function. This application does not limit the specific form of the model structure of the second AI model. For example, when the first network device indicates the model structure of the second AI model, information 3 includes the model file of the second AI model, such as the number of layers included in the second AI model and the number of parameters in each layer. In one possible implementation, the protocol may predefine one or more model structures for the second AI model, with each model structure under the second AI model corresponding to a unique identifier. In this case, the first network device can indicate the identifier of the second AI model structure when indicating the model structure of the second AI model. The information used to indicate the model structure of the second AI model can also be replaced with any of the following: information used to indicate the model structure of supported second AI models.

[0169] The identifier of the aforementioned second information can also be described as a version number of the second information. Different version numbers of the second information are different; one version number can be understood as corresponding to one version of the second information. Understandably, there is a possibility that the first target node may update the first dual-end AI model, in which case the second AI model and the second information may also be updated. Therefore, the first network device can indicate to the terminal the version of the second information upon which the first AI model matching the second AI model should be based, so that the terminal can know the version of the second information associated with the second AI model matching the first AI model, thereby further obtaining the first AI model matching the second AI model. For example, taking AI CSI feedback as the first AI function, after training the first dual-end AI model, the first target node can indicate the decoder (which can be considered the second AI model) used by the first network device to the first network device, and indicate the version of the encoder (which can be considered the first AI model) used to obtain the terminal to the first network device. Furthermore, the first network device indicates this version to the terminal to indicate the version of the second information supported by the first network device.

[0170] The aforementioned information 1, information 2, information 3, and information 4 can be sent in different ways. For example, the first network device may send information 1, information 2, information 3, and information 4 to the terminal in multiple messages. Alternatively, the first network device may send information 1, information 2, information 3, and information 4 to the terminal in a single message. Another example is that the first network device may send information 1, information 2, information 3, and information 4 to the terminal simultaneously. Yet another example is that the first network device may not send information 1, information 2, information 3, and information 4 to the terminal simultaneously.

[0171] Optionally, the first network device may carry information 1, information 2, information 3 and information 4 in a broadcast message to indicate to the terminal, for example, information 1, information 2, information 3 and information 4 may be carried in a system message of the first network device.

[0172] Optionally, the first network device sends at least one of information 1, information 2, information 3, and information 4 to the terminal before instructing the terminal to the first target node via the first information. Alternatively, the first network device simultaneously instructs the terminal to the first target node and at least one of information 1, information 2, information 3, and information 4.

[0173] There are different ways to implement the first network device sending the first information.

[0174] In one implementation, the first information may be carried in a broadcast message, and the terminal receives the first information based on the broadcast message of the first network device.

[0175] In another implementation, referring to Figure 5, method 500 may include S502: the terminal sends third information to the first network device, the third information being used to request the first AI model. Correspondingly, after receiving the third information, the first network device sends first information to the terminal. The third information being used to request the first AI model can also be understood as: the third information being used to request second information, or the third information being used to request the second information required for training the first AI model, or the third information being used to request training information.

[0176] There are different ways for a terminal to determine that the first network device is sending third information:

[0177] For example, the terminal stores information indicating the network device sets corresponding to at least one target node. For instance, the terminal includes the identifiers of the network devices in the network device sets corresponding to the addresses of at least one target node. After the terminal accesses the first cell, it can obtain the identifier of the first network device associated with the first cell through the system information of the first cell. If the identifier of the first network device is not included in any of the stored network device sets, the terminal determines that there is no address of the first target node associated with the first network device, thus determining that there is no / does not have / does not contain the first AI model, and then sends third information to the first network device.

[0178] For example, the terminal stores information indicating the cell sets corresponding to at least one target node, such as the identifiers of the cells included in the cell sets corresponding to the addresses of at least one target node. After the terminal accesses the first cell, it can obtain the identifier of the first cell through the system information of the first cell, such as the cell global identifier (CGI). If the identifier of the first cell is not included in any of the stored cell sets, the terminal determines that there is no address of the first target node associated with the first network device, thus determining that there is no / no first AI model, and then sends third information to the first network device.

[0179] For example, the terminal can determine whether to send third information based on whether it supports the first AI function. For instance, if the terminal does not have the address of the first target node and the terminal supports the first AI function, the terminal determines to send third information to the first network device. Understandably, if the terminal determines whether to send third information to the first network device based on whether it supports the first AI function, the information 1 sent by the first network device will precede the third information.

[0180] For example, in one implementation, the terminal can determine whether to send the first information based on a first method indicated by the first network device. For instance, if the terminal does not have the address of the first target node and the terminal supports the first method, the terminal determines to send the third information to the first network device. The terminal supporting the first method can be understood as the terminal supporting the first AI model trained based on the first method.

[0181] For example, if the terminal supports obtaining the first AI model based on the first model structure and / or the first model parameters, and the first method indicated by the first network device is that the first AI model is obtained based on the first dataset, the terminal determines not to send the third information to the first network device. Understandably, if the terminal determines whether to send the fourth information to the first network device based on the first method indicated by the network device, the information 2 sent by the first network device will precede the fourth information.

[0182] S520: The terminal obtains the first AI model based on the first information.

[0183] One implementation of the terminal obtaining the first AI model based on the first information includes: the terminal sending information indicating the first target node to a terminal-side server (hereinafter referred to as the terminal-side server), such as the terminal sending the address of the first target node to the terminal-side server; the terminal-side server finding the first target node based on the terminal's indication and requesting second information from the first target node; the first target node responding to the terminal-side server's request sending the second information to the terminal-side server; and the terminal-side server training the first AI model based on the second information and indicating the first AI model to the terminal. The aforementioned terminal-side server can be deployed by the terminal manufacturer or the chip manufacturer.

[0184] Based on method 500, the first network device indicates the first target node used to obtain the second information of the first AI model to the terminal. In this way, the terminal can indicate the first target node used to obtain the second information of the first AI model to the end-side server, causing the end-side server to request the second information from the first target node, then train the first AI model based on the second information, and indicate the first AI model to the terminal. This allows the terminal to obtain the first AI model that, together with the second AI model used by the first network device, completes / implements the first AI function. Furthermore, based on this technical solution, the first network device does not need to send the second information to the terminal, thus saving air interface resources.

[0185] On the other hand, since the first network device can send at least one of information 1, information 2, information 3, and information 4 to the terminal, the terminal can obtain at least one of the first method and the version number of the second information corresponding to the first AI model that matches the second AI model. This helps to improve the matching degree between the first AI model and the second AI model obtained by the terminal, thereby improving the inference performance of the dual-end AI model including the first AI model and the second AI model.

[0186] On the other hand, a single edge server may support training AI models for multiple terminals from the same vendor. Therefore, only one terminal needs to request the address of the first target node from the first network device. After the edge server completes model training and obtains the first AI model, it can push the first AI model to multiple terminals. This eliminates the need for each terminal to request the first AI model from the network separately, thus saving signaling overhead. Understandably, the decision of which terminal requests the first target node from the first network device can be left to the implementation; for example, it could be a terminal device with a special user identity.

[0187] Optionally, the first network device or the first target node may indicate the geographical range associated with the second information to the terminal. For example, the first network device or the first target node may send fourth information to the terminal, and the terminal receives the fourth information. The fourth information is used to indicate the first set of network devices associated with the second information, wherein the first set of network devices includes at least one network device, and the at least one network device includes the first network device; or, the fourth information is used to indicate the first set of cells associated with the first AI model, wherein the first set of cells includes at least one cell, and the at least one cell includes the first cell currently accessed by the terminal.

[0188] That is, the fourth piece of information can be considered as indicating to the terminal the geographical range to which the first AI model, determined based on the second piece of information mentioned above, can be applied.

[0189] Optionally, the aforementioned fourth information can also be sent from the end-side server to the terminal. For example, when the first target node sends the second information to the end-side server, it also indicates the first network device set / first cell set associated with the first AI model obtained based on the second information. Correspondingly, the terminal-side server or the first network device set / first cell set associated with the first AI model then instructs the terminal.

[0190] By indicating the geographical range associated with the second information to the terminal, when the terminal moves within the coverage area corresponding to the first set of network devices, the terminal only needs to obtain the address of the first target node once, without having to request the address of the first target node from each network device in the first set of network devices, thus saving the terminal's signaling overhead.

[0191] Understandably, in the embodiment shown in method 500, the first network device indicates a first target node to the terminal, and then the terminal indicates the first target node to the end-side server, thereby causing the end-side server to request second information from the first target node. Based on the second information, a first AI model is then trained and indicated to the terminal. Optionally, as another implementation, the terminal can establish a data transmission path with the first target node based on the address indicated by the first network device, then obtain second information (such as a dataset) from the first target node, and then train the first AI model based on the second information. That is, in this implementation, the first AI model is trained locally on the terminal, without requiring training on the end-side server. Optionally, as yet another implementation, the terminal can also directly perform inference using the model structure / model parameters locally, without training. In this implementation, the terminal can establish a data transmission path with the first target node based on the address indicated by the first network device to obtain second information (such as model structure and / or model parameters) from the first target node, and then obtain the first AI model. The aforementioned data transmission path can be a data plane path or a protocol data unit session (PDU) session, carrying the second information through one or more radio access bearers. The aforementioned radio access bearer can be a specific bearer.

[0192] Referring to Figure 6, which is a flowchart illustrating a communication method 600 provided in an embodiment of this application, the various steps of method 600 are described in detail below.

[0193] In S610, the terminal sends fifth information to the first network device, and the first network device receives the fifth information; the fifth information is used to indicate that the terminal has a third AI model, and the third AI model and the fourth AI model used by the first network device jointly complete the second AI function.

[0194] The fifth piece of information, indicating that the terminal possesses a third AI model, can also be understood as indicating the third AI model available to the terminal. The third AI model refers to the AI ​​model possessed by the terminal that, together with the fourth AI model used in the first network device, performs the second AI function. In other words, the AI ​​model used to perform the second AI function can be considered to be divided into two parts: the terminal's third AI model and the fourth AI model used in the first network device. That is, the AI ​​model used to perform the second AI function can be considered to consist of the third AI model and the fourth AI model; the third AI model is a part of the AI ​​models used to perform the second AI function, and the fourth AI model is another part of the AI ​​models used to perform the second AI function. For example, the AI ​​model composed of the third AI model and the fourth AI model can also be called the second dual-end AI model.

[0195] The joint completion of the second AI function by the third and fourth AI models can also be understood as: both the third and fourth AI models are related to / correspond to the second AI function. The meaning of the AI ​​function can be found in the description in Method 500, and will not be elaborated upon here.

[0196] For example, the terminal can include the fifth piece of information in an RRC message sent to the first network device. Alternatively, the terminal can include the fifth piece of information in UE assistance information or other dedicated RRC messages.

[0197] For example, the fifth information includes at least one of the following: the identifier of the sixth information, information for indicating the second method, or information for indicating the second target node.

[0198] The sixth information includes at least one of the following for obtaining the third AI model: a second dataset, a second model structure, and / or second model parameters. The identifier of the sixth information may also be described as a version number of the sixth information. Different version numbers of the sixth information are different. Understandably, it is possible that a network-side server (also called a second target node) updates the second dual-end AI model, where the second target node is a node used to train or store data related to the second dual-end AI model. In this case, the fourth AI model and the sixth information may also be updated. The terminal can indicate the version of the sixth information on which the third AI model is based to the first network device, thereby enabling the first network device to determine whether the third AI model and the fourth AI model match.

[0199] The second method can be any of the following: the third AI model is obtained based on the second dataset, or the third AI model is obtained based on the second model structure and / or the second model parameters. That is, it can be understood as the second method used when the terminal instructs the first network device to obtain the third AI model. The second method can also be described as a second pairing method, a second docking method, or a second matching method.

[0200] Information used to indicate the second target node is, for example, the IP address of the second target node.

[0201] S620, the first network device sends the seventh information to the terminal. The seventh information is used to indicate the mismatch between the third AI model and the fourth AI model or the reason for the mismatch. The reason for the mismatch includes one or more of the following: the identifier of the sixth information does not match, or the second method does not match.

[0202] In method 600, after the terminal reports that it has a third AI model, the first network device can determine whether the third AI model matches the fourth AI model. If they do not match, the first network device sends the seventh information to the terminal.

[0203] The first network device can determine whether the third AI model and the fourth AI model match in different ways.

[0204] For example, if the fifth information includes the identifier of the sixth information, that is, if the terminal reports a version of the sixth information associated with the third AI model, the first network device can determine whether the third AI model and the fourth AI model match based on the version of the sixth information.

[0205] This application does not restrict the specific implementation method of determining whether the third AI model and the fourth AI model of the terminal match based on the version of the sixth information reported by the terminal.

[0206] For example, if the sixth information based on the third AI model that matches the fourth AI model should be version #2, but the version of the sixth information based on the third AI model reported by the terminal is version #1, then it is considered that they cannot match.

[0207] For example, the dataset version associated with the third AI model that matches the fourth AI model should be version #2, while the dataset version associated with the third AI model reported by the terminal is version #1, thus indicating a mismatch. In one possible scenario, the first target node updates the second dual-end AI model, and the second target node instructs the first network device on the fourth AI model from the updated second dual-end AI model. However, the terminal does not promptly obtain the updated sixth information for training the third AI model from the second target node, resulting in the third AI model used by the terminal still being trained based on the old version of the dataset, leading to a mismatch between the third and fourth AI models. Understandably, determining whether the third and fourth AI models match based on inconsistent dataset version numbers is merely an example and does not constitute a limitation of this application. For instance, in another implementation, the first network device determines that the dataset version number on which the third AI model matches the fourth AI model should be version #1. Therefore, if the version number of the dataset on which the third AI model reported by the terminal is based is lower than or equal to (not higher than) version #1, then the third and fourth AI models can be considered mismatched.

[0208] For example, if the fifth piece of information includes information indicating the second method, then the first network device can determine whether the third AI model matches the fourth AI model based on the second method reported by the terminal. For instance, if the third AI model that matches the fourth AI model is based on method 1, then if the second method reported by the terminal is consistent with method 1, then the third AI model is considered to match the fourth AI model; otherwise, it is considered not to match. For example, in one possible scenario, the second target node corresponds to multiple network devices, and these multiple network devices may support different methods for obtaining the third AI model. This could lead to a situation where the network device accessed by the terminal supports a different method for obtaining the third AI model than the method used by the terminal to obtain the third AI model.

[0209] For example, the sixth piece of information includes the address of the second target node. If the address of the second target node is inconsistent with the address of the third target node associated with the fourth AI model, the first network device considers that the third AI model and the fourth AI model do not match.

[0210] In this application, after the first network device determines that the third AI model and the fourth AI model do not match, it sends a seventh message to the terminal to indicate to the terminal the reason for the mismatch between the third AI model and the fourth AI model. For example, the reason for the mismatch between the third AI model and the fourth AI model may include one or more of the following: mismatch of the identifier in the sixth message, or mismatch of the second method.

[0211] In this application, if the terminal receives the seventh information, the terminal can request the terminal-side server to retrain and update the third AI model in order to achieve matching with the fourth AI model.

[0212] In one implementation, when the first network device determines that the third AI model does not match the fourth AI model, it may send at least one of the following matching information (e.g., referred to as matching information) to the terminal: an identifier of the sixth information corresponding to the third AI model that matches the fourth AI model, and information indicating the acquisition method (e.g., referred to as the third method) corresponding to the third AI model that matches the fourth AI model. For example, the identifier of the sixth information corresponding to the third AI model that matches the fourth AI model is the identifier of the dataset or the identifier of the model structure and / or model parameters. After receiving the seventh information, the terminal can update the third AI model based on at least one of the above matching information. For example, the terminal or the terminal-side server may re-acquire the sixth information (dataset / model structure / model parameters), or the terminal or the terminal-side server may retrain the model using the third method to obtain the third AI model. Further, if the terminal-side server updates the third AI model, it can indicate the updated third AI model to the terminal. Understandably, after obtaining the updated third AI model, the terminal can report the updated third AI model to the first network device again.

[0213] For example, the first network device may carry at least one of the above matching information in the seventh information, or it may carry it in a message other than the seventh information.

[0214] In the communication method provided by method 600, the terminal reports its third AI model to the first network device. Correspondingly, the first network device determines whether the third AI model matches a fourth AI model used by the first network device. The third and fourth AI models jointly complete the second AI function. If a mismatch is determined, the terminal is informed of the mismatch or the reason for the mismatch. Thus, if the terminal receives the seventh information, it can request the terminal-side server to retrain and update the third AI model to achieve matching between the third and fourth AI models.

[0215] Understandably, in method 600, if the first network device determines that the third AI model matches the fourth AI model, then S620 may not be included. In the case where the third AI model matches the fourth AI model, the first network device may send an AI inference configuration to the terminal for the terminal to perform inference based on the third AI model. For example, under AI-based CSI feedback, the third AI model is used for CSI compression, and the AI ​​inference configuration for the terminal to perform inference based on the third AI model may include CSI resource configuration and CSI reporting configuration.

[0216] Optionally, in method 600, the first network device may also send at least one of the aforementioned matching information to the terminal before S610. Correspondingly, the terminal instructs the end-side server to train a third AI model that matches the fourth AI model based on the aforementioned at least one matching information. Understandably, in this implementation, S620 may not be included.

[0217] Referring to Figure 7, which is a flowchart illustrating a communication method 700 provided in an embodiment of this application, the various steps of method 700 are described in detail below.

[0218] S710, the terminal sends a first request to the first network device. The first request is used to request the model parameters of the first AI model whose model structure is the target model structure. The first AI model is used by the terminal. The first AI model and the second AI model used by the first network device jointly complete the first AI function.

[0219] Specifically, the aforementioned first AI model refers to an AI model used on a terminal that, together with a second AI model used on a first network device, performs the first AI function. In other words, the AI ​​model used to perform the first AI function can be considered to be divided into two parts: a first AI model for the terminal and a second AI model for the first network device. That is, the AI ​​model used to perform the first AI function can be considered to consist of a first AI model and a second AI model; the first AI model is a part of the AI ​​models used to perform the first AI function, and the second AI model is the other part. For example, the AI ​​model composed of the first AI model and the second AI model can also be called a first dual-end AI model. A dual-end AI model can also be called a bilateral AI model, etc.

[0220] For example, the first request includes an identifier of the target model structure.

[0221] Optionally, prior to S710, method 700 may further include S701: the terminal sends first capability information to the first network device, the first capability information indicating at least one of the following: the terminal supports a first AI function, whether the terminal supports the first network device providing model parameters of a first AI model to the terminal, at least one model structure supported by the terminal for the first AI model, and the maximum parameter threshold of the model parameters of the first AI model supported by the terminal.

[0222] When the first AI model terminal indicates at least one model structure supported by the first capability information, in one implementation, the at least one model structure can be, for example, predefined by the protocol, with each predefined model structure corresponding to a unique identifier. The terminal can then indicate the supported target model structure by indicating the identifier of the target model structure. In another implementation, the terminal can indicate the model file of the target model structure, such as the number of layers included in the target model structure and the number of parameters in each layer.

[0223] The maximum parameter threshold of the first AI model supported by the terminal can also be understood as the maximum storage space of the model parameters of the first AI model that the terminal can store.

[0224] There are different ways to implement the terminal sending the first request.

[0225] For example, if the terminal determines that there is no first AI model, it sends a first request to the first network device. For example, the terminal determines whether there is a first AI model based on stored information indicating the network device set / cell set corresponding to at least one target node. Detailed implementation of this part can be found in the description of the terminal determining whether there is a first AI model in method 500, and will not be repeated here.

[0226] For example, the terminal may determine whether to send a first request based on at least one of the following information sent by the first network device: information indicating that the first network device has a first AI function, information indicating the model structure corresponding to the model parameters of the first AI model that the first network device can provide, an identifier indicating the model parameters of the first AI model that the first network device can provide, and information indicating the set of network devices or the set of cells associated with the model parameters of the first AI model.

[0227] The first network device having the first AI function can also be understood as the first network device supporting the first AI function.

[0228] Information used to indicate the model structure corresponding to the model parameters of the first AI model that the first network device can provide can also be understood as: information used to indicate the AI ​​model structure corresponding to the model parameters of the first AI model that matches the second AI model.

[0229] An identifier used to indicate the model parameters of a first AI model that can be provided by a first network device, which can also be understood as an identifier used to indicate the model parameters of a first AI model that matches a second AI model. The identifier can also be described as a version number.

[0230] Information used to indicate the set of network devices or cell sets associated with the model parameters of the first AI model can also be understood as: the geographical range to which the model parameters of the first AI model are applicable.

[0231] For example, if the terminal supports a first AI function, and the terminal determines that there is no first AI model, but the first network device indicates that the model structure corresponding to the model parameters of the available first AI model is the target model structure supported by the terminal, then the terminal sends a first request to request the first network device to send the model parameters corresponding to the target model structure.

[0232] For example, if the terminal supports the first AI function and the terminal determines that it has the first AI model, but the identifier of the model parameter of the first AI model that the terminal has is inconsistent with the identifier of the model parameter of the first AI model that the first network device can provide, then the terminal sends a first request to request the first network device to send the model parameter corresponding to the target model structure.

[0233] In one implementation, when the terminal requests model parameters of a first AI model whose model structure is the target AI model structure from the first network device via a first request, the first request is also used to indicate a first parameter quantity threshold, so that when the terminal requests the first network device to send the model parameters of the first AI model, the first model parameters whose parameter quantity is less than or equal to the first parameter quantity threshold are sent.

[0234] S720: The first network device sends the model parameters of the first AI model to the terminal.

[0235] In one implementation, if the terminal indicates to the first network device the maximum parameter threshold of the model parameters of the first AI model supported by the terminal, then the first network device can send to the terminal the model parameters of the first AI model whose parameter count is less than or equal to the maximum parameter threshold, so that the total size of the model parameters of the first AI model sent by the first network device does not exceed the maximum parameter threshold supported by the terminal.

[0236] In another implementation, if the first request also indicates a first parameter quantity threshold, then the first network device can send the model parameters of the first AI model whose parameter quantity is less than or equal to the first parameter quantity threshold to the terminal, so that the total size of the model parameters of the first AI model sent by the first network device does not exceed the first parameter quantity threshold of the model parameters of the first AI model requested by the terminal in this transaction.

[0237] After receiving the first request, the first network device may instruct the network-side server to train the first dual-end AI model, wherein the model structure of the first AI model in the trained first dual-end AI model is the target AI model structure. Optionally, if the terminal indicates a maximum parameter threshold / first parameter threshold to the first network device, the first network device may also indicate the maximum parameter threshold / first parameter threshold to the network-side server, so that the model parameters of the first AI model in the trained first dual-end AI model are less than or equal to the maximum parameter threshold / first parameter threshold. Correspondingly, after receiving the instruction from the first network device, the network-side server trains the first dual-end AI model based on the target AI model structure, sends the second AI model in the first dual-end AI model to the first network device, and indicates the model parameters of the first AI model to the first network device, so that the first network device knows the model parameters of the second AI model and the first AI model matching the second AI model.

[0238] Understandably, after the first network device sends the parameters of the first AI model to the terminal, the terminal can obtain the first AI model with the target model structure. That is, the terminal can complete the deployment of the first AI model based on the model parameters provided by the first network device. In other words, after the first network device sends the parameters of the first AI model to the terminal, it can be considered that the terminal has / supports / deploys / obtained the first AI model.

[0239] Optionally, the first network device may provide the terminal with AI inference configuration for the terminal to perform inference based on the first AI model. For example, under AI-based CSI feedback, the first AI model is used for CSI compression, and the AI ​​inference configuration may include CSI resource configuration and CSI reporting configuration. For instance, the first network device may provide the terminal with AI inference configuration for the terminal to perform inference based on the first AI model when providing the first model parameters; or, the first network device may provide the terminal with the first model parameters first, and then provide the terminal with AI inference configuration for the terminal to perform inference based on the first AI model.

[0240] Optionally, after obtaining the first AI model, the terminal can also indicate to the first network device that it has the first AI model. If the first network device also provides an AI inference configuration for the terminal to perform inference based on the first AI model when providing the first model parameters to the terminal, then after the terminal reports the first AI model, for use cases that do not require dynamic activation by the first network device (e.g., periodic CSI reporting), AI inference will begin by activating the first AI model (e.g., providing AI-based CSI feedback).

[0241] Based on the communication method provided by method 700, when the terminal requests the model parameters of the first AI model from the first network device, it also indicates the target model structure of the requested first AI model to the first network device. This enables the first network device to instruct the network-side server to train the first dual-end AI model after receiving the first request. In this way, the first AI model in the trained first dual-end AI model can be applied to the terminal, thereby increasing the probability that the first AI model trained by the network side can be applied to the terminal.

[0242] Referring to Figure 8, which is a flowchart illustrating a communication method 800 provided in an embodiment of this application, the various steps of method 800 are described in detail below.

[0243] S810, the terminal sends a first instruction information to the first network device, and the first network device receives the first instruction information; the first instruction information is used to indicate the third AI model that the terminal has, and the third AI model and the fourth AI model used by the first network device jointly complete the second AI function.

[0244] The first indication information is used to indicate that the terminal has a third AI model, or it can be understood as the first indication information being used to indicate the third AI model available to the terminal. A detailed description of the third AI model and the fourth AI model can be found in S610 of method 600, and will not be repeated here. Similarly, the AI ​​model composed of the third AI model and the fourth AI model is also referred to as the second dual-end AI model.

[0245] For example, the first indication information includes at least one of the following: an identifier for indicating the model structure and / or model parameters of the third AI model possessed by the terminal.

[0246] For the information used to indicate the model structure of the third AI model possessed by the terminal, one implementation may be an identifier of the model structure of the third AI model. Another implementation may be a model file of the third AI model, including information such as the number of layers in the third AI model and the number of parameters in each layer.

[0247] The identifier for the model parameters can also be described as the version number of the model parameters. Model parameters differ across version numbers. Understandably, there may be a situation where a network-side server (also called a second target node) updates the second dual-end AI model, where the second target node is used to train or store data related to the second dual-end AI model. By indicating the identifier of the model parameters associated with the third AI model to the first network device via the terminal, the first network device can determine whether the third AI model and the fourth AI model match.

[0248] There are different ways for the terminal to send the first indication information. For example, the terminal can carry the first indication information in an RRC message sent to the first network device. Alternatively, the terminal can carry the first indication information in UE assistance information or other dedicated RRC messages.

[0249] S820, the first network device sends a second indication message to the terminal. The second indication message is used to indicate the mismatch between the third AI model and the fourth AI model or the reason for the mismatch. The reason for the mismatch includes one or more of the following: the identifiers of the model parameters of the third AI model do not match, or the model structure of the third AI model does not match.

[0250] In method 800, after the terminal reports the third AI model, the first network device can determine whether the third AI model matches the fourth AI model. If they do not match, the first network device sends a second instruction message to the terminal.

[0251] The first network device can determine whether the third AI model and the fourth AI model match in different ways.

[0252] For example, if the first indication information includes an identifier for the model parameters of the third AI model, meaning the terminal reported the version of the model parameters of the third AI model it possesses, the first network device can determine whether the third AI model and the fourth AI model match based on the version of the model parameters of the third AI model possessed by the terminal. For example, the version of the model parameters of the third AI model that matches the fourth AI model should be version #2, while the version reported by the terminal is version #1, then it is considered that they do not match.

[0253] For example, if the first indication information includes information indicating the model structure of the third AI model possessed by the terminal, then the first network device can determine whether the third AI model possessed by the terminal matches the fourth AI model based on the model structure of the third AI model reported by the terminal. For example, if the model structure of the third AI model that matches the fourth AI model is model structure 1, but the model structure of the third AI model possessed by the terminal reported by the terminal is model structure 1, then the first network device can determine that the third AI model that matches the fourth AI model does not match.

[0254] In this application, after the first network device determines that the third AI model and the fourth AI model do not match, it can indicate to the terminal the reason for the mismatch between the third AI model and the fourth AI model through the second instruction information.

[0255] In the communication method provided by method 800, the terminal reports to the first network device a third AI model that it possesses, which together with the fourth AI model on the first network device side, implements the second AI function. Correspondingly, the first network device determines whether the third AI model matches the fourth AI model. If a mismatch is determined, the device will indicate to the terminal the mismatch or the reason for the mismatch. Thus, if the terminal receives the second indication information, it can re-determine a third AI model that matches the fourth AI model.

[0256] Understandably, in method 800, if the first network device determines that the third AI model matches the fourth AI model, then S820 may not be included. Optionally, if the third AI model matches the fourth AI model, the first network device may send an AI inference configuration to the terminal for the terminal to perform inference based on the third AI model. For example, under AI-based CSI feedback, the third AI model is used for CSI compression, and the first AI inference configuration may include the configuration of CSI resources and the configuration of CSI reporting.

[0257] The communication method provided in this application was described above using a terminal and a first network device as examples. It is understood that the terminal described above can also be replaced by a communication module within the terminal, or by components within the terminal responsible for communication functions (such as circuits, chips (e.g., modem chips, also known as baseband chips, or system-on-chip (SoC) chips or system-in-package (SIP) chips containing modem cores), chip systems, or processors, etc.), and this application does not limit this. Similarly, the first network device can also be replaced by a communication module within the first network device, or by components within the first network device responsible for communication functions (e.g., circuits, chips (e.g., modem chips, also known as baseband chips, or system-on-chip (SoC) chips or system-in-package (SIP) chips containing modem cores), chip systems, or processors, etc.), and this application does not limit this.

[0258] The communication method provided in this application has been described above with reference to the accompanying drawings. The communication device provided in this application will now be described.

[0259] Figure 9 is a structural schematic diagram of the communication device provided in an embodiment of this application. Specifically, as shown in Figure 9, the device 900 includes: a transceiver module 901 and a processing module 902.

[0260] In an embodiment of the first device, device 900 can be applied to a terminal.

[0261] Specifically, the transceiver module 901 is used to receive first information from the first network device. The first information is used to instruct the first target node, and the first target node is used to obtain second information. The second information is used to obtain the first AI model, and the first AI model is used by the terminal. The first AI model and the second AI model used by the first network device jointly complete the first AI function. The processing module 1102 is used to obtain the first AI model based on the first information.

[0262] In one possible implementation, the transceiver module 901 is further configured to: receive at least one of the following information from the first network device: information indicating that the first network device has a first AI function, information indicating a first mode, information indicating the model structure of the second AI model, or an identifier of the second information; the first mode is any one of the following: obtaining the first AI model based on the first dataset, or obtaining the first AI model based on the first model structure and / or the first model parameters.

[0263] In one possible implementation, the transceiver module 901 is further configured to: send third information to the first network device, the third information being used to request the first AI model.

[0264] In one possible implementation, the transceiver module 901 is specifically used to: send third information to the first network device if the terminal supports the first AI function and / or the terminal supports obtaining the first AI model based on the first method.

[0265] In one possible implementation, the transceiver module 901 is further configured to: receive fourth information, the fourth information being used to indicate the first network device set or the first cell set associated with the second information, the first network device set including at least one network device, the at least one network device including the first network device, the first cell set including at least one cell, and the at least one cell including the first cell currently accessed by the terminal.

[0266] In one possible implementation, the transceiver module 901 is further configured to: send fifth information to the first network device, the fifth information being used to indicate that the terminal has a third AI model, the third AI model and the fourth AI model used by the first network device jointly completing the second AI function.

[0267] In one possible implementation, the transceiver module 901 is further configured to: receive seventh information from the first network device, the seventh information being used to indicate whether the third AI model and the fourth AI model are mismatched or the reason for the mismatch; the reason for the mismatch includes one or more of the following: the identifier of the sixth information is mismatched, or the second method is mismatched.

[0268] In an embodiment of the second device, device 900 can be applied to the first network device.

[0269] Specifically, the transceiver module 901 is used to: send first information to the terminal, the first information is used to instruct the first target node, the first target node is used to obtain second information, wherein the second information is used to obtain the first AI model, the first AI model is used by the terminal, and the first AI model and the second AI model used by the first network device jointly complete the first AI function.

[0270] In one possible implementation, the transceiver module 901 is further configured to: send at least one of the following information to the terminal: information indicating that the first network device has a first AI function, information indicating a first mode, information indicating the model structure of the second AI model, or an identifier of the second information; the first mode is any one of the following: obtaining the first AI model based on the first dataset, or obtaining the first AI model based on the first model structure and / or the first model parameters.

[0271] In one possible implementation, the transceiver module 901 is further configured to: receive third information from the terminal, the third information being used to request the first AI model.

[0272] In one possible implementation, the transceiver module 901 is further configured to: send fourth information to the terminal, the fourth information being used to indicate the first network device set or the first cell set associated with the second information, the first network device set including at least one network device, the at least one network device including the first network device, the first cell set including at least one cell, and the at least one cell including the first cell currently accessed by the terminal.

[0273] In one possible implementation, the transceiver module 901 is further configured to: receive fifth information from the terminal, the fifth information indicating that the terminal has a third AI model, the third AI model and the fourth AI model for the first network device jointly perform the second AI function.

[0274] In one possible implementation, the transceiver module 901 is further configured to: send a seventh message to the terminal, the seventh message indicating the mismatch between the third AI model and the fourth AI model or the reason for the mismatch; the reason for the mismatch includes one or more of the following: the identifier of the sixth message is mismatched, or the second method is mismatched.

[0275] In an embodiment of the third device, device 900 can be applied to a terminal.

[0276] Specifically, the transceiver module 901 is used to: send a first request to the first network device, the first request being used to request model parameters of a first AI model whose model structure is the target model structure, the first AI model being used by the terminal, and the first AI model and a second AI model used by the first network device jointly performing the first AI function; and receive the first model parameters from the first network device.

[0277] In one possible implementation, the transceiver module 901 is further configured to: send first capability information of the terminal, the first capability information being used to indicate at least one of the following: the terminal supports a first AI function, whether the terminal supports the first network device providing model parameters of the first AI model to the terminal, at least one model structure supported by the terminal for the first AI model, and the maximum parameter threshold of the model parameters of the first AI model supported by the terminal.

[0278] In one possible implementation, the transceiver module 901 is further configured to: receive at least one of the following information from the first network device: information indicating that the first network device has a first AI function, information indicating the model structure associated with the model parameters of the first AI model provided by the first network device, an identifier indicating the model parameters of the first AI model provided by the first network device, and a set of network devices or a set of cells associated with the model parameters of the first AI model.

[0279] In one possible implementation, the transceiver module 901 is further configured to: send first indication information to the first network device, the first indication information being used to indicate a third AI model possessed by the terminal, the third AI model and a fourth AI model used by the first network device jointly completing the second AI function.

[0280] In one implementation where the terminal sends a first instruction message to the first network device, the transceiver module 901 is further configured to: receive a second instruction message sent by the first network device, the second instruction message being used to indicate the mismatch between the third AI model and the fourth AI model or the reason for the mismatch; the reason for the mismatch includes one or more of the following: the identifiers of the model parameters of the third AI model are mismatched, the model structure of the third AI model is mismatched.

[0281] In an embodiment of the fourth device, device 900 may be applied to the first network device.

[0282] Specifically, the transceiver module 901 is used to: receive a first request from the terminal, the first request being used to request model parameters of a first AI model whose model structure is the target model structure, the first AI model being used by the terminal, and the first AI model and a second AI model used by the first network device jointly completing the first AI function; the transceiver module 901 is also used to: send the model parameters of the first AI model whose model structure is the target model structure.

[0283] In one possible implementation, the transceiver module 901 is further configured to: receive first capability information from the terminal, the first capability information indicating at least one of the following: the terminal supports a first AI function, whether the terminal supports the first network device providing model parameters of the first AI model to the terminal, at least one model structure supported by the terminal for the first AI model, and the maximum parameter threshold of the model parameters of the first AI model supported by the terminal.

[0284] In one possible implementation, the transceiver module 901 is further configured to send at least one of the following information: information indicating that the first network device has a first AI function; information indicating the AI ​​model structure associated with the model parameters of the first AI model provided by the first network device; an identifier indicating the model parameters of the first AI model provided by the first network device; and a set of network devices associated with the model parameters of the first AI model.

[0285] In one possible implementation, the transceiver module 901 is further configured to: receive first indication information from the terminal, the first indication information being used to indicate a third AI model possessed by the terminal, the third AI model and a fourth AI model for the first network device jointly performing a second AI function.

[0286] In one implementation, the transceiver module 901 is further configured to send second indication information, which indicates a mismatch between the third AI model and the fourth AI model or the reason for the mismatch. The reason for the mismatch includes one or more of the following: mismatch in the identifiers of the model parameters of the third AI model, or mismatch in the model structure of the third AI model. Thus, if the terminal receives the second indication information, it can re-determine the third AI model that matches the fourth AI model, thereby improving the inference performance when performing the second AI function based on the third and fourth AI models. More detailed information regarding the first, second, third, fourth, fifth, sixth, and seventh information, the first indication information, and the second indication information can be found in the descriptions in the communication method embodiments provided above, and will not be repeated here.

[0287] Figure 10 is a structural schematic diagram of another communication device provided in an embodiment of this application. The device shown in Figure 10 can be used to perform the methods described in the foregoing embodiments.

[0288] As shown in Figure 10, the device 1000 of this embodiment includes a memory 1001 and a processor 1002. In one implementation, the device 1000 further includes a communication interface 1003 and a bus 1004. The memory 1001, processor 1002, and communication interface 1003 are interconnected via the bus 1004.

[0289] The memory 1001 may be a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1001 may store a program, and when the program stored in the memory 1001 is executed by the processor 1002, the processor 1002 performs the various steps of the communication method described above.

[0290] The processor 1002 may be a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, used to execute relevant programs to implement the various steps of the communication method described above.

[0291] The processor 1002 can also be an integrated circuit chip with signal processing capabilities. In implementation, each step of the method in Figures 5 to 8 of the embodiments of this application can be completed by the integrated logic circuitry in the processor 1002 or by software instructions.

[0292] The processor 1002 described above can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or a conventional processor, etc.

[0293] The steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory 1001. The processor 1002 reads the information in memory 1001 and, in conjunction with its hardware, completes the functions required by the units included in the device of this application. For example, it can execute the various steps / functions of the communication method described above.

[0294] The communication interface 1003 can use, but is not limited to, transceivers to enable communication between the device 1000 and other devices or communication networks.

[0295] Bus 1004 may include a pathway for transmitting information between various components of device 1000 (e.g., memory 1001, processor 1002, communication interface 1003).

[0296] It should be understood that the device 1000 shown in the embodiments of this application can be deployed in access network equipment or terminals.

[0297] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or 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., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be a usable medium accessible to a computer or a data storage device such as a server or data center containing one or more sets of usable media. The usable medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0298] It should be understood that the term "and / or" in this article 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. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0299] In this application, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0300] It should be understood that 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 limit the implementation process of the embodiments of this application.

[0301] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0302] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0303] 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.

[0304] 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.

[0305] 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.

[0306] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

Claims

1. A communication method, characterized in that, Applied to terminals, including: The system receives first information from a first network device, the first information being used to instruct a first target node, the first target node being used to obtain second information, wherein the second information is used to obtain a first artificial intelligence (AI) model, the first AI model being used by the terminal, and the first AI model and a second AI model used by the first network device jointly performing a first AI function. The first AI model is obtained based on the first information.

2. The method according to claim 1, characterized in that, The second information includes at least one of the following for obtaining the first AI model: a first dataset, a first model structure, and / or first model parameters.

3. The method according to claim 2, characterized in that, The method further includes: Receive at least one of the following information from the first network device: information indicating that the first network device has a first AI function, information indicating a first mode, information indicating the model structure of the second AI model, or an identifier of the second information; The first method is any one of the following: obtaining the first AI model based on the first dataset, or obtaining the first AI model based on the first model structure and / or the first model parameters.

4. The method according to any one of claims 1 to 3, characterized in that, The method further includes: A third message is sent to the first network device, the third message being used to request the first AI model.

5. The method according to claim 4, characterized in that, Sending the third information to the first network device includes: If the terminal supports the first AI function and / or the terminal supports obtaining the first AI model based on the first method, it sends third information to the first network device.

6. The method according to any one of claims 1 to 5, characterized in that, The method further includes: The system receives fourth information, which indicates a first set of network devices or a first set of cells associated with the second information. The first set of network devices includes at least one network device, and the first set of cells includes at least one cell, and the first cell includes the first cell currently accessed by the terminal.

7. The method according to any one of claims 1 to 6, characterized in that, The method further includes: A fifth message is sent to the first network device, the fifth message being used to indicate that the terminal has a third AI model, the third AI model and the fourth AI model used by the first network device together complete the second AI function.

8. The method according to claim 7, characterized in that, The fifth information includes at least one of the following: the identifier of the sixth information, information for indicating the second method, or information for indicating the second target node; The sixth piece of information includes at least one of the following for obtaining the third AI model: a second dataset, a second model structure, and / or second model parameters; The second method is any one of the following: the third AI model is obtained based on the second dataset, or the third AI model is obtained based on the second model structure and / or the second model parameters; The second target node is used to obtain the sixth information.

9. The method according to claim 7 or 8, characterized in that, The method further includes: Receive seventh information from the first network device, the seventh information being used to indicate the mismatch or reason for the mismatch between the third AI model and the fourth AI model; The reasons for the mismatch include one or more of the following: the identifier of the sixth information does not match, or the second method does not match.

10. A communication method, characterized in that, Applied to the first network device, including: Send first information to the terminal, the first information is used to instruct a first target node, the first target node is used to obtain second information, wherein the second information is used to obtain a first artificial intelligence (AI) model, the first AI model is used by the terminal, and the first AI model and a second AI model used by the first network device jointly complete the first AI function.

11. The method according to claim 10, characterized in that, The second information includes at least one of the following for obtaining the first AI model: a first dataset, a first model structure, and / or first model parameters.

12. The method according to claim 10 or 11, characterized in that, The method further includes: Send at least one of the following information to the terminal: information indicating that the first network device has a first AI function, information indicating a first method, information indicating the model structure of the second AI model, or an identifier of the second information; The first method is any one of the following: obtaining the first AI model based on the first dataset, or obtaining the first AI model based on the first model structure and / or the first model parameters.

13. The method according to any one of claims 10 to 12, characterized in that, The method further includes: Receive third information from the terminal, the third information being used to request the first AI model.

14. The method according to any one of claims 10 to 13, characterized in that, The method further includes: The terminal is sent a fourth piece of information, which is used to indicate the first set of network devices or the first set of cells associated with the second information. The first set of network devices includes at least one network device, and the first set of cells includes at least one cell, and the first cell includes the first cell currently accessed by the terminal.

15. The method according to any one of claims 10 to 14, characterized in that, The method further includes: The terminal receives a fifth piece of information, which indicates that the terminal has a third AI model, and the third AI model, together with a fourth AI model used in the first network device, performs a second AI function.

16. The method according to claim 15, characterized in that, The fifth information includes at least one of the following: the identifier of the sixth information, information for indicating the second method, or information for indicating the second target node; The sixth piece of information includes at least one of the following for obtaining the third AI model: a second dataset, a second model structure, and / or second model parameters; The second method is any one of the following: the third AI model is obtained based on the second dataset, or the third AI model is obtained based on the second model structure and / or the second model parameters; The second target node is used to obtain the sixth information.

17. The method according to claim 15 or 16, characterized in that, The method further includes: Send a seventh message to the terminal, the seventh message being used to indicate the mismatch between the third AI model and the fourth AI model or the reason for the mismatch; The reasons for the mismatch include one or more of the following: the identifier of the sixth information does not match, or the second method does not match.

18. A communication device, characterized in that, It includes modules for implementing the method as described in any one of claims 1 to 9; or, it includes modules for implementing the method as described in any one of claims 10 to 17.

19. A communication device, characterized in that, Includes a processor for causing the communication device to implement the method as described in any one of claims 1 to 9, or to cause the communication device to implement the method as described in any one of claims 10 to 17, by executing computer programs or instructions, and / or by logic circuitry.

20. The apparatus according to claim 19, characterized in that, It also includes a memory for storing computer programs or instructions, and / or, configuration files for the logic circuitry.

21. The apparatus according to claim 19 or 20, characterized in that, It also includes a communication interface for inputting and / or outputting signals.

22. A computer-readable storage medium storing a computer program or instructions thereon, characterized in that, When the computer program or instructions are executed, the method of any one of claims 1 to 9 is performed, or the method of any one of claims 10 to 17 is performed.

23. A computer program product, characterized in that, Includes a computer program, wherein when the computer program or instructions are run, the method of any one of claims 1 to 9 is performed, or the method of any one of claims 10 to 17 is performed.

Images