Communication method and related apparatus

Abstract

Provided in the present application are a communication method and a related apparatus. By means of providing an evaluation model, the performance of a task model can be predicted without the need for a label. Therefore, it is only necessary to input input data and / or output data of the task model into the evaluation model, such that the performance of the task model can be acquired from the evaluation model, thereby reducing the amount of data that needs to be acquired and lowering the overhead.

Description

Communication methods and related devices

[0001] This application claims priority to Chinese Patent Application No. 202411520078.6, filed on October 28, 2024, entitled "Communication Method and Related Apparatus", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of communication technology, and in particular to a communication method and related apparatus. Background Technology

[0003] With the development of communication technology, communication equipment in communication systems can now perform not only traditional communication services but also other new types of services, such as artificial intelligence (AI) services. Generally, a communication system capable of handling AI services can also be called an AI system.

[0004] When using AI technology, the first step is to train the AI ​​model using a training dataset. Only a well-trained AI model can be used for data inference. For the same inference function or the same task in a communication system, multiple AI models may exist, each with different model structures, parameters, computational overhead, and transmission overhead. During communication, for multiple AI models performing the same inference function, typically one model is activated based on its performance, while the others are not used for inference or to execute the task.

[0005] Evaluating the performance of multiple AI models requires collecting a large amount of data, and reducing data collection overhead has become an urgent problem to be solved. Summary of the Invention

[0006] Analysis revealed that evaluating the performance of a task model requires obtaining its input data, output data, and expected results or labels, which in turn determines the model's performance. This necessitates acquiring a substantial amount of data.

[0007] To address this, this application provides a communication method that, by providing an evaluation model, can predict the model's performance without requiring labels. Therefore, by simply inputting the task model's input data and / or output data into the evaluation model, the performance of the task model can be obtained from the evaluation model, reducing the amount of data required and lowering overhead. The solution is described below in conjunction with the method provided in the first aspect.

[0008] Firstly, this application provides a communication method. This method is executed by a first communication device. The first communication device can be an AI network element. An AI network element can also be called an AI node, AI device, AI entity, AI module, AI model, or AI unit, etc. The AI ​​network element is used to implement AI-related functions. The AI ​​network element can be a network element in a communication system or a module (referred to as an AI module) built into a network element in a communication system. Network elements in a communication system can be terminal devices, access network devices, core network devices, cloud servers, or network management (OAM), etc. The OAM can be a network management system for core network devices and / or for access network devices. An AI module can be a component within a network element, and this component can be a processor, chip (or chip system), logic module, or software product, etc.

[0009] In the method provided in the first aspect, the first communication device can process first data based on an evaluation model. The first data includes input data and / or output data of one or more task models, where the one or more task models correspond to a first task in the wireless communication system. The evaluation model is used to obtain prediction results of the performance data of the corresponding model based on the input data and / or output data of one or more models. Therefore, even if the first data does not include the expected results of the output data, after the evaluation model processes the first data, the first communication device can obtain prediction results of the performance data of one or more task models. This reduces the amount of data the first communication device needs to acquire and lowers data acquisition overhead in scenarios where the first communication device predicts the performance data of one or more task models.

[0010] Optionally, the evaluation model is an artificial intelligence (AI) model. Optionally, the task model is an artificial intelligence (AI) model. Optionally, the evaluation model may also be called a reward model.

[0011] Optionally, performance data from one or more task models may be used to indicate the performance of one or more task models. This application does not limit the manner in which performance data from one or more task models indicates the performance of one or more task models.

[0012] Optionally, in the first indication method, performance data from one or more task models are used to indicate the performance values ​​of one or more task models respectively. For example, performance data from the first task model and the second task model are used to indicate the performance values ​​of the first task model (denoted as the first value) and the performance values ​​of the second task model (denoted as the second value), respectively.

[0013] The performance values ​​of the two task models can be used to determine their performance relative. Optionally, when the first value is greater than the second value, the first task model performs better than the second task model. Alternatively, optionally, when the first value is less than the second value, the first task model performs better than the second task model.

[0014] The performance values ​​of the two task models can be used to determine the magnitude of the performance difference between them. Assume that the performance of the first and second task models is better than that of the third task model, and the performance value of the third task model is a third value. Optionally, when the difference between the first and third values ​​is greater than the difference between the second and third values, the performance difference between the first and third task models is greater than the performance difference between the second and third task models.

[0015] Optionally, based on the first data including input data and / or output data of multiple task models, in the second indication method, the performance data of the multiple task models indicates the performance superiority or inferiority relationship of different task models, and / or indicates the magnitude of the performance differences between different task models. For example, the performance data of multiple task models indicates the order in which the multiple task models are arranged according to their performance superiority or inferiority relationship; for example, it indicates that the sequence obtained by arranging the multiple task models in order of performance from best to worst is the first task model, the second task model, and the third task model.

[0016] Optionally, in the third indication method, the performance data of one or more task models respectively indicate the performance superiority and / or difference between the one or more task models and a reference model, wherein the reference model is a non-AI model, a default AI model, or a fallback AI model corresponding to the first task. For example, the performance data of one or more task models respectively indicate the performance superiority and inferiority between the first task model and the reference model, the performance superiority and inferiority between the second task model and the reference model, and the performance superiority and inferiority between the third task model and the reference model.

[0017] Optionally, based on the third indication method, the first data may also include the input data and / or output data of the reference model.

[0018] This application does not limit the use of the prediction results of the performance data acquired by the first communication device.

[0019] Optionally, in the first application, the prediction results of the performance data acquired by the first communication device are used to manage a set of task models corresponding to the first task. Managing the set of task models may include at least one of the following: model selection, model enabling, model updating, model switching, or model rollback of the task models in the set of task models. The relationship between one or more task models and the set of task models may be that one or more task models include all or some of the task models in the set of task models, or that the set of task models includes all or some of the task models in one or more task models.

[0020] Optionally, in the second use, the prediction results of the performance data acquired by the first communication device are used to train or validate the evaluation model.

[0021] Optionally, in this method, the first communication device may further acquire target results of the performance data of one or more of the task models, the target results being associated with the first data. Then, the first communication device can train the evaluation model and / or verify the accuracy of the evaluation model based on the predicted results of the performance data and the target results.

[0022] Here, the target result of the performance data can be understood as the actual or real result of the performance data. The correlation between the target result and the first data can be understood as the target result being obtained by evaluating the performance of one or more task models based on the first data.

[0023] Optionally, when the difference between the predicted result and the target result is less than a threshold, the evaluation model is used to evaluate the performance of one or more task models, and the performance data of one or more task models output by the evaluation model can be used to manage one or more task models. Optionally, when the difference between the predicted result and the target result is greater than a threshold, the evaluation model is not used to evaluate the performance of one or more task models, and the performance data of one or more task models output by the evaluation model is not used to manage one or more task models.

[0024] Optionally, the trained evaluation model is used to evaluate the performance of one or more task models, and the performance data of one or more task models output by the evaluation model can be used to manage one or more task models.

[0025] In managing a set of task models, the best-performing task model is typically enabled, meaning it's used to execute the first task. However, analysis reveals that when the performance of the task model used to execute the first task (called the target model) is no longer superior to other task models, switching to that model can lead to a large number of model switching operations and significant model management overhead.

[0026] To address this, this application provides a communication method that allows the target model's performance to be lower than other models within a certain threshold. This not only ensures the performance of the first task performed using the target model but also reduces the number of model switching operations and lowers model management overhead. The solution is described below in conjunction with the method provided in the second aspect.

[0027] Secondly, this application provides a communication method. This method is executed by a first communication device. The first communication device can be an AI network element. The meaning of an AI network element can be understood by referring to the relevant content introduced above, and will not be repeated here.

[0028] In the method provided in the second aspect, a first communication device determines a performance difference between a first model and a second model among a plurality of models, wherein the plurality of models respectively correspond to a first task in the wireless communication system. Furthermore, the first model is used to perform the first task, while the second model is not used to perform the first task. In other words, the target model corresponding to the first task is the first model. Then, based on the difference satisfying a first condition, the first communication device switches among the plurality of models to the model used to perform the first task. The first condition includes: the difference indicating that the performance of the second model is superior to that of the first model, and the absolute value of the difference exceeding a first threshold. Compared to the prior art, where the first communication device switches the target model as long as the performance of the second model is superior to the first model, the method provided in the second aspect is advantageous in continuing to use the first model as the target model when the absolute value of the difference does not exceed the first threshold, without performing model switching, thus reducing the number of model switching operations and lowering the overhead of model management.

[0029] Switching the model used to perform the first task can be understood as the first communication device replacing the model used to perform the first task from the first model to another model corresponding to the first task. This application does not limit the specific manner in which the first communication device switches the model used to perform the first task.

[0030] In the first switching method, the first communication device switches the model used to perform the first task by updating first management information. The first management information indicates the model used to perform the first task, and the first communication device enables the model indicated by the first management information to perform the first task. For example, based on the first management information indicating that the model used to perform the first task is a first model, the first communication device enables the first model, causing the first model to perform the first task.

[0031] In the second switching method, the first communication device switches the model used to perform the first task by updating second management information, whereby the second management information indicates the state of multiple models. Alternatively, the first communication device switches the model used to perform the first task by changing the state of at least one of the multiple models. This at least one model includes the first model.

[0032] The first communication device can store the states of multiple models, where each model's state is one of those states. Assume the multiple states include a first state and a second state, where the model in the first state is used to perform the first task, and the model in the second state is not used to perform the first task.

[0033] The states of multiple models indicated by the second management information may include the state of a first model and the state of a third model, where the third model corresponds to the first task. Switching the model used to execute the first task by the first communication device can be understood as the first communication device switching the state of the first model from a first state among multiple states to another state among the multiple states, and switching the state of the third model from a second state among the multiple states back to the first state. Accordingly, after the first communication device switches the model used to execute the first task, the third model is used to execute the first task.

[0034] The third model can be the second model. Both the first and second models can be AI models corresponding to the first task. Alternatively, one of the first and second models can be an AI model, and the other model can be a non-AI model (such as the reference model introduced above).

[0035] Alternatively, the third model can be a model other than the first and second models, and this other model corresponds to the first task. The first, second, and third models are all AI models, or the first, second, and third models include both AI models and non-AI models (e.g., a reference model).

[0036] Optionally, the multiple states correspond to different multiple model management information, and the first communication device can manage the model according to the model's state. The model management information may indicate at least one of the following: first indication information, second indication information, conditions for switching states among the multiple states, or configuration information for measuring model performance. Specifically, the first indication information may indicate whether the first task is executed or not; the second indication information may indicate the configuration for executing the first task; and the configuration information for measuring model performance may indicate one or more parameters for measuring model performance, which may include the measurement period.

[0037] Optionally, based on the performance difference between the first model and the second model satisfying a first condition, the first communication device switches the state of the first model from the first state to the second state. Subsequently, the first communication device may also switch the state of the first model from the second state to the first state, thereby enabling the first model, when the performance of the first model satisfies a certain condition (referred to as the second condition).

[0038] Optionally, the multiple states may include a third state in addition to the first and second states. The model in the third state is not used to perform the first task and does not support direct switching to the first state. Optionally, the measurement period indicated by the model management information corresponding to the third state is longer than the measurement period indicated by the model management information corresponding to the first state, and also longer than the measurement period indicated by the model management information corresponding to the second state.

[0039] Optionally, based on the performance difference between the first model and the second model satisfying a first condition, and the absolute value of the difference exceeding a second threshold, wherein the second threshold is greater than the first threshold, the first communication device can switch the state of the first model from the first state to the third state. By switching the first model, whose absolute value of the difference exceeds the second threshold, to the third state, it is advantageous for the performance of the model in the second state to be superior to that of the model in the third state. This allows for the use of a longer measurement cycle to detect the model in the third state, thus reducing the performance overhead of the measurement model.

[0040] The methods provided in the first aspect and the methods provided in the second aspect can be combined with each other. For example, referring to the method provided in the first aspect, the first communication device can process the first data based on the evaluation model and obtain the prediction results of the performance data of one or more task models. Then, referring to the method provided in the second aspect, it can determine the first difference in performance corresponding to the first model and the second model based on the prediction results. After that, the first communication device can switch the model used to perform the first task based on the first difference satisfying a first condition.

[0041] Optionally, the models mentioned in the method provided in the second aspect can be understood by referring to the task models in the method provided in the first aspect, and the multiple models mentioned in the method provided in the second aspect can be understood by referring to the set of task models in the method provided in the first aspect.

[0042] Optionally, in the method provided in the first aspect, one or more task models may include a first model and / or a second model.

[0043] This application does not limit the specific method by which the first communication device determines the first difference based on the prediction results of performance data from one or more task models.

[0044] Based on the first indication method described above, optionally, the prediction result of the performance data of one or more task models acquired by the first communication device may include the prediction result of the performance value of the first model (referred to as the first value) and the prediction result of the performance value of the second model (referred to as the second value). The first communication device determines the first difference based on the prediction result of the first value and the prediction result of the second value. Alternatively, optionally, the prediction result of the performance data of one or more task models acquired by the first communication device may include the prediction result of the performance value of one of the first and second models. The first communication device may acquire the performance value of the other model through a method other than the method provided in the first aspect.

[0045] Based on the second indication method described above, optionally, the prediction results of the performance data of one or more of the task models acquired by the first communication device indicate the performance difference between the first model and the second model.

[0046] Based on the third indication method described above, optionally, the prediction results of the performance data of one or more of the task models acquired by the first communication device indicate the performance differences between the first model and the reference model, and the performance differences between the second model and the reference model, respectively. Then, the first communication device determines the first difference based on the performance differences between the first model and the reference model, and the performance differences between the second model and the reference model. Alternatively, optionally, one of the first model and the second model is the task model, and the other model is the reference model. Taking the first model as the task model and the second model as the reference model as an example, the prediction results of the performance data of one or more of the task models acquired by the first communication device can indicate the first difference.

[0047] In this application, "difference" can be understood as a difference or ratio.

[0048] In this application, the performance of the model may include the accuracy of the model and / or the performance of the first task performed using the model. The model corresponds to the first task; for example, the model is a task model, a reference model, or any of a plurality of models.

[0049] In this application, the type of the first task is not limited. For example, the first task includes at least one of the following: channel state information feedback task, channel estimation task, channel prediction task, beam management task, positioning task, sensing task, network energy saving task, load balancing task, or mobility management task.

[0050] Optionally, the task model is deployed on the first communication device.

[0051] Alternatively, the task model can be deployed on a communication device other than the first communication device. Accordingly, the first communication device obtains the first data from the other communication device and then processes the first data based on the evaluation model.

[0052] The preceding text introduced a method by which the first communication device uses an evaluation model to obtain prediction results of performance data for one or more task models (referred to as the task model performance prediction method), a method by which the first communication device manages a set of task models (referred to as the task model management method), and a method by which the first communication device trains or validates the evaluation model (referred to as the evaluation model management method). In some examples, at least two of the task model performance prediction method, task model management method, or evaluation model management method can be executed by different communication devices. For example, the communication device executing the task model performance prediction method may be different from the communication device executing the task model management method, and / or, the communication device executing the task model management method may be different from the communication device executing the evaluation model management method.

[0053] This application does not limit the task model to be deployed on the communication device used to execute the management method of the task model. Given that the communication device used to deploy the task model and the communication device used to execute the management method of the task model are different, or that they are built into different network elements, the execution of the task model management method by the communication device can be understood as sending management commands for the task model to the communication device that deploys the task model, and the communication device that deploys the task model managing the set of task models according to the management commands.

[0054] In one possible example, after the first communication device obtains the prediction result of the performance data, this application does not limit the first communication device to managing multiple models based on the prediction result of the performance data. Optionally, after obtaining the prediction result of the performance data, the first communication device can send the prediction result to the second communication device, and the second communication device manages multiple models based on the prediction result. The method by which the second communication device manages the task models can be understood by referring to the method by which the first communication device manages the task models described in this application.

[0055] The second communication device can be an AI network element. For an explanation of AI network elements, please refer to the previous text; further details will not be provided here.

[0056] For example, the first communication device can be a terminal device or an AI module built into a terminal device, and the second communication device can be a network device or an AI module built into a network device. Alternatively, the second communication device can be a terminal device or an AI module built into a terminal device, and the first communication device can be a network device or an AI module built into a network device. Alternatively, the first and second communication devices can be two different network devices or AI modules built into two different network devices; for example, the two different network devices can be different network elements in a centralized unit (CU), distributed unit (DU), radio unit (RU), or integrated access and backhaul (IAB) node. Alternatively, the first and second communication devices can be two different terminal devices or AI modules built into two different terminal devices; these two different terminal devices can be devices in a device-to-device (D2D) scenario.

[0057] After the first communication device obtains the prediction result of the performance data, this application does not limit the first communication device to training or verifying the evaluation model based on the prediction result of the performance data. Optionally, after obtaining the prediction result of the performance data, the first communication device may send the prediction result to the third communication device, and the third communication device may train or verify the evaluation model based on the prediction result. The method by which the third communication device executes the management method of the evaluation model can be understood by referring to the method by which the first communication device executes the management method of the evaluation model described in this application.

[0058] The third communication device can be an AI network element. For an explanation of AI network elements, please refer to the previous text; further details will not be provided here.

[0059] For example, the first communication device may be a terminal device or an AI module built into a terminal device, and the third communication device may be a network device or an AI module built into a network device. Alternatively, the third communication device may be a terminal device or an AI module built into a terminal device, and the first communication device may be a network device or an AI module built into a network device.

[0060] The second communication device and the third communication device may be the same or different communication devices.

[0061] A third aspect of this application provides a communication device comprising a plurality of interacting functional modules, wherein, exemplarily, the plurality of functional modules include an acquisition unit and a processing unit.

[0062] In some examples, the communication device is used to implement the method described in the first aspect or any possible implementation of the first aspect, or the second aspect or any possible implementation of the second aspect, and to achieve the corresponding technical effects. Accordingly, the communication device can be a first communication device. For details, please refer to the foregoing corresponding methods, which will not be repeated here. Exemplarily, the acquisition unit is used to execute the steps corresponding to the "acquisition" operation in the above method, and the processing unit is used to execute the steps corresponding to other operations besides the "acquisition" operation in the above method.

[0063] A fourth aspect of this application provides a communication device comprising at least one processor coupled to a memory for storing programs or instructions. The at least one processor executes the program or instructions to cause the device to implement the methods described in the first aspect or any possible implementation thereof, or the second aspect or any possible implementation thereof, and to achieve the corresponding technical effects. Accordingly, the communication device may be a first communication device.

[0064] Optionally, the communication device may also include the memory.

[0065] A fifth aspect of this application provides a communication device including at least one logic circuit and an input / output interface. The logic circuit is used to implement the method described in the first aspect or any possible implementation thereof, or the second aspect or any possible implementation thereof, and to achieve the corresponding technical effects. Accordingly, the communication device can be a first communication device.

[0066] A sixth aspect of this application provides a chip or chip system including at least one processor. For example, the chip may be a SoC chip (such as a SoC chip containing a modem core), a SIP chip, or a communication module. In one possible design, the chip or chip system may further include a memory for storing program instructions and data necessary for the communication device. The chip system may be composed of chips or may include chips and other discrete devices. Optionally, the chip system may also include an interface circuit that provides program instructions and / or data to the at least one processor. The chip or chip system is used to implement the method described in the first aspect or any possible implementation of the first aspect, or the second aspect or any possible implementation of the second aspect, and to achieve the corresponding technical effects. Accordingly, the chip or chip system may be a first communication device.

[0067] A seventh aspect of this application provides a communication system. Optionally, the communication system includes the first communication device described above, and may also include the second and / or third communication device described above.

[0068] An eighth aspect of this application provides a computer-readable storage medium for storing one or more computer-executable instructions, which, when executed by a processor, implement the method described in the first aspect or any possible implementation thereof, or the second aspect or any possible implementation thereof.

[0069] The ninth aspect of this application provides a computer program product (or computer program) that, when executed by a processor, implements the method described in the first aspect or any possible implementation of the first aspect, or the second aspect or any possible implementation of the second aspect.

[0070] The technical effects of any of the design methods in aspects three through nine can be found in the technical effects of the corresponding design methods in aspects one through four above, and will not be repeated here. Attached Figure Description

[0071] Figure 1-1 is a schematic diagram of a communication system in this application;

[0072] Figure 1-2 is a schematic diagram of the application framework involving the RIC module under the O-RAN architecture;

[0073] Figure 1-3 shows an example diagram of an O-RAN system;

[0074] Figure 1-4 schematically illustrates another architecture of the communication system to which this application applies;

[0075] Figure 2 schematically illustrates the framework of the AI / ML system;

[0076] Figure 3-1 schematically illustrates the methods performed by AI-related functions;

[0077] Figure 3-2 schematically illustrates the method by which the management module in Figure 3-1 determines the performance of the task model;

[0078] Figure 4-1 schematically illustrates the process by which the management module acquires performance data for two task models;

[0079] Figure 4-2 schematically illustrates another possible process by which the management module obtains performance data for two task models;

[0080] Figure 5 schematically illustrates the process of the management module performing model management;

[0081] Figure 6 schematically illustrates the multiple states of the model and the transition conditions between different states;

[0082] Figure 7 schematically illustrates the timing process of AI / ML system performance monitoring and model management;

[0083] Figure 8 schematically illustrates a possible flow of communication methods in an AI / ML system;

[0084] Figures 9-13 schematically illustrate the possible flow of communication methods in AI / ML systems;

[0085] Figure 14 shows a simplified structural diagram of a terminal device;

[0086] Figure 15 shows a simplified schematic diagram of a network device. Detailed Implementation

[0087] To facilitate understanding of the embodiments of this application, some of the terms used in this application will be explained below.

[0088] (1) In the embodiments of this application, "send" and "receive" indicate the direction of signal transmission. For example, "send information to XX" can be understood as the destination of the information being XX, which may include sending directly through the air interface or sending indirectly through the air interface by other units or modules. "Receive information from YY" or "receive information sent by YY" may include receiving directly from YY through the air interface or receiving indirectly from YY through the air interface by other units or modules. The information may be generated by YY, or the information may be generated by other devices besides YY, and YY is responsible for sending the information. "Send" can also be understood as the "output" of the chip interface, and "receive" can also be understood as the "input" of the chip interface.

[0089] In other words, sending and receiving can occur between devices, such as between network devices and terminal devices, or within a device, such as between components, modules, chips, software modules, or hardware modules within the device via buses, wiring, or interfaces.

[0090] It is understandable that information may undergo necessary processing, such as encoding and modulation, between the source and destination, but the destination can understand the valid information from the source. Similar statements in this application can be interpreted in a similar way and will not be elaborated further.

[0091] (2) In the embodiments of this application, "instruction" may include direct instruction and indirect instruction, as well as explicit instruction and implicit instruction. The information indicated by a certain piece of information (hereinafter referred to as instruction information) is called the information to be instructed. In the specific implementation process, there are many ways to indicate the information to be instructed, such as, but not limited to, directly indicating the information to be instructed, such as the information to be instructed itself or its index. It can also indirectly indicate the information to be instructed by indicating other information, where there is an association between the other information and the information to be instructed; or it can only indicate a part of the information to be instructed, while the other parts of the information to be instructed are known or pre-agreed upon. For example, the instruction of specific information can be achieved by using a pre-agreed (e.g., protocol predefined) arrangement order of various information, thereby reducing the instruction overhead to a certain extent. This application does not limit the specific method of instruction. It is understood that for the sender of the instruction information, the instruction information can be used to indicate the information to be instructed, and for the receiver of the instruction information, the instruction information can be used to determine the information to be instructed.

[0092] (3) Access network equipment is a device deployed in a wireless access network to provide wireless communication functions for terminal devices. Access network equipment can connect terminal devices to the RAN node of the wireless network, and can also be called wireless access equipment, access network equipment, RAN entity, access node, network node, or communication device, etc.

[0093] Specifically, the access network equipment can be access network equipment for cellular systems related to the 3rd Generation Partnership Project (3GPP). For example, 4G or 5G communication systems. The access network equipment can also be access network equipment in open RAN (O-RAN or ORAN) or cloud radio access network (CRAN). Alternatively, the access network equipment can also be access network equipment in a communication system resulting from the integration of two or more of the above communication systems.

[0094] Access network equipment includes, but is not limited to: next generation node B (gNB), evolved node B (eNB), RNC, node B (NB), base station controller (BSC), base transceiver station (BTS), home base station (e.g., home evolved Node B, or home Node B (HNB), baseband unit (BBU), access point (AP) in wireless fidelity (WIFI) system, macro base station, micro base station, wireless relay node, donor node, radio controller in CRAN scenario, wireless backhaul node, transmission point (TP) or transmission and reception point (TRP), etc., and can also be access network equipment in 5G mobile communication system. For example, a next-generation NodeB (gNB), TRP, or TP in an NR system; or one or a group of antenna panels (including multiple antenna panels) in a base station in a 5G mobile communication system; or, access network equipment can also be network nodes constituting a gNB or transmission point. Examples include centralized units (CU), distributed units (DU), centralized unit control planes (CU-CP), centralized unit user planes (CU-UP), or radio units (RU), etc. CUs and DUs can be separate or included in the same network element, such as a BBU. RUs can be included in radio equipment or radio units. For example, in remote radio units (RRU), active antenna units (AAU), or remote radio heads (RRH). Alternatively, access network equipment can also be servers, wearable devices, vehicles, or in-vehicle equipment, etc. For example, the access network equipment in V2X technology can be a roadside unit (RSU). It should be understood that the aforementioned TRP can be a device or module located on the network side of the communication system and possessing corresponding communication functions. The TRP typically contains communication modules, circuits, or chips that perform the corresponding communication functions.The TRP can also be configured with program instructions for the corresponding communication functions.

[0095] It should be noted that CU (or CU-CP and CU-UP), DU, or RU may have different names in different systems, but those skilled in the art will understand their meaning. For example, in an open radio access network (ORAN) system, CU can also be called an open centralized unit (O-CU) or an open CU, DU can also be called an open distributed unit (O-DU), CU-CP can also be called an open centralized unit control plane (O-CU-CP), CU-UP can also be called an open centralized unit user plane (O-CU-UP), and RU can also be called an open radio unit (O-RU). This application does not impose any specific limitations on these details. Any of the units CU, 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.

[0096] Optionally, for network elements in the ORAN system, each network element can implement the protocol layer functions shown in Table 1 below.

[0097] Table 1

[0098] It should be noted that in the ORAN system, the access network equipment in this application can be one or more network elements listed in Table 1 above.

[0099] The architecture of the CU and DU of the access network equipment is described below. An access network equipment includes at least one CU and at least one DU. Optionally, the access network equipment may also include at least one RU.

[0100] The following example uses an access network device consisting of one CU and one DU. The CU has some core network functions and can include CU-CP and CU-UP. The CU and DU can be configured according to the protocol layer functions of the wireless network they implement. For example, the CU may be configured to implement the Packet Data Convergence Protocol (PDCP) layer and above (e.g., RRC and / or SDAP layers). The DU may be configured to implement protocol layers below the PDCP layer (e.g., RLC, MAC, and / or physical (PHY) layers). Alternatively, the CU may be configured to implement protocol layers above the PDCP layer (e.g., RRC and / or SDAP layers), and the DU may be configured to implement protocol layers below the PDCP layer (e.g., RLC, MAC, and / or PHY layers).

[0101] When a CU includes CU-CP and CU-UP, CU-CP is used to implement the control plane functions of the CU, and CU-UP is used to implement the user plane functions of the CU. For example, when a CU is configured to implement the functions of the PDCP layer, RRC layer, and SDAP layer, CU-CP is used to implement the RRC layer functions and the control plane functions of the PDCP layer, and CU-UP is used to implement the SDAP layer functions and the user plane functions of the PDCP layer.

[0102] The CU-CP can interact with network elements in the core network used to implement control plane functions. These network elements can be access and mobility function (AMF) network elements, such as the AMF in a 5G system. The AMF is responsible for mobility management in the mobile network, such as terminal location updates, terminal registration with the network, and terminal handover.

[0103] CU-UP can interact with network elements in the core network used to implement user plane functions. These network elements, such as the user plane function (UPF) in a 5G system, are responsible for forwarding and receiving data in the terminal.

[0104] 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. For example, based on 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.

[0105] 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.

[0106] Network equipment may also include core network equipment, such as the Mobility Management Entity (MME), Home Subscriber Server (HSS), Serving Gateway (S-GW), Policy and Charging Rules Function (PCRF), and Public Data Network Gateway (PDN Gateway, P-GW) in 4th generation (4G) networks; and access and mobility management function (AMF), user plane function (UPF), or session management function (SMF) in 5G networks. Furthermore, this core network equipment may also include other core network equipment in 5G networks and next-generation networks of 5G networks.

[0107] (4) Terminal equipment, also known as user equipment (UE), mobile station (MS), mobile terminal (MT), fixed wireless access (FWA), customer premise equipment (CPE), or 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 grid, smart furniture, smart office, smart wearables, smart transportation, smart cities, etc.

[0108] Terminal devices can include mobile phones, tablets, laptops, PDAs, mobile internet devices (MIDs), wearable devices, virtual reality (VR) devices, augmented reality (AR) devices, wireless terminals in industrial control, wireless terminals in self-driving (e.g., drones, vehicles), wireless terminals in remote medical surgery, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, IoT terminals or wireless terminals in smart homes, repeaters, or integrated access and backhaul (IAB) nodes, etc. For example, wireless terminals in self-driving can be drones, helicopters, or airplanes. For example, wireless terminals in vehicle-to-everything (V2X) can be in-vehicle equipment, vehicle equipment, in-vehicle modules, vehicles, or ships, etc. Wireless terminals in industrial control can be cameras, robots, or robotic arms, etc. Wireless terminals in a smart home can be televisions, air conditioners, robot vacuums, speakers, or set-top boxes, etc. Terminal devices can also be devices or modules that connect to the communication systems shown above and have corresponding communication functions. Terminal devices typically contain communication modules, circuits, or chips that perform the corresponding communication functions, and they also contain program instructions for performing these functions. The embodiments of this application do not limit the specific technology or device form used in the terminal.

[0109] (5) Artificial Intelligence (AI) Model: Used to implement corresponding AI functions. The AI ​​model can be configured based on one or more of the following parameters: structural parameters (e.g., at least one of the following: number of neural network layers, neural network width, inter-layer connections, neuron weights, neuron activation function, or bias in the activation function), input parameters (e.g., type and / or dimension of input parameters), or output parameters (e.g., type and / or dimension of output parameters). The bias in the activation function can also be called the bias of the neural network.

[0110] A model can infer an output, which includes one or more parameters. The learning, training, or inference processes of different models can be deployed on different nodes or devices, or they can be deployed on the same node or device.

[0111] The neural network of an AI model can be a neural network composed of an embedding layer and a multilayer perception (MLP), or it can be a deep neural network (DNN), a convolutional neural network (CNN), a recurrent neural network (RNN), a residual network, or other neural networks.

[0112] The technical solution of this application can be applied to various communication systems. For example, 5th generation (5G) systems, new radio (NR) systems, long term evolution (LTE) systems, LTE frequency division duplex (FDD) systems, LTE time division duplex (TDD) systems, universal mobile telecommunication system (UMTS), mobile communication systems after 5G networks (e.g., 6G mobile communication systems), vehicle to everything (V2X) communication systems, device to device (D2D) communication systems, Internet of Things (IoT) communication systems, industrial Internet (IIoT) communication systems, satellite communication systems, short-range communication systems, or narrow band Internet of Things (NB-IoT) systems.

[0113] The communication system to which this application applies may include a radio access network (RAN), which includes at least one access network device and at least one terminal device. This application does not limit the number or type of access network devices and terminal devices in the communication system.

[0114] Figure 1-1 is a schematic diagram of a communication system according to this application. In Figure 1-1, the communication system may include RAN 100, which includes access network equipment 101 and terminal equipment 102 to 107. In the example shown in Figure 1-1, terminal equipment 102 is a vehicle, terminal equipment 103 is a smart air conditioner, terminal equipment 104 is a smart fuel dispenser, terminal equipment 105 is a mobile phone, terminal equipment 106 is a smart teacup, and terminal equipment 107 is a printer.

[0115] As shown in Figure 1-1, the communication system may also include a CN 200. The access network device 101 in the RAN 100 can be connected to the CN 200 wirelessly or via a wired connection. The core network 200 may include one or more core network devices. The core network device in the CN 200 and the access network device 101 in the RAN 100 may be independent and different physical devices, or they may be the same physical device that integrates the logical functions of the core network device and the logical functions of the access network device.

[0116] As shown in Figure 1-1, the communication system may further include an Internet 300, which can be connected to the CN 200 via wired or wireless means, and / or, the Internet 300 can be connected to the access network device 101 in the RAN 100 via wired or wireless means. Optionally, the service requester can connect to the CN 200 and / or the RAN 100 via the Internet 300.

[0117] Figure 1-2 is a schematic diagram of the application framework involving the RIC module under the O-RAN architecture. The communication system includes a RAN intelligent controller (RIC), which can be at least one of real-time (RT RIC), near-real-time (near-RT RIC), or non-real-time (non-RT RIC). Figure 1-2 schematically illustrates near-real-time RIC and non-real-time RIC.

[0118] As an example, the near real-time RIC in Figure 1-2 is used for model training and inference. For instance, it can be used to train an AI model, which is then used for inference. The near real-time RIC can obtain network-side and / or terminal-side information from RAN nodes (e.g., CU, CU-CP, CU-UP, DU, and / or RU) and / or terminal devices. This information can be used as training data or inference data. Optionally, the near real-time RIC can deliver the inference results to the RAN nodes and / or terminal devices. Optionally, inference results can be exchanged between CU and DU, and / or between DU and RU. For example, the near real-time RIC delivers the inference results to the DU, and the DU sends them to the RU.

[0119] As another example, the non-real-time RIC in Figure 1-2 is used for model training and inference. For example, it can be used to train an AI model and then use that AI model for inference. The non-real-time RIC can obtain network-side and / or terminal-side information from RAN nodes (e.g., CU, CU-CP, CU-UP, DU, and / or RU) and / or terminal devices. This information can be used as training data or inference data, and the inference results can be delivered to the RAN nodes and / or terminal devices. Optionally, inference results can be exchanged between CU and DU, and / or between DU and RU; for example, the non-real-time RIC delivers the inference results to the DU, which then forwards them to the RU.

[0120] As another example, the near real-time RIC and non-real-time RIC in Figure 1-2 can also be configured as separate network elements. Optionally, the near real-time RIC and non-real-time RIC can also be part of other devices. For example, the near real-time RIC can be set in the RAN node (e.g., in the CU or DU), while the non-real-time RIC can be set in the operation, administration and maintenance (OAM) system, cloud server, core network device, or other network device.

[0121] Figure 1-3 illustrates an example of an O-RAN system. As shown in Figure 1-3, access network devices communicate with the core network via a backhaul link and with terminal devices via an air interface. An O-RAN system may include components other than those shown in Figure 1-3. Figure 1-3 uses the example of CU and DU included in the BBU; in some examples, CU and DU can be configured separately.

[0122] As shown in Figure 1-3, the CU and DU can communicate via at least one midhaul link. The CU can connect to network nodes such as the core network through interfaces, for example, the E2 interface. Optionally, the CU can have some core network functions. The CU (e.g., the PDCP layer and / or higher) connects to the DU (e.g., the radio link control (RLC) layer and lower layers of the DU) through interfaces, for example, the F1 interface. Optionally, the F1 interface 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 signaling procedures of F1 in some examples. The F1 interface supports control plane F1-C and user plane F1-U.

[0123] Figure 1-4 schematically illustrates another architecture of the communication system to which this application applies. As shown in Figure 1-4, the communication system includes a base station and two terminal devices. The base station can provide services to the terminals, and correspondingly, communication-related services and / or AI-related services can be performed between the base station and the terminal devices. The communication system shown in Figure 1-4 can support point-to-point communication; for example, communication-related services and / or AI-related services can be performed between different terminal devices.

[0124] The base station in the communication system shown in Figure 1-4 can be replaced with other types of network equipment. The communication system shown in Figure 1-4 can include fewer or more terminal devices, and it can also include a larger number of network devices. The communication system shown in Figure 1-4 can include one or more cells, and each cell can include one or more network devices.

[0125] 3GPP 5G Release 18 research introduces AI / machine learning (ML) technology into the radio air interface for wireless channel information compression and reconstruction, beam management, and positioning enhancement. The communication system provided in this application can incorporate AI network elements to implement some or all AI-related services, operations, or functions. AI network elements can also be referred to as AI nodes, AI devices, AI entities, AI modules, AI models, or AI units, etc. These AI network elements can be built into network elements within the communication system. For example, an AI network element can be an AI module built into access network equipment, core network equipment, RT RIC, Non-RT RIC, cloud server, CU, DU, terminal equipment, or operation, administration, and maintenance (OAM) network management system. The OAM can be the network management system for core network equipment and / or access network equipment. Alternatively, the AI ​​network element can also be a network element independently configured within the communication system.

[0126] Figure 2 schematically illustrates the framework of an AI / ML system. The AI / ML system shown in Figure 2 modularly demonstrates the AI-related functions of the AI ​​network elements. As shown in Figure 2, the AI-related functions may include a data collection module, a model training module, a management module, an inference module, and a model storage module. In the communication system to which this application applies, the AI-related functions of the AI ​​network elements may include at least one of the functions shown in Figure 2. It should be noted that the different functions shown in Figure 2 can be executed by the same or different AI network elements, and a single function shown in Figure 2 can be executed independently by a single AI network element or jointly by multiple AI network elements. This application does not limit the implementation method of the modules shown in Figure 2; for example, the modules shown in Figure 2 can be implemented by software, hardware, or a combination of both.

[0127] The inference module can be used to process or infer inference data using AI models. Hereinafter, the AI ​​model involved in the inference module will be referred to as the task model. The inference output data obtained from the task model can be used to execute tasks in the communication system. Inference data can be understood as the input data of the task model, and inference output data can be understood as the output data of the task model. For example, the tasks executed by the AI ​​network element using the task model in the communication system may include, but are not limited to, at least one of the following tasks: channel state information (CSI) feedback enhancement, channel estimation, channel prediction, beam management enhancement, sensing, positioning accuracy enhancement, network energy saving, load balancing, and mobility optimization. Some of these tasks will be explained below.

[0128] 1. Enhanced CSI feedback

[0129] Channel State Information (CSI) is the channel attribute of a communication link, representing channel state information reported by the terminal device to the network device. By reporting this CSI information, the terminal device can select appropriate precoding matrices or codebooks, modulation and coding schemes (MCS), etc., to adapt to changing wireless channels. For example, the terminal device performs channel estimation based on the received channel state information reference signal (CSI-RS) and then feeds back the CSI information to the network device. This information serves as input to the network device's model, enabling AI model training. By applying AI to CSI feedback enhancement, overhead can be reduced, accuracy improved, and prediction capabilities enhanced.

[0130] CSI feedback enhancement may include at least one sub-function, such as: CSI compressed reconstruction, CSI prediction, and CSI-RS overhead reduction. CSI compressed reconstruction may further include CSI compressed reconstruction in at least one domain: spatial, temporal, and frequency.

[0131] 2. Enhanced Beam Management

[0132] Enhanced beam management primarily aims to discover the strongest transmit / receive beams or beam pairs. AI-based beam prediction can improve accuracy. Based on AI training and inference, this can include AI beam prediction on the network side and AI beam prediction on the terminal device side. Taking terminal device-side AI beam prediction as an example, the pre-trained AI model on the terminal device can be provided by the network side or pre-stored on the terminal device. During the training phase, the network device scans all possible beams, and then the terminal device reports the reference signal received power (RSRP) of the beams or the index of the strongest beam with a subset of RSRPs. When the model training is complete, the network device only needs to scan a small subset of beams, and then the terminal device feeds back the inference results to the network device. AI-based beam management can achieve beam prediction in, for example, the temporal and / or spatial domains, reducing overhead and latency and improving beam selection accuracy.

[0133] Beam management enhancements may include at least one sub-function, such as: beam scan matrix prediction, optimal beam prediction, and beam reference signal received power prediction.

[0134] 3. Enhanced positioning accuracy

[0135] Also known as localization tasks, in line-of-sight (LOS) or non-line-of-sight (NLOS) scenarios, AI-based localization can improve localization accuracy with a smaller number of TRP antennas. Localization enhancement can include at least one sub-function, such as: localization enhancement based on access network devices, localization enhancement based on localization management function network elements, localization enhancement based on terminal devices, and localization assistance information prediction.

[0136] 4. Network energy saving

[0137] Network energy conservation can be achieved through cell activation / deactivation, load reduction, coverage improvement, or other RAN setting adjustments. AI technology can be used to optimize energy-saving decisions by leveraging data collected within the RAN network. AI algorithms can predict energy efficiency and load status for the next cycle, which can be used to assist in cell activation / deactivation decisions to save energy. Based on the predicted load, the system can dynamically configure energy-saving strategies to maintain a balance between system performance and energy efficiency, and reduce energy consumption.

[0138] 5. Load balancing

[0139] Load balancing can distribute the load evenly between cells and across different areas within a cell, or transfer some traffic from congested cells, or offload users across a single cell, carrier, or access standard, thereby improving network performance. Using AI models to enhance load balancing performance—such as inputting various measurements and feedback from terminal devices and network nodes, as well as historical data—can provide a higher quality user experience and increase system capacity.

[0140] 6. Mobility Management

[0141] Mobility management is a solution that ensures service continuity for mobile devices by minimizing dropped calls, radio link failures (RLFs), unnecessary handovers, and ping-pong effects. AI can enhance mobility management by, for example, reducing the probability of unexpected events, predicting device location / mobility / performance, and routing traffic.

[0142] It should be understood that the definitions of the above technical terms are merely illustrative. For example, as technology continues to develop, the scope of the above definitions may also change, and the embodiments of this application are not intended to limit the scope.

[0143] The model training module can be used to train or update a task model using training data.

[0144] The model storage module can be used to store the trained model or the updated model, which is a task model.

[0145] The management module can be used to perform performance prediction (or performance monitoring) on ​​at least one of multiple task models using monitoring data and / or inference output data, and / or to manage multiple task models based on the performance prediction results. In this application, model performance may include model accuracy.

[0146] The data collection module can be used to collect data.

[0147] Different AI-related modules can be interconnected. For example, the data collection module can collect and provide inference data to the inference module, and / or collect and provide monitoring data to the management module, and / or collect and provide training data to the model training module. For example, the inference module can provide inference output data to the management module. For example, the model training module can provide trained or updated task models to the model storage module. For example, the management module can send performance feedback or retraining requests to the model training module, and the model training module can perform model training based on the performance prediction results of at least one task model indicated by the performance feedback or the retraining request. For example, the management module can send model transfer or delivery requests to the model storage module, and the model storage module can perform model transfer or delivery to the inference module based on the request, such as sending the task model or its parameters to the inference module. For example, the management module can send management instructions to the inference module, which can be used to instruct the inference module to perform one or more model management operations.

[0148] Model management operations can include activation / deactivation, switching, selection, or rollback. Assuming the task model used by the inference module is task model 1 (meaning the inference module uses task model 1 to process inference data), and assuming the inference module receives a transferred or transmitted task model 2 from the model storage module, the management instructions received by the inference module from the management module can be used to instruct the inference module to deactivate task model 1 and activate task model 2, or to switch the used task model from task model 1 to task model 2, or to select task model 2 as the task model to use, or to roll back to using the reference model to process the inference data. The reference model can be a non-AI model, a default AI model, or a rollback AI model. In this application, a non-AI model can be understood as a model used to process inference data, and this model is not an AI model. This application does not limit the implementation method of the non-AI model; for example, the non-AI model can be implemented through software, hardware, or a combination of both. Hereinafter, the non-AI model will be referred to as a legacy model. The default AI model or the rollback AI model can be understood as the task model that can be used by default.

[0149] Assuming the data collection module, inference module, and management module are deployed in the base station, the task model is used to perform the spatial beam prediction task. Figure 3-1 schematically illustrates the methods performed by the AI-related functions.

[0150] As shown in process 1 of Figure 3-1, the base station can perform the function of the data collection module by sending the CSI-RS corresponding to each beam in beam set A to the terminal device, and collecting the RSRP corresponding to each beam in beam set A measured by the terminal device. The base station can also perform the function of the inference module by processing the collected RSRP corresponding to each beam in beam set A based on the task model, and obtaining the processing result of the task model, that is, the predicted RSRP corresponding to each beam in beam set B.

[0151] As shown in process 2 of Figure 3-1, in addition to processing the RSRP (i.e., inference data) corresponding to each beam in beam set A based on the task model using the method in process 1 to obtain the predicted RSRP (i.e., inference output data) corresponding to each beam in beam set B, the base station can also collect the tags corresponding to the inference data, i.e., the target results of the RSRP corresponding to each beam in beam set B. As shown in process 2 of Figure 3-1, the base station can execute the data collection module function, collecting the RSRP corresponding to each beam in beam set B measured by the terminal device by sending the CSI-RS corresponding to each beam in beam set B to the terminal device. Afterwards, the base station can execute the management module function, monitoring the performance of the task model based on the target results of the RSRP corresponding to each beam in beam set B and the predicted results of the RSRP corresponding to each beam in beam set B output by the task model.

[0152] The label corresponding to the inference data can be understood as the expected value of the result obtained by the task model after processing the inference data. This expected value can also be called the expected result, the true value, the target output data, the target result, or the measurement result.

[0153] Taking the task model used to perform a time-domain beam prediction task as an example, the inference data of the task model can include the index (ID) or identification (ID) of the best beam in the past T time moments, the inference output data of the task model can include the prediction results of the ID of the best beam in the future K time moments, and the label corresponding to the inference data can include the target result of the ID of the best beam in the future K time moments.

[0154] Taking the task model used to perform a positioning task as an example, the inference data of the task model may include measured channel information, such as channel impulse response (CIR) and / or power delay profile (PDP). The inference output data of the task model may include the predicted UE location (e.g., the predicted UE location coordinates). The label corresponding to the inference data may include the actual UE location.

[0155] Taking a task model that includes Task Model-1 for performing CSI compression and Task Model-2 for performing CSI reconstruction as an example, the inference data of Task Model-1 (i.e., the inference data of the task model) can include measured channel state information. The inference output data of Task Model-1 includes the compression result of the channel state information. The inference data of Task Model-2 includes the inference output data of Task Model-1, i.e., the compression result of the channel state information. The inference output data of Task Model-2 (i.e., the inference output data of the task model) includes the reconstruction result of the channel state information based on the compression result, or in other words, the prediction result of the channel state information based on the compression result. The label corresponding to the inference data of the task models can include the measured channel state information, i.e., the inference data of Task Model-1. Channel state information can be understood by referring to the channel information introduced above.

[0156] Taking a task model used to execute a scheduling task as an example, the inference data of the task model can include the network state, and the inference output data of the task model can include resource scheduling information. The network state generally includes the UE's channel information, the delay information of the data packets to be transmitted, historical average rate information, etc. The scheduling information generally indicates at least one of the following: the UE selected for scheduling, the allocated time-frequency resources, the MIMO precoding codebook, or MCS, etc.

[0157] Figure 3-2 schematically illustrates the method used by the management module in Figure 3-1 to determine the performance of the task model. As shown in Figure 3-2, the management module can acquire sample data ({x, y} as shown in Figure 3-2), which includes inference data (such as x) and the corresponding labels (such as y) of the inference data. Then, it acquires the inference output data obtained by the task model processing the inference data (y as shown in Figure 3-2). Next, it determines the difference between the two (d(y, y) as shown in Figure 3-2), and uses this difference to determine the accuracy of the model. It should be noted that the difference mentioned in this application can also be replaced by a ratio or other parameters used to reflect the magnitude of the difference between the two.

[0158] To evaluate the accuracy of the task model, the data collection module needs to collect inference data and the corresponding labels. Not only is the amount of data required large, but the data collection module also incurs significant overhead because the labels generally require additional measurement or feedback.

[0159] In the method provided in the first aspect above, the first communication device may use an evaluation model when evaluating the performance of one or more task models corresponding to the first task. The evaluation model is used to obtain prediction results of the performance data of the corresponding model based on first data, which includes input data and / or output data of one or more models. The input data and inference data can be cross-referenced and understood, the output data can be cross-referenced and understood with the inference output data, the first communication device can be cross-referenced and understood with the management module, or the management module is deployed within the first communication device.

[0160] Thus, referring to the method provided in the first aspect, the data collection module can collect first data (such as inference data and / or inference output data), and the management module can process the first data based on the evaluation model to obtain the prediction results of the performance data of one or more task models. In scenarios where the management module predicts the performance of one or more of the task models, this helps to reduce the amount of data required by the management module and lowers the overhead of the data collection module.

[0161] As described above, performance data from one or more task models are used to indicate the performance of one or more task models. This application does not limit the manner in which performance data from one or more task models indicates the performance of one or more task models.

[0162] Taking the management module's processing of inference data and inference output data based on the evaluation model as an example, and using the evaluation of the performance of two task models (task model 1 and task model 2) as an example, Figure 4-1 schematically illustrates the process by which the management module obtains the performance data of the two task models. This application does not limit the number of task models evaluated by the management module based on the evaluation model.

[0163] As shown in Figure 4-1, the inference module can input inference data (denoted as x) into task model 1 and task model 2 respectively. Task model 1 processes x to obtain inference output data y1, and task model 2 processes x to obtain inference output data y2. Then, the management module obtains x, y1, and y2 respectively, and based on the data obtained by the evaluation model, obtains the performance data (denoted as r) of task model 1 and task model 2.

[0164] Referring to the first indication method introduced above, the performance data r can indicate the performance value of task model 1 (denoted as r1) and the performance value of task model 2 (denoted as r2), respectively.

[0165] Referring to the second indication method introduced above, the performance data r can indicate the superiority or inferiority relationship between task model 1 and task model 2 in terms of performance, and / or, the performance data r can indicate the magnitude of the performance difference between task model 1 and task model 2, that is, the difference between r1 and r2 (denoted as DeltaPerf). This difference can be a ratio or a difference value.

[0166] This application does not limit the way performance data r indicates the performance superiority / inferiority relationship between task model 1 and task model 2. For example, performance data r can indicate the order of task model 1 and task model 2, indicating their performance superiority / inferiority relationship through their order. For instance, the management module can include R(x, y1, y2), where R(x, y1, y2) = 1 indicates that the performance of task model 1 is better than that of task model 2, and R(x, y1, y2) = 0 indicates that the performance of task model 1 is worse than that of task model 2.

[0167] Optionally, the performance data r may include DeltaPerf and / or R(x, y1, y2).

[0168] Referring to the third indication method described above, the performance data of one or more task models respectively indicate the superiority and / or difference in performance between one or more task models and the reference model. This reference model can be cross-referenced and understood with the traditional model, default AI model, or fallback AI model described above. Thus, referring to the third indication method, Figure 4-2 schematically illustrates another possible process by which the management module obtains the performance data of the two task models. As shown in Figure 4-2, in addition to inputting the inference data (i.e., x) into task model 1 and task model 2 respectively, the inference module also inputs x into the traditional model. Task model 1 processes x to obtain inference output data y1, task model 2 processes x to obtain inference output data y2, and the traditional model processes x to obtain data y0. Then, the management module obtains x, y1, y2, and y0 respectively, and based on the data obtained from the evaluation model, obtains the performance data (i.e., r) of task model 1 and task model 2. Among them, the performance data r can indicate the superiority and / or difference between the performance value of task model 1 (i.e., r1) and the performance value of the traditional model (denoted as r0), and can also indicate the superiority and / or difference between the performance value of task model 2 (i.e., r2) and r0.

[0169] The previous text introduced the types of inference data and inference output data in various tasks, which will not be repeated here.

[0170] Referring to the preceding text, the performance of a task model can include its accuracy. Taking the first task as temporal beam prediction, task model accuracy can be understood as the proportion of the predicted IDs of the best beams at the next K time points obtained from the task model to the target results of those IDs. Taking the first task as spatial beam prediction, task model accuracy can be understood as the difference between the RSRP (Recovery Rate Reduction) of each beam in beam set B obtained from the task model and the target or measured RSRP of each beam in beam set B. Taking the first task as CSI compression reconstruction, task model accuracy can be understood as the difference between the reconstructed channel state information based on the task model and the measured channel state information. Optionally, the performance of a task model can include the performance of the first task performed by the task model. Taking the first task as temporal beam prediction, the performance of the first task performed by the task model can be understood as the performance (e.g., bit error rate) of transmitting data on the selected beam based on the predicted IDs of the best beams at the next K time points obtained from the task model. Taking the first task as spatial beam prediction as an example, the performance of the first task executed by the task model can be understood as the performance of transmitting data on the RSRP-selected beam corresponding to each beam in the beam set B obtained by the task model. Taking the first task as CSI compression reconstruction as an example, the performance of the first task executed by the task model can be understood as the performance of the link or system after determining the scheduling information based on the channel state information reconstructed by the task model and executing the scheduling according to the scheduling information. The link performance can include block error rate (BLER) and / or signal to interference plus noise ratio (SINR), etc. The system performance can include system throughput and / or latency, etc.

[0171] After the management module obtains performance data for one or more task models, how to manage these task models is a pressing issue. The management module typically enables the best-performing task model to execute the first task. However, when the performance of the task model used to execute the first task (called the target model) is no longer superior to other task models, switching the task model used for the first task can easily lead to a large number of model switching operations. Switching models involves operations such as loading or unloading models, resulting in significant overhead for model management.

[0172] In the method provided in the second aspect above, a first communication device determines the performance difference between a first model and a second model among multiple models, wherein the target model corresponding to the first task is the first model. Then, based on the difference satisfying a first condition, the first communication device switches among the multiple models to use the model for performing the first task. The first condition includes: the difference indicating that the performance of the second model is better than that of the first model, and the absolute value of the difference exceeding a first threshold. The first communication device can mutually refer to and understand a management module, or the management module is deployed within the first communication device. The first model and the second model belong to multiple models, which may include one or more task models, or the multiple models may include one or more task models and a traditional model.

[0173] Thus, referring to the method provided in the second aspect, when the performance of the first model is inferior to that of the second model, the management module will instruct the inference module to switch the model used to perform the first task only when the difference between the two exceeds a first threshold. This helps to reduce the number of model switching and reduce the overhead of the management module.

[0174] Referring to the second aspect, this application does not limit the specific method by which the management module switches the model used to perform the first task. For example, referring to the first switching method, the management module can switch the model used to perform the first task by updating the first management information, where the first management information indicates the model used to perform the first task. For example, referring to the second switching method, the management module can switch the model used to perform the first task by updating the second management information, where the second management information indicates the status of multiple models respectively. The following description uses the second switching method as an example to illustrate the method executed by the management module.

[0175] Referring to the second aspect, the management module can store the states of multiple models. A model's state is one of these multiple states, which includes a first state, a second state, and optionally, a third state. The definitions of the first, second, and third states are illustrated below with examples.

[0176] The first state, also known as the active state, is where the model is used for inference in the first task. The second state, also known as the monitoring state, is where the inference module does not use the model in the monitoring state to perform the first task, but the management module needs to monitor its performance. The third state, also known as the de-active state, is where the inference module does not use the model in the de-active state to perform the first task, and the management module does not need to monitor its performance, or it does need to monitor its performance, but the frequency with which the management module monitors the performance of the de-active model is lower than the frequency with which it monitors the performance of the model in the monitoring state.

[0177] Referring to the second aspect, the multiple states can correspond to different multiple model management information, and the management module can manage the model according to the model's state.

[0178] As described above, the model management information may indicate at least one of the following: first indication information, second indication information, conditions for switching states among the plurality of states, or configuration information for measuring the model's performance. The first indication information may indicate whether the first task is executed or not; the second indication information may indicate the configuration for executing the first task; and the configuration information for measuring the model's performance may indicate one or more parameters for measuring the model's performance, which may include the measurement cycle (or the frequency of monitoring the model's performance).

[0179] For example, in the model management information corresponding to the active state, the first indication information can indicate that the first task should be executed, and the configuration information can indicate that the frequency of monitoring the model's performance is frequency 1. In the model management information corresponding to the monitoring state, the first indication information can indicate that the first task should not be executed, and the configuration information can indicate that the frequency of monitoring the model's performance is frequency 2, where frequency 2 is less than or equal to frequency 1. In the model management information corresponding to the deactivated state, the first indication information can indicate that the first task should not be executed, and the configuration information can indicate that the model's performance should not be monitored, or the configuration information can indicate that the frequency of monitoring the model's performance is frequency 3, where frequency 3 is less than frequency 2.

[0180] From a deployment or implementation perspective, model management information can include deployment information indicating the model's storage location. For example, in model management information corresponding to the active state and the monitoring state, the deployment information can instruct the model to be loaded into the memory of a processing unit, which can be a central processing unit (CPU), graphics processing unit (GPU), or neural processing unit (NPU) for inference or real-time switching. Conversely, in model management information corresponding to the deactivated state, the deployment information can instruct the model to be stored on an external storage device such as a hard drive or on a server. When the model's state changes from deactivated to active or monitoring, the model needs to be loaded from external storage into memory or downloaded from the server to the device, which involves a certain delay.

[0181] Model management involves determining which models to activate, maintaining and monitoring models, and switching model states. Maintaining and monitoring models can be understood as determining which models need to be monitored, such as updating the list of models that need monitoring, removing some models from the list, and adding others.

[0182] Figure 5 schematically illustrates the model management process performed by the management module. As shown in Figure 5, model management can be divided into three stages based on the model's state: Stage 1: Initial Model Selection -> Stage 2: Model Tracking -> Stage 3: Model Determination. Stage 1: Initial Model Selection can be understood as ranking multiple initial models by performance to determine K candidate models. Figure 5 uses an example with N+1 initial models and K = 3, where N is a positive integer, and this application does not limit the number of N and K. Stage 2: Model Tracking can be understood as monitoring and tracking the performance of candidate models. Stage 3: Model Determination can be understood as determining the active model from the K candidate models, or, more specifically, determining the state of each of the K candidate models. As shown in Figure 5, initial models not selected as candidate models in Stage 1 can be in an idle or deactivated state. Figure 5 uses model 1 as an example of a model selected as active. Optionally, one or more of the K candidate models can be in a monitoring state.

[0183] As mentioned earlier, the K candidate models can include traditional models (or traditional methods) for performing fallback.

[0184] As introduced above, the switching of the model state generally involves operations such as model loading or unloading. To avoid the extra overhead caused by unnecessary switching, the conditions for state switching can be configured. Among them, the management module can configure performance difference thresholds (N1 and N2), where N1 < N2. The management module can determine the state of the model to be monitored based on the difference (referred to as the performance difference) between the performance of a certain model (referred to as the model to be monitored) and the performance of other models (referred to as the reference model). Hereinafter, it is assumed that when the value of the performance of the model to be detected is greater than the value of the performance of the reference model, the performance of the model to be detected is better than the performance of the reference model. On the contrary, when the value of the performance of the model to be detected is less than the value of the performance of the reference model, the performance of the model to be detected is worse than the performance of the reference model.

[0185] FIG. 6 schematically shows multiple states of the model and the transition conditions between different states.

[0186] As shown in FIG. 6, based on the model to be detected being an active model, when the performance difference is greater than -N1, the management module can transfer the state of the model to the monitoring state. When the performance difference is greater than -N2, the management module can transfer the state of the model to the deactivated state. The type of the reference model is not limited in this application. For example, based on the model to be detected being an active model, the reference model can be a model in the monitoring state or a candidate model.

[0187] As shown in FIG. 6, based on the model to be detected being a model in the monitoring state, when the performance difference is greater than +N1, the management module can transfer the state of the model to the active state. When the performance difference is greater than -N2, the management module can transfer the state of the model to the deactivated state. The type of the reference model is not limited in this application. For example, based on the model to be detected being a model in the monitoring state, the reference model can be an active model.

[0188] Continuing to refer to FIG. 6, based on the model to be detected being a deactivated model, when the performance difference is greater than +N2, the management module can transfer the state of the model to the monitoring state. The type of the reference model is not limited in this application. For example, based on the model to be detected being a deactivated model, the reference model can be an active model or a model in the monitoring state.

[0189] The first threshold introduced above can be understood in reference to N1, and the second threshold can be understood in reference to N2.

[0190] In this context, a performance difference greater than -N1 means that the performance value of the model being tested is less than the performance value of the reference model, and the absolute value of the difference between the two values ​​is greater than N1. Similarly, a performance difference greater than -N2 means that the performance value of the model being tested is less than the performance value of the reference model, and the absolute value of the difference between the two values ​​is greater than N2. Likewise, a performance difference greater than +N2 means that the performance value of the model being tested is greater than the performance value of the reference model, and the absolute value of the difference between the two values ​​is greater than N2. Similarly, a performance difference greater than +N1 means that the performance value of the model being tested is greater than the performance value of the reference model, and the absolute value of the difference between the two values ​​is greater than N1.

[0191] Alternatively, the management module can use the traditional model or the default model as a reference model, and use the performance of the reference model as a baseline to manage the state of the model. In other words, it can determine the state of the model based on the performance difference between the model to be monitored and the traditional model.

[0192] For example, based on the model to be monitored being an active model (let's say model 1), when the difference between the performance value of model 1 and the baseline is greater than +N², the management module can continue to monitor the performance of model 1. When the difference between the performance value of model 1 and the baseline is less than +N², the management module can use the model in the monitoring state (let's say model 2) as the model to be monitored. When the difference between the performance value of model 2 and the baseline is greater than +N², the management module can change the state of model 2 from monitoring to active, and change the state of model 1 from active to monitoring. When the difference between the performance value of model 2 and the baseline is less than +N², the management module can continue to estimate the performance difference. When the difference between the performance value of model 1 and the baseline is less than -N¹, the management module can set the state of the traditional model or the default model to active, and change the state of model 1 from active to monitoring or deactivated.

[0193] As described above, the methods provided in the first aspect and the methods provided in the second aspect can be combined. For example, the management module can use the evaluation model to determine the difference between the performance of the model to be monitored and the performance of the reference model, and then determine the status of the model to be monitored based on this difference.

[0194] To ensure the accuracy of the performance of the task model evaluated by the evaluation model, the management module needs to verify or update the evaluation model. The evaluation model can be an AI model; therefore, the data processed by the evaluation model can be understood as its inference data, and the performance data obtained by the evaluation model can be understood as its inference output data. For clarity, the inference data of the evaluation model will be referred to as evaluation data, and the inference output data as evaluation output data. Optionally, the management module can process the evaluation data (e.g., the first data mentioned earlier) based on the evaluation model to obtain evaluation output data (e.g., the prediction results of the performance data mentioned earlier). The management module can then verify or update the evaluation model based on the evaluation output data and its expected results. Correspondingly, to verify or update the evaluation model, the data collection module needs to collect the expected results of the evaluation output data, and then verify or update the evaluation model based on the evaluation output data and its expected results.

[0195] Figure 7 schematically illustrates the timeline of performance monitoring and model management in an AI / ML system. As shown by the white rectangles in Figure 7, the data collection module can perform data collection; for example, it can collect inference data for the task model for the inference module. As shown by the black rectangles in Figure 7, the inference module can process the inference data based on the task model to obtain inference output data. Referring to the arrows between the rectangles in Figure 7, the inference module can provide inference output data and / or inference data (i.e., evaluation data) to the management module. As shown by the black striped rectangles in Figure 7, the management module can process the evaluation data based on the evaluation model to obtain evaluation output data. As shown by the white rectangles in Figure 7, the data collection module can also collect the expected results of the evaluation output data for the management module. As shown by the black striped rectangles in Figure 7, the management module can also verify or update the task model based on the evaluation output data and its expected results.

[0196] Once validated or updated, the evaluation model can be used for subsequent task model monitoring. Data collection can be performed periodically or based on events, where an event could be a decline in system performance (such as throughput).

[0197] Figure 8 schematically illustrates a possible flow of a communication method in an AI / ML system. In the method shown in Figure 8, the data collection module, inference module, and management module are deployed on the terminal device. As shown in Figure 8, this method may include steps S801 to S807.

[0198] S801, The terminal device collects inference data from the base station;

[0199] S802, The terminal device processes inference data based on the task model;

[0200] S803, Terminal equipment collects data from base station 8-1;

[0201] Data 8-1 may include inference data and / or inference output data, and may also include the expected results of evaluating the output data.

[0202] S804, terminal equipment verification or training evaluation model;

[0203] The terminal device can process inference data and / or inference output data based on the evaluation model to obtain the predicted results of the performance data of the task model (i.e., evaluation output data). After that, the terminal device can verify or train the evaluation model based on the evaluation output data and the expected results of the collected evaluation output data.

[0204] S805, terminal devices process inference data based on task models;

[0205] S806, Terminal Equipment Usage Evaluation Model Management Task Model;

[0206] S807, The terminal device sends the model management results to the base station.

[0207] Taking beam management on the terminal device side as an example, the terminal device measures the RSRP of beam set A based on its task model and the reference signal transmitted by the base station, and predicts the RSRP of beam set B. Then, the terminal device can calculate the performance of the beam prediction model based on the measured RSRP of beam set B, or transmit based on the predicted beams, with the base station providing feedback on the transmission performance. The terminal device verifies the evaluation model based on the collected performance data. For example, if the performance error predicted by the evaluation model is below a threshold, the verification passes; otherwise, the evaluation model is updated based on the collected performance data.

[0208] Taking terminal device-side positioning as an example, the terminal device measures channel information and predicts its location based on its task model and reference signals transmitted by the base station. The terminal device can then calculate model performance based on its location measured by the base station. The terminal device validates the evaluation model using the collected performance data. For example, if the performance error predicted by the evaluation model is below a threshold, the validation passes; otherwise, the evaluation model is updated based on the collected performance data.

[0209] The embodiments of this application evaluate the relative performance of the model prediction task model, and manage the model and switch states based on the relative performance, which helps to save monitoring overhead and unnecessary model state switching.

[0210] Figure 9 schematically illustrates a possible flow of a communication method in an AI / ML system. In the method shown in Figure 9, a data collection module is deployed on both the terminal device and the base station, an inference module is deployed on the terminal device, and management modules are deployed on both the terminal device and the base station. As shown in Figure 9, this method may include steps S901 to S909.

[0211] S901, The terminal device collects inference data from the base station;

[0212] S902, Terminal devices process inference data based on task models;

[0213] S903, The base station collects data from the terminal equipment 9-1;

[0214] Data 9-1 may include inference data and / or inference output data, and may also include the expected results of evaluating the output data.

[0215] S904, base station verification or training evaluation model;

[0216] The base station can process inference data and / or inference output data based on the evaluation model to obtain the predicted results of the performance data of the task model (i.e., evaluation output data). After that, the base station can verify or train the evaluation model based on the evaluation output data and the expected results of the collected evaluation output data.

[0217] S905, The base station sends the evaluation model to the terminal equipment;

[0218] S906. The base station sends model management configuration to the terminal device;

[0219] The model management configuration can refer to N1 and N2 as described above.

[0220] S907, Terminal devices process inference data based on task models;

[0221] S908, Terminal Equipment Usage Evaluation Model Management Task Model;

[0222] S909, The terminal device sends the model management results to the base station.

[0223] The embodiments of this application evaluate the relative performance of the model prediction task model, and manage the model and switch states based on the relative performance, which helps to save monitoring overhead and unnecessary model state switching.

[0224] Figure 10 schematically illustrates a possible flow of a communication method in an AI / ML system. In the method shown in Figure 10, a data collection module is deployed at the terminal device and the base station, respectively; the inference module is deployed at the terminal device; and the management module is deployed at the base station. As shown in Figure 10, this method may include steps S1001 to S1008.

[0225] S1001, The terminal device collects inference data from the base station;

[0226] S1002, The terminal device processes inference data based on the task model;

[0227] S1003, The base station collects data from the terminal equipment 10-1;

[0228] Data 10-1 may include inference data and / or inference output data, and may also include the expected results of evaluating the output data.

[0229] S1004, Base station verification or training evaluation model;

[0230] S1004 can be understood by referring to S904.

[0231] S1005. Terminal devices process inference data based on task models;

[0232] S1006, The terminal device sends data 10-2 to the base station;

[0233] Data 10-2 may include inference output data and / or inference data of the task model.

[0234] S1007, Base Station Usage Evaluation Model Management Task Model;

[0235] S1008. The base station sends the model management results to the terminal equipment.

[0236] The base station verifies the evaluation model, the terminal device feeds back the model inference results to the base station, the base station monitors the performance and manages the model based on the inference results, and the base station notifies the terminal device of the model management results.

[0237] Figure 11 schematically illustrates a possible flow of a communication method in an AI / ML system. In the method shown in Figure 11, a data collection module is deployed on the terminal device, an inference module is deployed on the terminal device, and management modules are deployed on both the base station and the terminal device. As shown in Figure 11, this method may include steps S1101 to S1109.

[0238] S1101, The terminal device collects inference data from the base station;

[0239] S1102. The terminal device processes inference data based on the task model;

[0240] S1103, Terminal equipment collects data from base station 11-1;

[0241] Data 11-1 may include inference data and / or inference output data, and may also include the expected results of evaluating the output data.

[0242] S1104, Terminal equipment verification or training evaluation model;

[0243] S1104 can be understood by referring to the content of S804.

[0244] S1105. Terminal devices process inference data based on task models;

[0245] S1106. The terminal device obtains the prediction results of performance data based on the evaluation model;

[0246] S1107, Prediction results of performance data sent by terminal equipment to base station;

[0247] S1108, Base station management task model based on performance data prediction results;

[0248] S1109. The base station sends the model management results to the terminal equipment.

[0249] The terminal device validates or updates the evaluation model based on the collected performance data. Based on the evaluation model, the terminal device predicts the relative performance of the task model and sends the model performance data to the base station. The base station manages the model based on the performance data and notifies the terminal device of the management results.

[0250] Figure 12 schematically illustrates a possible flow of a communication method in an AI / ML system. In the method shown in Figure 12, a data collection module is deployed on the terminal device and a data base station, an inference module is deployed on the terminal device, and a management module is deployed on the base station. It is assumed that the task model includes the previously described task model-1 (denoted as the ENC model) and task model-2 (denoted as the DEC model). As shown in Figure 12, this method may include steps S1201 to S1211.

[0251] S1201, The terminal device collects inference data from the base station;

[0252] S1202, The terminal device processes inference data based on the ENC model;

[0253] S1203, The terminal device sends compressed data to the base station;

[0254] S1204. The base station processes the compression model based on the DEC model.

[0255] S1205, Base station verification or training evaluation model;

[0256] S1206. The base station sends the evaluation model to the terminal equipment.

[0257] S1207. Terminal devices process inference data based on the ENC model;

[0258] S1208, The terminal device obtains the prediction results of performance data based on the evaluation model;

[0259] S1209, Prediction results of performance data sent by the terminal equipment to the base station;

[0260] S1210, Base station management task model based on performance data prediction results;

[0261] S1211, The base station sends the model management results to the terminal equipment.

[0262] The base station verifies the evaluation model based on the decoding performance results and the output of the terminal device's coding model, and sends the verified evaluation model, along with the model management configuration, to the terminal device. Subsequently, the terminal device monitors the relative performance of different coding models based on the outputs of the evaluation model and the local coding model, and feeds back the predicted performance to the base station. The base station performs model management (including model state switching) based on the relative performance of the models and notifies the terminal device of the management results (the terminal device then switches the state of the paired coding models).

[0263] Figure 13 schematically illustrates a possible flow of a communication method in an AI / ML system. As shown in Figure 13, the method may include steps S1301 to S1310.

[0264] S1301, The terminal device collects inference data from the base station;

[0265] S1302. Terminal devices process inference data based on the ENC model;

[0266] S1303, The terminal device sends compressed data to the base station;

[0267] S1304. The base station processes the compression model based on the DEC model.

[0268] S1305, The base station sends data 13-1 to the terminal device;

[0269] Data 13-1 may include the inference output data of the DEC model. As mentioned earlier, the inference output data of the DEC model is the inference output data of the task model corresponding to the ENC model and the DEC model. Therefore, the inference output data of the DEC model can be used to train or validate and evaluate the model.

[0270] S1306, Terminal equipment verification or training evaluation model;

[0271] The terminal device can process the inference data of the ENC model and / or the inference output data of the DEC model based on the evaluation model to obtain the prediction results of the performance data of the ENC model and / or DEC model (i.e., evaluation output data). After that, the terminal device can verify or train the evaluation model based on the evaluation output data and the expected results of the collected evaluation output data.

[0272] S1307. The base station sends model management configuration to the terminal device;

[0273] The model management configuration can refer to N1 and N2 as described above.

[0274] S1308. Terminal devices process inference data based on the ENC model;

[0275] S1309, Terminal Equipment Usage Evaluation Model Management Task Model;

[0276] S1310, The terminal device sends the model management results to the base station.

[0277] The terminal device collects performance data from the base station's decoding model and verifies the evaluation model. Then, the terminal device receives the base station's model management configuration and uses this configuration to predict and manage the performance of the coding model. The terminal device and the base station synchronize the model management results.

[0278] In the method shown in Figure 12 or Figure 13, the management module can perform performance monitoring and model management on both sides based on the evaluation model, reducing the overhead of monitoring data collection.

[0279] In the methods shown in Figures 8 to 13, a network device is used as an example of a base station. The solution provided in this application does not limit the specific type of network device; correspondingly, in the methods shown in Figures 8 to 13, the base station can be replaced by a network device.

[0280] The first and second communication devices have been introduced previously. In the methods shown in Figures 8 to 13, the terminal device can be replaced by the first communication device, and the base station can be replaced by the second communication device. As described above regarding the possible configurations of the first and second communication devices, in the methods shown in Figures 8 to 13, optionally, the terminal device can be replaced by a network device, and the base station can be replaced by the terminal device; alternatively, optionally, the terminal device and the base station can be replaced by different network devices (e.g., network device 1 and network device 2); or alternatively, optionally, the terminal device and the base station can be replaced by different terminal devices (e.g., terminal device 1 and terminal device 2).

[0281] The methods shown in Figures 8 to 13 illustrate several processes involved in the AI / ML system described above. These processes include data collection, inference of the task model, validation or training of the evaluation model, and management of the task model based on the evaluation model. Figure 8 illustrates a method where a terminal device executes these processes alone; Figures 9 to 13 illustrate methods where a terminal device and network devices work together to execute these processes. Optionally, these processes can be executed solely by the network device, or they can be executed jointly by a greater number of communication devices, such as servers in addition to the terminal device and network device.

[0282] The communication apparatus provided in the third aspect of this application has been described above. This communication apparatus may include an acquisition unit and a processing unit.

[0283] Optionally, the communication device can be used to execute the steps or processes performed by the terminal device or base station in any of the examples in Figures 8 to 13. For details, please refer to the relevant descriptions in the foregoing method examples.

[0284] The preceding text also describes a communication apparatus provided in the fourth aspect of this application. This communication apparatus includes at least one processor, which executes a computer program stored in a memory, such that the processor performs the steps or processes executed by a terminal device or base station in any of the examples shown in Figures 8 to 13.

[0285] The preceding text also describes a communication apparatus provided in the fifth aspect of this application. This communication apparatus includes at least one logic circuit and an input / output interface. The logic circuit is used to execute steps or processes performed by a terminal device or base station in any of the examples shown in Figures 8 to 13.

[0286] The preceding text also describes a chip (or chip device or chip system) provided in the sixth aspect of this application. This chip includes a processor for calling computer programs or instructions stored in memory, causing the processor to execute the steps or processes executed by a terminal device or base station in any of the examples shown in Figures 8 to 13. Optionally, the processor is coupled to the memory via an interface.

[0287] In this application, the processor mentioned anywhere may be a general-purpose central processing unit, a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of a program that controls the methods provided in any of the above embodiments. The memory mentioned anywhere above may be read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, such as random access memory (RAM).

[0288] The preceding text also describes a computer-readable storage medium provided in the eighth aspect of this application, which includes computer instructions that, when executed on a computer, cause the computer to perform steps or processes performed by a terminal device or base station in any of the examples shown in Figures 8 to 13.

[0289] The preceding text also describes a computer program product including computer instructions provided in the ninth aspect of this application, which, when run on a computer, causes the computer to perform the steps or processes performed by a terminal device or base station in any of the examples shown in Figures 8 to 13.

[0290] The preceding text also describes the communication system provided in the seventh aspect of this application, which includes a terminal device and a base station as shown in any of the examples in Figures 8 to 13.

[0291] In this application, the processing unit can be implemented by at least one processor or processor-related circuitry. Specifically, the processor may include a modem chip, or a SoC chip or SIP chip containing a modem core. The transceiver unit can be implemented by a transceiver or transceiver-related circuitry. The transceiver unit may also be referred to as a communication module or communication interface. The storage module can be implemented by at least one memory.

[0292] Optionally, in this application, when the communication device is a circuit or chip responsible for communication functions, such as a modem chip or a SoC chip or SIP chip containing a modem core, the function of the processing unit can be implemented by a circuit system in the aforementioned chip that includes one or more processors or processing cores. The function of the transceiver unit can be implemented by the interface circuit or data transceiver circuit on the aforementioned chip.

[0293] In this application, when the communication device is a terminal device, Figure 14 shows a simplified structural schematic diagram of a terminal device. As shown in Figure 14, the terminal includes a processor, a memory, and a transceiver. The memory can store computer program code, and the transceiver includes a transmitter 1431, a receiver 1432, a radio frequency circuit (not shown in the figure), an antenna 1433, and input / output devices (not shown in the figure).

[0294] The processor is primarily used for processing communication protocols and data; controlling the terminal; executing software programs; and processing data from those programs. The memory is primarily used for storing software programs and data. The radio frequency (RF) circuitry is primarily used for converting baseband signals to RF signals and processing RF signals. The antenna is primarily used for transmitting and receiving RF signals in the form of electromagnetic waves. Input / output devices may include touchscreens, displays, or keyboards. These devices are primarily used for receiving user input and outputting data to the user. It should be noted that some types of terminals may not have input / output devices.

[0295] When data needs to be transmitted, the processor performs baseband processing on the data to be transmitted and outputs a baseband signal to the radio frequency (RF) circuit. The RF circuit then processes the baseband signal and transmits it outwards via an antenna as electromagnetic waves. When data is sent to the terminal, the RF circuit receives the RF signal through the antenna. The RF circuit converts the RF signal back into a baseband signal and outputs it to the processor. The processor converts the baseband signal back into data and processes the data. For ease of explanation, Figure 14 only shows one memory, processor, and transceiver. In actual terminal products, there may be one or more processors and one or more memories. Memory can also be called storage medium or storage device, etc. Memory can be independent of the processor or integrated with the processor; this embodiment does not limit this.

[0296] In the embodiments of this application, the antenna and radio frequency circuit with transceiver function can be regarded as the transceiver unit of the terminal, and the processor with processing function can be regarded as the processing unit of the terminal.

[0297] As shown in Figure 14, the terminal includes a processor 1410, a memory 1420, and a transceiver 1430. The processor 1410 may also be referred to as a processing unit, processing board, processing unit, or processing device, etc. The transceiver 1430 may also be referred to as a transceiver unit, transceiver, or transceiver device, etc.

[0298] Optionally, the devices in transceiver 1430 used to implement the receiving and / or transmitting functions can be considered as transceiver units. A transceiver may also be referred to as a transceiver module, transceiver circuit, etc.

[0299] The processor 1410 is used to execute the steps or processes performed by the terminal device in any of the examples in Figures 8 to 13. The transceiver 1430 is used to execute the transmit and receive actions of the terminal device in any of the examples in Figures 8 to 13.

[0300] It should be understood that Figure 14 is merely an example and not a limitation, and the terminal described above, including the transceiver unit and the processing unit, may not depend on the structure shown in Figure 14.

[0301] When the communication device 1400 is a chip, the chip includes a processor, a memory, and a transceiver. The transceiver can be an input / output circuit or a communication interface. The processor can be a processing unit integrated on the chip, a microprocessor, or an integrated circuit. In the above method embodiments, the terminal's sending operation can be understood as the chip's output, and the terminal's receiving operation in the above method embodiments can be understood as the chip's input.

[0302] In this application, when the communication device is a network device, such as a gNB or a base station, Figure 15 shows a simplified schematic diagram of a base station structure. The base station includes part 1510, part 1520, and part 1530.

[0303] The 1510 section is mainly used for baseband processing and base station control; the 1510 section is usually the control center of the base station, which can be called the processor, and is used to control the base station to perform the processing operations on the access network equipment side in the above method embodiments.

[0304] Section 1520 is primarily used to store computer program code and data.

[0305] Section 1530 is primarily used for transmitting and receiving radio frequency (RF) signals, as well as converting RF signals to baseband signals. Section 1530 is commonly referred to as a transceiver unit, transceiver module, transceiver, transceiver circuit, or transceiver. The transceiver module of section 1530, also known as a transceiver, includes antenna 1533 and RF circuitry (not shown in the figure), where the RF circuitry is mainly used for RF processing. Optionally, the device in section 1530 that performs the receiving function can be considered a receiver, and the device that performs the transmitting function can be considered a transmitter; that is, section 1530 includes receiver 1532 and transmitter 1531. The receiver can also be called a receiving unit, receiver circuit, or receiving unit, and the transmitter can be called a transmitting module, transmitter, or transmitting circuit.

[0306] Sections 1510 and 1520 may include one or more circuit boards, each of which may include one or more processors and one or more memories. The processors are used to read and execute programs from the memories to implement baseband processing functions and control the base station. If multiple circuit boards exist, they can be interconnected to enhance processing capabilities. As an alternative implementation, multiple circuit boards may share one or more processors, multiple circuit boards may share one or more memories, or multiple circuit boards may simultaneously share one or more processors.

[0307] For example, in one implementation, the transceiver module in section 1530 is used to execute the transceiver-related processes performed by the network device side in any of the examples in Figures 8 to 13. The processor in section 1510 is used to execute the processing-related processes performed by the network device side in any of the examples in Figures 8 to 13.

[0308] It should be understood that Figure 15 is merely an example and not a limitation, and the network devices described above, including processors, memory, and transceivers, may not depend on the structure shown in Figure 15.

[0309] When the communication device 1500 is a chip, the chip includes a transceiver, a memory, and a processor. The transceiver can be an input / output circuit or a communication interface; the processor can be a processor integrated on the chip, a microprocessor, or an integrated circuit. In the above method embodiments, the transmitting operation of the network device can be understood as the chip's output, and the receiving operation of the network device in the above method embodiments can be understood as the chip's input.

[0310] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the explanations and beneficial effects of the relevant contents in any of the above-mentioned devices can be referred to the corresponding method embodiments provided above, and will not be repeated here.

[0311] 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 an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0312] 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.

[0313] Furthermore, 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. The integrated unit can be implemented in hardware or as a software functional unit.

[0314] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the essential contribution of the technical solution of this application, or all or part 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, ROM, RAM, magnetic disks, or optical disks.

[0315] References to "one embodiment" or "some embodiments" as described in this application mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0316] In the description of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. "And / or" in this document 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 alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" 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 represent: a, b, c; a and b; a and c; b and c; or a and b and c. Where a, b, and c can be single or multiple.

[0317] Furthermore, in the embodiments of this application, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design scheme described as an "example" in this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the term "example" is intended to present concepts in a concrete manner. In the embodiments of this application, "of," "corresponding, relevant," and "corresponding" may sometimes be used interchangeably, and it should be noted that their intended meanings are consistent unless their distinction is emphasized.

[0318] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A communication method, characterized in that, A first communication device applied in a wireless communication network, the method comprising: The evaluation model processes the first data, wherein the first data includes input data and / or output data of one or more task models, the one or more task models corresponding to a first task in the wireless communication system, and the evaluation model is used to obtain the prediction result of the performance data of the corresponding model based on the input data and / or output data of one or more models, wherein the evaluation model and the task model are artificial intelligence (AI) models. Obtain the prediction results of the performance data of one or more of the task models.

2. The method according to claim 1, characterized in that, The first data includes input data and / or output data of multiple task models, and the performance data of the multiple task models indicates the superiority and / or difference in performance of different task models.

3. The method according to claim 1, characterized in that, The performance data of one or more of the task models respectively indicate the superiority and / or difference in performance between one or more of the task models and a reference model, wherein the reference model is a non-AI model or a default AI model corresponding to the first task.

4. The method according to claim 2 or 3, characterized in that, The performance of the task model or the reference model includes the accuracy of the corresponding model and / or the performance of the first task performed using the corresponding model.

5. The method according to any one of claims 1-4, characterized in that, The first task includes at least one of the following: Channel state information feedback task, channel estimation task, channel prediction task, beam management task, positioning task, sensing task, network energy saving task, load balancing task, or mobility management task.

6. The method according to any one of claims 1-5, characterized in that, The method further includes: Based on the prediction results, a first performance difference is determined between the first model and the second model corresponding to the first task, wherein the first model is the task model used to perform the first task, and the second model is not used to perform the first task; Based on the first difference satisfying the first condition, switch the model used to perform the first task; The first condition includes: the first difference indicates that the performance of the second model is better than that of the first model, and the absolute value of the first difference exceeds a first threshold.

7. The method according to claim 6, characterized in that, The switching of the model used to perform the first task includes: The state of the first model is switched from a first state among multiple states to another state among the multiple states, and the state of the third model is switched from a second state among the multiple states to the first state, wherein the model in the first state is used to perform the first task, the model in the second state is not used to perform the first task, and the third model is the second model or another model other than the first model and the second model, and the other model corresponds to the first task.

8. The method according to any one of claims 1-5, characterized in that, The method further includes: Obtain a target result of the performance data of one or more of the task models, the target result being associated with the first data; Based on the prediction results and the target results, train the evaluation model and / or verify the accuracy of the evaluation model.

9. A communication method, characterized in that, A first communication device applied in a wireless communication system, the method comprising: The performance difference between a first model and a second model among multiple models is determined, wherein the multiple models respectively correspond to a first task in the wireless communication system, and the first model is used to perform the first task, while the second model is not used to perform the first task; Based on the fact that the difference satisfies the first condition, switch the model used to perform the first task among the plurality of models; The first condition includes: the difference indicates that the performance of the second model is better than that of the first model, and the absolute value of the difference exceeds a first threshold.

10. The method according to claim 9, characterized in that, Switching between the multiple models for performing the first task includes: The state of the first model is switched from the first state among multiple states to other states among multiple states, and the state of the third model is switched from the second state among multiple states to the first state, wherein the model in the first state among multiple models is used to perform the first task, the model in the second state among multiple models is not used to perform the first task, and the third model is the second model or other models among multiple models other than the first model and the second model.

11. The method according to claim 10, characterized in that, When the absolute value of the difference exceeds the second threshold, and the second threshold is greater than the first threshold, the other state is the third state among the plurality of states, wherein the model in the third state among the plurality of models is not used to perform the first task and does not support direct switching to the first state.

12. The method according to claim 10 or 11, characterized in that, The multiple states correspond to different multiple model management information, and the model management information indicates at least one of the following: First indication information, second indication information, conditions for switching states among the plurality of states, or configuration information for measuring the performance of the model, wherein the first indication information indicates whether the first task is executed or not, and the second indication information indicates the configuration for executing the first task.

13. A communication device, characterized in that, The communication device includes multiple functional modules that interact with each other to implement the method as described in any one of claims 1 to 12.

14. A communication device, characterized in that, It includes at least one processor, said at least one processor being used to perform the method as described in any one of claims 1 to 12.

15. A chip, characterized in that, Includes a processor for invoking a computer program or computer instructions in memory to cause the processor to perform the method as described in any one of claims 1 to 12.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program or instructions that, when executed by a communication device, implement the method as described in any one of claims 1 to 12.

17. A computer program product, characterized in that, It includes a computer program or instructions that, when executed by a computer, implement the method as described in any one of claims 1 to 12.

Images