AI model switching or update method and communication device

Abstract

This application provides a method for switching or updating an AI model. Changes in the AI ​​model's operating environment (such as the movement speed of inference network elements, channel environment, and so on) can be learned by monitoring the AI ​​model's input or intermediate performance indicators, for example, by monitoring the AI ​​model's input indicator or the AI ​​model's intermediate performance indicator. In this way, whether or not the AI ​​model needs to be switched or updated is determined based on corresponding information and adapted to the changes in the AI ​​model's operating environment. This helps to mitigate the problem of the AI ​​model's performance degrading or becoming worse due to significant changes in the AI ​​model's operating environment.

Description

Technical Field

[0001]

[0002] Technical Field The embodiments of this application are related to the field of machine learning, and in particular, to a method for switching or updating an AI model and a communication device.

Background Art

[0002]

[0003] Currently, artificial intelligence (AI) has been introduced into wireless communication networks and is widely applied to many application scenarios of air interface technology, such as AI-based channel state information (CSI) prediction, AI-based beam management, and AI-based CSI feedback, and is playing an increasingly important role.

[0003]

[0004] However, the performance evaluation results of the network where the AI model is deployed indicate that the network performance is unstable and the network performance may deteriorate severely in some cases.

Summary of the Invention

[0004]

[0005] This application provides a method for switching or updating an AI model and a communication device, that is, a method and a communication device for monitoring an AI model, in order to improve the performance of the network where the AI model is deployed.

[0005]

[0006] According to a first aspect, a method for switching or updating an AI model is provided. The method may be executed by a first network element or may be executed by a chip or a circuit. This is not limited in this application. Hereinafter, the first network element is used as an example for explanation. The method may include the following.

[0006]

[0007] The first network element acquires the first information, which represents the estimation result of the first parameter, the estimation result of the first parameter is based on the channel measurement result, and the channel measurement result is the input to the AI ​​model.

[0007]

[0008] The first network element decides whether to switch or update the first AI model based on correspondence information and the first information, the first AI model is deployed in the first or second network element, the correspondence information indicates the correspondence between M AI models and N values ​​of the first parameter, where M is an integer greater than or equal to 1 and N is an integer greater than or equal to 1.

[0008]

[0009] In the technical solution of this application, the first network element can determine whether the operating environment of the AI ​​model has changed, for example, whether the movement speed of the network element to which the AI ​​model is deployed has changed, or whether the channel environment has changed, by monitoring the input indicators of the AI ​​model (e.g., the first parameters) of the AI ​​model. In this way, the first network element can decide whether to switch or update the AI ​​model to adapt to the change in the operating environment. This helps to mitigate the problem of the network to which the AI ​​model is deployed degrading due to changes in the operating environment of the AI ​​model.

[0009]

[0010] With respect to the first aspect, in a partial implementation of the first aspect, the first network element deciding whether to switch or update the first AI model based on correspondence information and first information includes the following:

[0010]

[0011] The first network element determines that the estimation result of the first parameter corresponds to the first value out of N values, and that the first value corresponds to the second AI model out of M AI models.

[0011]

[0012] The first network element decides to switch from the first AI model to the second AI model.

[0012]

[0013] In this implementation, the first network element can decide whether to switch the currently used AI model based on correspondence information and the estimation results of the first parameter. In this way, by monitoring the estimation results of the first parameter, the first network element can be aware of changes in the environment in which the AI ​​model is used in a timely manner and switch the AI ​​model in a timely manner to ensure the performance of the AI ​​model.

[0013]

[0014] With respect to the first embodiment, in a partial implementation of the first embodiment, the first network element deciding whether to switch or update the first AI model based on correspondence information and the first information includes the following:

[0014]

[0015] The first network element determines that the estimation result for the first parameter does not correspond to any of the N values.

[0015]

[0016] The first network element decides to update the first AI model.

[0016]

[0017] In this implementation, the first network element can determine whether to update the currently used AI model based on correspondence information and the estimation results of the first parameter. In this way, by monitoring the estimation results of the first parameter, the first network element can be aware of changes in the AI ​​model's operating environment in a timely manner, and if there is no AI model in the current model library that is suitable for the current operating environment, it can update the AI ​​model in a timely manner to ensure the AI ​​model's performance.

[0017]

[0018] With respect to the first embodiment, in a partial implementation of the first embodiment, the first AI model is deployed in a first network element, and the first network element is a terminal device.

[0018]

[0019] In this implementation, the terminal device can decide whether to switch or update the currently used first AI model by monitoring input indicators of the artificial intelligence model, such as changes in the terminal device's movement speed. This ensures the performance of the AI ​​model deployed on the terminal device in different usage environments.

[0019]

[0020] With respect to the first embodiment, in a partial implementation of the first embodiment, before the first network element obtains the first information, the method further includes the following:

[0020]

[0021] The first network element receives the second piece of information, and the second piece of information indicates the first parameter.

[0021]

[0022] In this implementation, if the correspondence information records the correspondence between the values ​​of multiple monitoring indicators and the AI ​​model, the first network element can retrieve the second information and determine the specified monitoring indicator, i.e., the first parameter. This implementation can establish a correspondence between the AI ​​model and multiple monitoring indicators, improving the diversity of the AI ​​model's deployment environment. Based on different deployment environments, the corresponding monitoring indicator is selected for monitoring, and the AI ​​model is switched or updated in a timely manner.

[0022]

[0023] With respect to the first embodiment, in a partial implementation of the first embodiment, after the first network element decides to switch the first AI model to the second AI model based on correspondence information and first information, the method further includes the following:

[0023]

[0024] The first network element sends first instruction information to the second network element, and the first instruction information indicates that the first network element is requesting to switch from the first AI model to the second AI model.

[0024]

[0025] Regarding the first aspect, in some implementations of the first aspect, the method further includes the following.

[0025]

[0026] A first network element receives a second AI model from a second network element.

[0026]

[0027] The first network element switches the first AI model to the second AI model.

[0027]

[0028] In this implementation, after determining to switch the AI model, the first network element sends a switching request to the second network element to obtain the second AI model from the second network element. As a result, the AI model can be switched in a timely manner, the performance of the AI model can be guaranteed, and the performance of the network in which the AI model is deployed can be stabilized.

[0028]

[0029] Regarding the first aspect, in some implementations of the first aspect, after the first network element determines to update the first AI model based on the corresponding information and the first information, the method further includes the following.

[0029]

[0030] The first network element sends second instruction information to the second network element, and the second instruction information is used to request to update the first AI model.

[0030]

[0031] [[ID=3o]]Regarding the first aspect, in some implementations of the first aspect, the method further includes the following.

[0031]

[0032] The first network element sends training data to the second network element, and the training data is used to update the first AI model. Optionally, the training data includes the estimation result of the first parameter.

[0032]

[0033] With respect to the first embodiment, in a partial implementation of the first embodiment, the method further includes the following:

[0033]

[0034] The first network element receives the third AI model from the second network element.

[0034]

[0035] The first network element updates the first AI model to the third AI model.

[0035]

[0036] In this implementation, after deciding to update the AI ​​model, the first network element sends an update request to the second network element, which then retrieves the new third AI model. As a result, the AI ​​model can be updated in a timely manner, ensuring the performance of the AI ​​model and stabilizing the performance of the network on which the AI ​​model is deployed.

[0036]

[0037] With respect to the first embodiment, in a partial implementation of the first embodiment, the acquisition of first information by a first network element includes the following:

[0037]

[0038] The first network element measures a reference signal to obtain channel measurement results.

[0038]

[0039] The first network element obtains an estimated result of the first parameter based on the channel measurement result, and the first information includes the estimated result of the first parameter.

[0039]

[0040] In this implementation, the UE monitors the input to the AI ​​model by measuring a reference signal to determine the value of a monitoring indicator (i.e., the first parameter), determines changes in the environment in which the AI ​​model is used, and makes a timely decision as to whether the AI ​​model needs to be switched or updated.

[0040]

[0041] With respect to the first embodiment, in a partial implementation of the first embodiment, before the first network element decides to switch or update the first AI model based on the correspondence information and the first information, the method further includes:

[0041]

[0042] The first network element obtains all or part of the correspondence information from the second or third network element.

[0042]

[0043] In this implementation, the first network element can pre-acquire and store correspondence information from the second or third network element, and then decide whether to switch or update the AI ​​model. The correspondence information can be flexibly configured. If the correspondence information changes, the updated settings can also be delivered to the first network element in a timely manner via the second or third network element.

[0043]

[0044] In the first embodiment, in a partial implementation of the first embodiment, the first network element is a network device, the first AI model is deployed on the second network element, and the second network element is a terminal device.

[0044]

[0045] With respect to the first embodiment, in a partial implementation of the first embodiment, the acquisition of first information by a first network element includes the following:

[0045]

[0046] A first network element receives first information from a second network element, the first information including an estimated result of a first parameter, or the first information including information used to determine the estimated result of the first parameter, the estimated result of the first parameter being based on channel measurement results obtained by measuring a reference signal on the side of the second network element.

[0046]

[0047] In this implementation, the first network element is a network-side device. The first network element obtains first information from the second network element, decides whether to switch or update the first AI model deployed on the second network element, and can also switch or update the AI ​​model deployed on the terminal device on the network side, thereby ensuring the performance of the AI ​​model on the terminal side and further ensuring the performance of the network on which the AI ​​model is deployed.

[0047]

[0048] With respect to the first embodiment, in a partial implementation of the first embodiment, if a first network element decides to switch the first AI model to a second AI model based on correspondence information and first information, the method further includes the following:

[0048]

[0049] The first network element sends the second AI model to the second network element.

[0049]

[0050] In this implementation, the first network element is the network-side device, and the second network element is the storage network element for the AI ​​model library on the terminal side. This reduces the storage overhead on the terminal side compared to storing the AI ​​model library on the terminal side.

[0050]

[0051] With respect to the first embodiment, if in a partial implementation of the first embodiment the first network element decides to update the first AI model based on correspondence information and first information, the method further includes the following:

[0051]

[0052] The first network element acquires the training data.

[0052]

[0053] The first network trains an AI model based on the training data to obtain a third AI model.

[0053]

[0054] The first network element sends the third AI model to the second network element.

[0054]

[0055] In this implementation, the first network element is the network-side device. If it is decided to update the AI ​​model deployed on the second network element, the first network element acquires a new AI model through training and provides the new AI model to the terminal side to ensure the performance of the AI ​​model on the terminal side.

[0055]

[0056] With respect to the first embodiment, in a partial implementation of the first embodiment, after the first network element decides to switch the first AI model to the second AI model based on correspondence information and first information, the method further includes the following:

[0056]

[0057] The first network element sends first instruction information to the second network element, and the first instruction information indicates that the first network element is requesting to switch from the first AI model to the second AI model.

[0057]

[0058] The first network element receives third instruction information from the second network element, and the third instruction information instructs the first network element to switch the first AI model.

[0058]

[0059] The first network element switches the first AI model to the second AI model based on the third instruction information.

[0059]

[0060] In this implementation, the first and second network elements can collaboratively switch between AI models. This is primarily applicable to switching between AI models in bilateral models. In this way, timely switching of AI models in bilateral model application scenarios can be ensured, and the stability of the performance of the network where the bilateral model is deployed can be guaranteed.

[0060]

[0061] With respect to the first embodiment, in a partial implementation of the first embodiment, after the first network element decides to update the first AI model based on correspondence information and first information, the method further includes the following:

[0061]

[0062] The first network element transmits first instruction information to a second network element, wherein the first instruction information indicates that the first network element is requesting to switch from a first AI model to a second AI model.

[0062]

[0063] The first network element receives third instruction information from the second network element, and the third instruction information instructs the first network element not to switch the first AI model.

[0063]

[0064] In this implementation, if it is not possible to switch AI models in a bilateral model, for example, if there is no AI model to switch to, neither the first nor the second network element will switch AI models.

[0064]

[0065] With respect to the first embodiment, in a partial implementation of the first embodiment, after the first network element decides to update the first AI model based on the correspondence information and the first information, the method further includes the following:

[0065]

[0066] The first network element transmits the second instruction information and the estimation result of the first parameter to the second network element, and the second instruction information indicates that the first network element is requesting that the first AI model be updated.

[0066]

[0067] The first network element receives fourth instruction information from the second network element, and the fourth instruction information instructs the first network element to update the first AI model.

[0067]

[0068] The first network element updates the first AI model to the third AI model based on the fourth instruction information, and the third AI model is acquired through training based on the estimation results of the first parameters.

[0068]

[0069] In this implementation, the first and second network elements can collaboratively update the AI ​​model to ensure timely updates of the AI ​​model in bilateral model application scenarios and to ensure the stability of the network performance on which the bilateral model is deployed.

[0069]

[0070] With respect to the first embodiment, in a partial implementation of the first embodiment, after the first network element decides to update the first AI model based on correspondence information and first information, the method further includes the following:

[0070]

[0071] The first network element transmits the second instruction information and the estimation result of the first parameter to the second network element, and the second instruction information indicates that the first network element is requesting that the first AI model be updated.

[0071]

[0072] The first network element receives fourth instruction information from the second network element, and the fourth instruction information instructs the first network element not to update the first AI model.

[0072]

[0073] In this implementation, if there is no training data available to use for model updates, neither the first nor the second network element in the bilateral model will update the AI ​​model.

[0073]

[0074] With respect to the first embodiment, in a partial implementation of the first embodiment, the AI ​​model is applied to CSI prediction, beam management, or CSI feedback.

[0074]

[0075] With respect to the first embodiment, in a partial implementation of the first embodiment, the first parameter is: The movement speed of the terminal device; Channel signal-to-interference ratio plus noise ratio (SINR); or A parameter that reflects the degree of non-line-of-sight (NLOS) in the channel; Includes one or more of the following.

[0075]

[0076] In this implementation, the performance of the AI ​​model changes with the environment in which it is used. Thus, different primary parameters can be selected and monitored based on different usage environments. The most important indicators affecting the AI ​​model's performance are monitored, resulting in more accurate decisions to switch or update the AI ​​model in a timely manner. This helps ensure the performance of both the AI ​​model and the network on which it is deployed.

[0076]

[0077] With respect to the first embodiment, in some implementations of the first embodiment, the reference signal is a channel state information-reference signal (CSI-RS).

[0077]

[0078] With respect to the first embodiment, in a partial implementation of the first embodiment, the fourth AI model is deployed on the first network element, and the fourth AI model and the first AI model are used in a matching manner.

[0078]

[0079] The first network element deciding whether to switch the first AI model based on correspondence information and first information includes the following:

[0079]

[0080] The first network element determines that the estimation result of the first parameter corresponds to the first value out of N values, and that the first value corresponds to the second AI model out of M AI models.

[0080]

[0081] The first network element determines whether the W stored AI models deployed in the first network element include an AI model that matches the second AI model, where W is an integer greater than or equal to 1.

[0081]

[0082] The first network element decides whether to switch the first AI model based on the judgment result.

[0082]

[0083] With respect to the first embodiment, in a partial implementation of the first embodiment, the determination by a first network element whether to switch the first AI model based on the determination result includes the following:

[0083]

[0084] The first network element decides to switch the first AI model to the second AI model if the determination result is that W AI models include a fifth AI model that matches the second AI model.

[0084]

[0085] The method further includes the following:

[0085]

[0086] The first network element transmits the third instruction information to the second network element, and the third instruction information indicates switching the first AI model to the second AI model.

[0086]

[0087] The first network element switches the fourth AI model to the fifth AI model.

[0087]

[0088] With respect to the first embodiment, in a partial implementation of the first embodiment, the determination by a first network element whether to switch the first AI model based on the determination result includes the following:

[0088]

[0089] The first network element decides not to switch the first AI model if the determination result is that none of the W AI models match the second AI model.

[0089]

[0090] With respect to the first embodiment, in a partial implementation of the first embodiment, a fourth AI model is deployed to a first network element, and the fourth AI model and the first AI model are used in a matching scheme.

[0090]

[0091] The first network element deciding whether to update the first AI model based on the correspondence information and the first information includes the following:

[0091]

[0092] The first network element determines that the estimation result for the first parameter does not correspond to any of the N values.

[0092]

[0093] The first network element determines whether it is possible to obtain training data to be used to update the first AI model, and based on the determination result, it decides whether to update the first AI model.

[0093]

[0094] With respect to the first embodiment, in a partial implementation of the first embodiment, if it is determined that it is possible to obtain training data to be used to update the first AI model, the first network element decides to update the first AI model.

[0094]

[0095] The method further includes the following:

[0095]

[0096] The first network element transmits the fourth instruction information to the second network element, and the fourth instruction information instructs the first network element to update the first AI model.

[0096]

[0097] The first network element updates the fourth AI model to the sixth AI model.

[0097]

[0098] With respect to the first embodiment, in a partial implementation of the first embodiment, determining whether a first network element updates the first AI model based on the determination result includes the following:

[0098]

[0099] If the first network element determines that there is no training data available to update the first AI model, it decides not to update the first AI model.

[0099]

[0100] In the implementation described above, the first network element makes a timely decision to switch or update the AI ​​models in the bilateral model (specifically, the fourth AI model deployed in the first network element and the first AI model deployed in the second network element) based on the estimation results of the first parameters provided by the second network element. This helps to ensure the performance of the AI ​​models and the performance of the network in which the AI ​​models are deployed.

[0100]

[0101] A second aspect provides a method for switching or updating an AI model. The method may be performed by a network element or by a chip or circuit. This is not limited to the present application. A second network element will be used as an example for illustrative purposes below.

[0101]

[0102] The methods may include the following:

[0102]

[0103] The second network element receives channel measurement results fed back from the first network element.

[0103]

[0104] The second network element, based on the fourth AI model, reconstructs the feedbacked channel measurement results to obtain the reconstructed channel measurement results.

[0104]

[0105] The second network element decides whether to switch or update the fourth AI model based on the feedbacked channel measurement results and the reconstructed channel measurement results.

[0105]

[0106] In this technical solution, the second network element can make a timely decision to switch or update the AI ​​model in the bilateral model based on the monitoring results of the bilateral model's intermediate performance indicators. This helps to ensure the performance of the AI ​​model in the bilateral model, as well as the performance of the network on which the bilateral model is deployed.

[0106]

[0107] Optionally, the feedback channel measurement results are obtained by the first network element by processing the channel measurement results obtained through measurement based on the first AI model.

[0107]

[0108] With respect to the second aspect, in a partial implementation of the second aspect, the step of deciding whether to switch or update the fourth AI model based on the feedbacked channel measurement results and the restored channel measurement results includes the following:

[0108]

[0109] The second network element determines the value of the error indicator for the validation dataset based on the feedbacked channel measurement results, the reconstructed channel measurement results, and the validation dataset.

[0109]

[0110] If the error indicator value does not meet the specified conditions, the second network element decides not to switch or update the fourth AI model.

[0110]

[0111] Alternatively, if the error indicator value meets the specified conditions, the second network element decides to switch to or update the fourth AI model.

[0111]

[0112] Optionally, the specified condition may include: the error indicator value being greater than or equal to a specified threshold T.

[0112]

[0113] For example, the error indicator is the GCS. It is known that the GCS measures the difference between two individual vectors using the cosine value of the included angle between those two vectors in vector space. If the cosine value is close to 1, it indicates that the included angle between the two vectors is close to 0 degrees, and the two vectors are more similar. If the cosine value is close to 0, it indicates that the included angle between the two vectors is close to 180 degrees, and the similarity between the two vectors is low. In this application, the feedbacked channel measurement results are reconstructed based on a fourth AI model, and the reconstructed channel measurement results are obtained. The GCS may be obtained through calculations based on the reconstructed channel measurement results and a validation dataset. Specifically, the validation dataset is the channel measurement results obtained through measurements on the side of the first network element and the labels of the fourth AI model. The GCS is obtained by comparing the reconstructed channel measurement results with the channel measurement results obtained through the measurements. A high GCS value indicates that the restored channel measurement result is closer to the channel measurement result obtained through the measurement; in other words, the fourth AI model has high accuracy in restoring the feedbacked channel measurement result. A threshold T is set for GCS. For example, threshold T = 0.95. If the GCS value is 0.95 or higher, it indicates that the accuracy of the fourth AI model in the current usage environment meets the requirements. In other words, the fourth AI model is suitable for the current usage environment. Therefore, switching or updating may not be necessary. Conversely, if the GCS value is less than 0.95, it indicates that the accuracy of the fourth AI model in the current usage environment does not meet the requirements. Therefore, the second network element needs to switch or update the fourth AI model.

[0113]

[0114] With respect to the second aspect, in a partial implementation of the second aspect, the decision by a second network element to switch or update a fourth AI model when the value of an error indicator meets a specified condition includes the following:

[0114]

[0115] The second network element decides to switch to the fourth AI model if the value of the error indicator meets the specified conditions, and the Q AI models stored in the second network element include the fifth AI model, and the complexity of the fifth AI model is higher than the complexity of the currently used fourth AI model.

[0115]

[0116] Alternatively, the second network element decides to update the fourth AI model if the error indicator value meets the specified conditions and the Q AI models stored in the second network element do not have a higher complexity than the currently used fourth AI model.

[0116]

[0117] With respect to the second aspect, in a partial implementation of the second aspect, if a second network element decides to switch or update a fourth AI model, the method further includes:

[0117]

[0118] The second network element sends a switch request or an update request to the first network element, where a switch request indicates a request to switch to the fourth AI model, and an update request is used to request an update to the fourth AI model.

[0118]

[0119] The second network element receives information from the first network element indicating that a switchover should be performed, or information indicating that an update should be performed.

[0119]

[0120] The second network element switches the fourth AI model to the fifth AI model based on information indicating that a switch should be performed. Alternatively, the second network element updates the fourth AI model to the sixth AI model based on information indicating that an update should be performed.

[0120]

[0121] A third aspect provides a method for switching or updating an AI model. The method may be performed by a network element or by a chip or circuit. This is not limited to the present application. A first network element is used as an example for illustrative purposes below. This method corresponds to the method in the second aspect.

[0121]

[0122] The methods may include the following:

[0122]

[0123] The first network element receives a switch request from the second network element, and the switch request indicates that the second network element is requesting a switch to an AI model.

[0123]

[0124] The first network element determines whether to switch the currently used first AI model based on the correspondence information and the switching request. The correspondence information indicates the correspondence between Q AI models and Q complexity information for each AI model, the complexity information corresponding to the first AI model indicates the first complexity level, and Q is an integer greater than or equal to 1.

[0124]

[0125] In the technical solution, the first network element decides whether to switch the currently used AI model based on a switch request. Since the switch request is sent to the first network element by the second network element based on monitoring results of the bilateral model's intermediate performance indicators, it becomes possible to decide in a timely manner whether to switch the AI ​​model in the bilateral model. This helps to ensure the performance of the AI ​​model in the bilateral model, as well as the performance of the network on which the bilateral model is deployed.

[0125]

[0126] With respect to the third aspect, in a partial implementation of the third aspect, the determination by the first network element whether to switch the currently used first AI model based on correspondence information and a switching request includes the following:

[0126]

[0127] The first network element determines, based on the switching request, whether the Q AI models include an AI model corresponding to a second complexity level, where the second complexity level is higher than the first complexity level.

[0127]

[0128] If Q AI models include a second AI model corresponding to a second complexity level, the first network element decides to switch the first AI model to the second AI model.

[0128]

[0129] The method further includes the following:

[0129]

[0130] The first network element sends information to the second network element indicating that a switchover is to be performed.

[0130]

[0131] The first network element switches the first AI model to the second AI model.

[0131]

[0132] With respect to the third aspect, in a partial implementation of the third aspect, the determination by the first network element whether to switch the currently used first AI model based on correspondence information and a switching request includes the following:

[0132]

[0133] The first network element determines, based on a switch request, whether the Q AI models include an AI model corresponding to a second complexity level, where the second complexity level is higher than the first complexity level.

[0133]

[0134] If none of the Q AI models include a second AI model corresponding to a second complexity level, the first network decides not to switch to the first AI model.

[0134]

[0135] The method further includes the following:

[0135]

[0136] The first network element sends instruction information to the first network element indicating that it will not perform a switch.

[0136]

[0137] Optionally, with respect to a third aspect, in some implementations of the third aspect, the first network element receives an update request from the second network element, and the update request indicates that the second network element is requesting an update to the AI ​​model.

[0137]

[0138] The first network element decides whether to update the first AI model based on the update request.

[0138]

[0139] For a specific implementation of how the first network element determines whether to update the first AI model based on an update request, please refer to the implementation where the first network element determines whether to update the first AI model in the first aspect based on correspondence information and the first information. The process is similar, and the details will not be explained again.

[0139]

[0140] For the technical effects of other implementations in the second and third embodiments, please refer to the explanation of the technical effects of the implementation of the bilateral model in the first embodiment. Further details will not be explained again.

[0140]

[0141] According to a fourth aspect, the present application provides a method for switching or updating an AI model, the method including:

[0141]

[0142] The second network element transmits correspondence information to the first network element, where the correspondence information indicates the relationship between M AI models and N values ​​of the first parameter, and M is an integer greater than or equal to 1, and N is an integer greater than or equal to 1.

[0142]

[0143] Alternatively, the correspondence information represents the relationship between Q AI models and R complexity information for those AI models, where the R complexity information represents different complexity levels, and both Q and R are integers greater than or equal to 1.

[0143]

[0144] With respect to the fourth aspect, in a partial implementation of the fourth aspect, the method further includes the following:

[0144]

[0145] The second network element transmits second information to the first network element, and the second information represents the first parameter.

[0145]

[0146] According to the fifth aspect, the present application provides a communication device. The communication device may include modules configured to correspond one-to-one with and perform the methods / operations / steps / actions described in the first to fourth aspects. The modules may be hardware circuits, software, or implemented by hardware circuits in combination with software. The communication device may be a first network element or a second network element.

[0146]

[0147] In implementation, the communication device is a communication device. For example, the communication device may include a communication unit and / or a processing unit. The communication unit may be a transceiver or an input / output interface, and the processing unit may be at least one processor. Optionally, the transceiver may be a transceiver circuit. Optionally, the input / output interface may be an input / output circuit.

[0147]

[0148] In another implementation, the device is a chip, chip system, or circuit used in a communication device. If the device is a chip, chip system, or circuit used in a terminal device, the communication unit may be an input / output interface, interface circuit, output circuit, input circuit, pin, associated circuit, or similar in the chip, chip system, or circuit, and the processing unit may be at least one processor, processing circuit, logic circuit, or similar.

[0148]

[0149] In one example, the communication device is a terminal device, or the communication device is a chip, chip system, circuit, or similar located within a terminal device. In another example, the communication device is an access network device, or the communication device is a chip, chip system, circuit, or similar located within an access network device.

[0149]

[0150] According to the sixth aspect, the present application provides a communication device. The communication device includes a processor, which is configured to execute computer programs or instructions stored in memory to perform any method provided in the first or third aspect and any implementation of the first or third aspect. Optionally, the communication device further includes memory. The communication device may also be a first network element.

[0150]

[0151] According to the seventh aspect, the present application provides a communication device. The communication device includes a processor and a communication interface and is configured to perform a method provided in any of the second or fourth aspects and implementations of the second or fourth aspects. For example, the communication interface may be a transceiver, hardware circuitry, a bus, a module, a pin, or another type of communication interface.

[0151]

[0152] According to the eighth aspect, the present application further provides a computer program. When the computer program is executed on a computer, the computer becomes capable of performing any method provided in the first through fourth aspects or in any implementation of the first through fourth aspects.

[0152]

[0153] According to the ninth aspect, the present application further provides a computer program product including instructions. When the instructions are executed on a computer, the computer is able to perform any of the methods provided in the first through fourth aspects or implementations of the first through fourth aspects.

[0153]

[0154] According to the tenth aspect, the present application further provides a computer-readable storage medium for storing computer programs or instructions. When the computer programs or instructions are executed on a computer, the computer becomes capable of performing any method provided in any of the first through fourth aspects or implementations thereof.

[0154]

[0155] According to the eleventh aspect, the present application further provides a chip configured to read a computer program stored in memory and to perform any method according to the first through fourth aspects or any implementation of the first through fourth aspects. Alternatively, the chip includes circuitry configured to perform any method provided in the first through fourth aspects or any implementation of the first through fourth aspects.

[0155]

[0156] According to the twelfth aspect, the present application further provides a chip system. The chip system includes a processor configured to support the device in implementing any of the methods provided in the first to fourth aspects or implementations of the first to fourth aspects. In possible designs, the chip system further includes memory, which is configured to store programs and data required by the device. The chip system may include a chip, or it may include a chip and other separate components.

[0156]

[0157] According to the 13th aspect, the present application provides a communication system including the first network element and the second network element described above.

[0157]

[0158] For example, one of the first and second network elements is a terminal device, and the other is a network device such as an access network device.

[0158]

[0159] For the technical effects of the solutions provided in any of the second through thirteenth embodiments or implementations of the second through thirteenth embodiments, please refer to the corresponding description in the first embodiment. Further details will not be provided again. [Brief explanation of the drawing]

[0159] [Figure 1]

[0160] Figure 1 is a diagram of a communication system to which the embodiment of this application can be applied. [Figure 2]

[0161] Figure 2 is a schematic flowchart of the method for switching or updating AI models according to this application. [Figure 3]

[0162] Figure 3 shows an example of a method for switching or updating an AI model according to this application. [Figure 4]

[0163] Figure 4 shows an example of a method for switching or updating an AI model according to this application. [Figure 5]

[0164] Figure 5 shows an AI-based CSI prediction. [Figure 6]

[0165] Figure 6 shows an example of the application of the AI ​​model switching or updating method described in this application. [Figure 7]

[0166] Figure 7 shows an example of the application of the AI ​​model switching or updating method described in this application. [Figure 8]

[0167] Figure 8 is a diagram of the AI-based CSI feedback process. [Figure 9]

[0168] Figure 9 is a schematic flowchart of the AI ​​model switching method related to this application. [Figure 10]

[0169] Figure 10 is a schematic flowchart of the AI ​​model update method according to this application. [Figure 11]

[0170] Figure 11 shows an example of the application of the method for switching or updating AI models in AI-CSI feedback according to this application. [Figure 12]

[0171] Figure 12 is a schematic flowchart of the method for switching or updating AI models according to this application. [Figure 13]

[0172] Figure 13 is a block diagram of the communication device according to this application. [Figure 14]

[0173] Figure 14 is a block diagram of the communication device according to this application. [Modes for carrying out the invention]

[0160]

[0174] The technical solution in the embodiment of this application will be described below with reference to the attached drawings.

[0161]

[0175] The technical solutions provided in this application may be applicable to a variety of communication systems. For example, the communication systems may be fourth-generation (4G) communication systems (e.g., long-term evolution (LTE) systems), fifth-generation (5G) communication systems, worldwide interoperability for microwave access (WiMAX) or wireless local area network (WLAN) systems, satellite communication systems, or future communication systems, such as 6G communication systems, or integrated systems of multiple systems. 5G communication systems are sometimes referred to as new radio (NR) systems.

[0162]

[0176] Network elements within a communication system may transmit signals to other network elements or receive signals from other network elements. Signals may be information, signaling, data, or similar. Network elements may be replaced by entities, network entities, devices, communication devices, communication modules, nodes, communication nodes, or similar entities. In this application, network elements are used as illustrative examples.

[0163]

[0177] The communication system to which this application applies may include a first network element and a second network element, and optionally further include a third network element. The number of first network elements, second network elements, and third network elements is not limited.

[0164]

[0178] In the embodiments of this application, the terminal device is a user-side entity configured to receive or transmit signals, such as a mobile phone. The terminal device is a handheld device with wireless connectivity, another processing device connected to a wireless modem, an in-vehicle device, or similar. The terminal device may be portable, pocket-sized, handheld, computer-integrated, or in-vehicle mobile device. Terminal device 120 can be widely used in a variety of scenarios, such as cellular communication, Wi-Fi systems, D2D, V2X, peer-to-peer (P2P), M2M, machine-type communication (MTC), Internet of Things (IoT), virtual reality (VR), augmented reality (AR), industrial control, autonomous driving, telemedicine, smart grids, smart furniture, smart offices, smart wearables, smart transportation, smart cities, unmanned aerial vehicles, robots, remote sensing, passive sensing, positioning, navigation and tracking, autonomous delivery and mobility, and similar applications.Some examples of communication devices 120 include user equipment (UE) in the GPP standard, stations (STA) in Wi-Fi systems, fixed devices, mobile devices, handheld devices, wearable devices, cellular phones, smartphones, session initiation protocol (SIP) phones, notebook computers, personal computers, smartbooks, vehicles, satellites, global positioning system (GPS) devices, target tracking devices, unmanned aerial vehicles, helicopters, aircraft, ships, remote control devices, smart home devices, industrial devices, personal communication service (PCS) phones, wireless local loop (WLL) stations, personal digital assistants (PDAs), wireless network cameras, tablet computers, palmtop computers, mobile internet devices (MIDs), wearable devices such as smartwatches, virtual reality (VR) devices, and augmented reality (AUGD) devices. Reality (AR) devices, wireless terminals in industrial control, terminals in vehicle internet systems, wireless terminals in autonomous driving, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, such as smart refueling stations, terminal devices in high-speed rail, and wireless terminals in smart homes, such as smart speakers, smart coffee machines, and smart printers.The terminal device 120 may be a wireless device in the various scenarios described above, or a device located within a wireless device, such as a communication module, modem, or chip within the aforementioned device. The terminal device may also be referred to as a terminal, user equipment (UE), mobile station (MS), mobile terminal (MT), or similar. Alternatively, the terminal device may be a terminal device in a future wireless communication system. Furthermore, the terminal device may further include a location-based device, such as an automated guided vehicle (AGV) or a device having similar functionality. The specific technologies and device forms used by the terminal device are not limited to the embodiments of this application. For ease of explanation, examples in which the terminal device is a terminal or UE are used below for illustrative purposes.

[0165]

[0179] In this application, the communication device configured to implement terminal device functions may be a terminal device, a terminal device having some of the functions of the aforementioned communication device, or a device capable of supporting the implementation of the aforementioned terminal device functions, such as a chip system. The device may be installed in the terminal device or used in a matching manner with the terminal device. In this application, the chip system may include a chip, or it may include a chip and other separate components.

[0166]

[0180] A network device may be a device that provides wireless communication functionality and services, may communicate with terminal devices, and is typically located on the network side. Network devices are sometimes called access network devices or wireless access network devices. For example, a network device may be a base station. For example, the access network devices in the embodiments of this application include, but are not limited to, the following: next-generation base stations (gNodeB, gNB) in 5th generation (5G) communication systems, base stations in 6th generation (6G) mobile communication systems, base stations in future mobile communication systems, access points (AP) in wireless fidelity (Wi-Fi) systems, evolved NodeB (eNB) in long-term evolution (LTE) systems, radio network controllers (RNC), NodeB (NB), base station controllers (BSC), home base stations (e.g., home evolved NodeB, or home NodeB, HNB), baseband units (BBU), transmission reception points (TRP), transmitting points (TP), and base transceiver base stations. This includes stations (BTS), satellites, unmanned aerial vehicles, and similar devices. In a network structure, network devices may include central unit (CU) nodes, distributed unit (DU) nodes, RAN devices including CU nodes and DU nodes, or RAN devices including control plane CU nodes, user plane CU nodes, and DU nodes.Alternatively, network devices may be radio controllers, relay stations, in-vehicle devices, wearable devices, and similar devices in cloud radio access network (CRAN) scenarios. Furthermore, base stations may be macro base stations, micro base stations, relay nodes, donor nodes, or a combination thereof. Alternatively, base stations may be communication modules, modems, or chips deployed in the aforementioned devices or equipment. Alternatively, base stations may be mobile switching centers, devices performing base station functions in device-to-device (D2D), vehicle-to-everything (V2X), or machine-to-machine (M2M) communications, network-side devices in 6G networks, devices performing base station functions in future communication systems, and similar devices. Base stations may support networks with the same or different access technologies, but are not limited to this. Base stations may be fixed or mobile. For example, a helicopter or unmanned aerial vehicle may be configured as a mobile base station, and one or more cells may move based on the location of the mobile base station. In another example, a helicopter or unmanned aerial vehicle may be configured as a device for communicating with another base station.

[0167]

[0181] In this application, the device configured to implement the functions of the aforementioned network device may be an access network device, a network device having some functions of an access network, or a device capable of supporting the implementation of access network functions, such as a chip system, hardware circuitry, software module, or a combination of hardware circuitry and software module. The device may be installed in an access network device or used in a matching manner with an access network device. In the method described in this application, an example is used in which the communication device configured to implement the functions of an access network device is an access network device.

[0168]

[0182] Figure 1 is a diagram of a communication system 100 to which embodiments of the present application are applicable. As shown in Figure 1, the wireless communication system includes a wireless access network 100. The wireless access network 100 may be a next-generation (e.g., 6G or higher) wireless access network or a conventional wireless access network (e.g., 5G, 4G, 3G). One or more terminal devices (120a to 120j, collectively referred to as 120) may be interconnected or connected to one or more network devices (110a and 110b, collectively referred to as 110) within the wireless access network 100. Figure 1 is merely an illustrative diagram. The wireless communication system may further include other devices, for example, core network devices, wireless relay devices, and / or wireless backhaul devices not shown in Figure 1.

[0169]

[0183] In actual applications, a wireless communication system may include multiple network devices (also called access network devices) or multiple terminal devices, but is not limited to these. One network device may serve one or more terminal devices. One terminal device may access one or more network devices. The number of terminal devices and network devices included in the wireless communication system is not limited to the present embodiments of this application.

[0170]

[0184] Optionally, the communication system further includes at least one AI node, which is not shown in Figure 1.

[0171]

[0185] Optionally, AI nodes may be deployed in one or more of the following locations within the communication system: access network devices, terminal devices, core network devices, or similar. Alternatively, AI nodes may be deployed independently, for example, in a location other than any of the aforementioned devices, such as a host or cloud server in an over-the-top (OTT) system. AI nodes can communicate with other devices within the communication system, which may be, for example, one or more of the following: network devices, terminal devices, network elements of the core network, or similar.

[0172]

[0186] Optionally, an AI node can be configured to perform AI-related operations. For example, AI-related operations may include one or more of the following: model failure testing, model performance testing, model training, data acquisition, or similar.

[0173]

[0187] For example, a network device can transfer data related to an AI model, reported by a terminal device, to an AI node, which then performs AI-related operations. In another example, a network device or terminal device can transfer data related to an AI model to an AI node, which then performs AI-related operations. In yet another example, an AI node can send the output of an AI-related operation, e.g., a trained neural network model, model evaluation results, model test results, or one or more of the same, to a network device and / or a terminal device. For example, an AI node can directly send the output of an AI-related operation to a network device and a terminal device. In yet another example, an AI node can send the output of an AI-related operation to a terminal device via a network device. In yet another example, an AI node can send the output of an AI-related operation to a network device via a terminal device.

[0174]

[0188] It should be understood that the number of AI nodes is not limited in this application. For example, if there are multiple AI nodes, they may be divided based on their function. For example, different AI nodes may be responsible for different functions.

[0175]

[0189] It will be further understood that an AI node may be an independent device, may be integrated within the same device to implement different functions, may be a network element within a hardware device, may be a software function running on dedicated hardware, or may be a virtualized function instantiated on a platform (e.g., a cloud platform). The specific form of an AI node is not limited in this application.

[0176]

[0190] To facilitate understanding, the relevant technologies or concepts in this application are briefly explained below.

[0177]

[0191] AI Model: An AI model is an algorithm or computer program capable of realizing AI functionality. An AI model represents a mapping relationship between the model's inputs and outputs, or it is a functional model that maps inputs of a specific dimension to outputs of a specific dimension. The parameters of a functional model can be obtained through machine learning training. For example, f(x) = ax 2 +b is a quadratic function model and can be considered as an AI model. A and b are parameters of the AI ​​model, and a and b may be obtained through machine learning training. For example, the AI ​​models referred to in the following embodiments of this application are not limited to neural networks, linear regression models, decision tree models, support vector machines (SVMs), Bayesian networks, Q-learning models, or other machine learning (ML) models.

[0178]

[0192] AI model design primarily includes a data acquisition phase (e.g., collection of training data and / or inference data), a model training phase, and a model inference phase. An inference result application phase may also be included. In the aforementioned data acquisition phase, a data source is used to provide training datasets and inference data. In the model training phase, the training data provided by the data source is analyzed or trained to obtain an AI model. Obtaining an AI model through learning using model training nodes is equivalent to obtaining a mapping relationship between the inputs and outputs of the AI ​​model through learning using the training data. In the model inference phase, the artificial intelligence model obtained through training in the model training phase is used to perform inference based on the inference data provided by the data source, and the inference result is obtained. This phase may be understood as follows: inference data is input to the AI ​​model, and an output is obtained through the AI ​​model. This output is the inference result. The inference result can indicate configuration parameters used (executed) by the execution object, and / or operations performed by the execution object. The inference result is published in the inference result application phase. For example, the inference result may be uniformly planned by the execution (actor) entity. For example, an execution entity can send inference results to one or more execution objects (e.g., a core network device, an access network device, or a terminal device) for execution. In another example, an execution entity can further feed back the performance of the AI ​​model to a data source to facilitate subsequent model updates of the AI ​​model.

[0179]

[0193] It will be understood that AI models may be implemented using hardware circuits, or using software, or a combination of software and hardware. This is not limited to these. Non-limiting examples of software include: program code, programs, subprograms, instructions, instruction sets, code, code segments, software modules, applications, software applications, or similar.

[0180]

[0194] It should be noted that in the embodiments of this application, “indication” may include direct, indirect, explicit, and implicit indications. If one indication indicates A, it may be understood that the indication carries A, and the indication may directly or indirectly indicate A. Indirect indication may mean that the purpose of using the indication is achieved by directly indicating B and the correspondence between B and A using the indication. The correspondence between B and A may be predefined in the protocol, pre-stored, or obtained by using a configuration between network elements.

[0181]

[0195] The methods for switching or updating AI models provided in this application will be described in detail below with reference to the attached drawings. The embodiments provided in this application may be applied to the wireless communication system shown in Figure 1, but are not limited thereto.

[0182]

[0196] In the following embodiments, the first, second, or third network element may be a network element that integrates one or more functions from among: an AI model inference network element, an AI model library storage network element, or an AI model training network element. This relates to specific implementations of different embodiments.

[0183]

[0197] Optionally, in some embodiments, the wireless communication system may include one or more of the following network elements, depending on the particular embodiment: a first network element, a second network element, and a third network element. These network elements may be logically deployed separately, but physically they may be deployed in the same network element or different network elements. This is not limited to these. For example, the first network element may be a module in a terminal device, and the second network element may be another module in the terminal device, and the modules may be located in the application layer. Alternatively, the first network element may be a module in an access network device, and the second network element may be another module in an access network device, and the modules may be located in the application layer. Alternatively, the first network element may be a terminal device or an access network device, and the second network element may be a host or cloud device in an over-the-top (OTT) system (or a system that provides various application services to a user over the internet). In another example, the first network element is a terminal device, and the second network element is an access network device. In another example, the first network element is an access network device, and the second network element is a terminal device.

[0184]

[0198] For example, the first network element may be a terminal device or a component of a terminal device (e.g., a chip or circuit). The second network element may be a network device or a component of a network device (e.g., a chip or circuit), or the second network element may be a host or cloud storage device in the OTT system or a component of a host or cloud storage device (e.g., a chip or circuit).

[0185]

[0199] The technical solutions provided in this application are described below.

[0186]

[0200] Figure 2 is a schematic flowchart of the method for switching or updating AI models according to this application.

[0187]

[0201] 210: The first network element obtains the first piece of information.

[0188]

[0202] The first piece of information shows the estimation result for the first parameter.

[0189]

[0203] The first piece of information indicating the estimation result of the first parameter includes the following:

[0190]

[0204] The first piece of information is the estimation result of the first parameter; in other words, the first network element obtains the estimation result of the first parameter.

[0191]

[0205] Alternatively, a correspondence may exist between the first piece of information and the estimated result of the first parameter, where the first piece of information may indicate the estimated result of the first parameter.

[0192]

[0206] The estimation result of the first parameter is obtained based on the channel measurement result, which is the input to the AI ​​model. In other words, the estimation result of the first parameter is an input indicator of the AI ​​model. In other words, the first parameter is related to the channel measurement result, which is the input to the AI ​​model. Therefore, the estimation result of the first parameter may reflect changes in the input to the AI ​​model. Changes in the input to the AI ​​model may be learned by monitoring the estimation result of the first parameter. For example, changes in the environment in which the AI ​​model is used may be learned further. For example, changes in the environment in which the AI ​​model is used may be one or more of the following: changes in channel state, changes in the movement speed of the AI ​​model's inference network elements, and so on.

[0193]

[0207] For example, in some application scenarios, the performance of an AI model is greatly influenced by the movement speed of the AI ​​model's inference network elements. If the movement speed of the inference network elements changes significantly, the generalization performance of the AI ​​model deteriorates considerably, and the performance of the network on which the AI ​​model is deployed also decreases or deteriorates accordingly.

[0194]

[0208] Optionally, the first parameter may include one or more of the following: channel time domain quality indicator, channel frequency domain quality indicator, the number of paths whose energy is greater than k times the energy of the first path in the channel impulse response (CIR), average power of multiple sampling points, LOS probability, a parameter reflecting the degree of channel NLOS, signal-to-interference plus noise ratio (SINR), reference signal received power (RSRP), received signal strength indication (RSSI), interference level, or the movement speed of the inference network elements on which the AI ​​model is deployed. It will be understood that these indicators may be obtained by measuring the reference signal or by performing corresponding processing on the measurement results of the reference signal. The specific processing process is not limited here. k is a number greater than 0. Optionally, k is a number greater than 0 and less than 1. For example, the parameter reflecting the degree of channel NLOS may be, for example, the Rician factor. The Rice factor represents the power ratio between LOS (Loss of Situation) and NLOS (Non-Limited Situation) paths within a set of paths. In an LOS scenario, the energy of the LOS path is higher than the total energy of the NLOS path. Generally, the Rice factor is used to define the ratio between LOS path power and NLOS path power. Here, NLOS path power represents the sum of the power of all NLOS paths. An NLOS scenario includes NLOS paths, but does not necessarily include LOS paths. From the above, it can be seen that, generally, a higher Rice factor indicates a higher LOS path energy relative to NLOS path energy, and a lower degree of NLOS.

[0195]

[0209] Optionally, CIR may be replaced with any of the following: time-aligned CIR, cross-correlation sequences of multiple CIR sequences, normalized CIR, or similar. This is not limited. The number of sampling points is not limited when the estimation result of the first parameter is determined based on CIR.

[0196]

[0210] For example, the first network element performs channel measurements based on a reference signal to obtain a channel frequency domain response (CFR) and obtains a channel frequency domain quality indicator through calculations based on the CFR. In this embodiment of the present application, when the first parameter is calculated based on the CFR, the number of CFR bandwidths, subbands, and ports are not limited. The CFR may also be replaced with a normalized CFR. Further details are not described below.

[0197]

[0211] For example, the channel frequency domain quality indicator may be a Doppler frequency shift or a Doppler frequency. The Doppler shift may reflect the movement speed of the inference network elements on which the AI ​​model is deployed. For example, the movement speed of the inference network elements may be estimated based on the Doppler frequency shift obtained by measurement, and a decision may be made to switch or update the currently used AI model based on the movement speed. In particular, in AI model application scenarios that are sensitive to movement speed, monitoring changes in the movement speed of the inference network elements allows for a timely decision to switch or update the AI ​​model, adapting to the change in movement speed and reducing the impact on the performance of the network on which the AI ​​model is deployed.

[0198]

[0212] 220: The first network element decides whether to switch or update the first AI model based on the correspondence information and the estimation results of the first parameters.

[0199]

[0213] The correspondence information shows the relationship between M AI models and N values ​​of the first parameter, where M is an integer greater than or equal to 1 and N is an integer greater than or equal to 1. Optionally, M=N or M>N.

[0200]

[0214] Optionally, M may be the total number of AI models included in the AI ​​model library, or the number of AI models included in one AI model group within the AI ​​model library. For example, the AI ​​model library may include one or more AI model groups, where different AI model groups correspond to different first parameters. For instance, the AI ​​model library may include AI model group 1 and AI model group 2, with a correspondence between the AI ​​models in AI model group 1 and the SINR, and a correspondence between AI model group 2 and the movement speed of the UE. If M may be the total number of AI models included in the AI ​​model library, the case in which the M AI models indicated by the correspondence information belong to one or more AI model groups is not limited to this embodiment of the present application. In other words, the correspondence information may include a correspondence between each of the one or more first parameters and the AI ​​model group corresponding to the first parameter.

[0201]

[0215] Optionally, M AI models and N values ​​of the first parameter may be replaced with M AI model groups and N values ​​of the first parameter. In other words, the same value of the first parameter may correspond to one or more AI model groups, and in other words, the same value of the first parameter may correspond to one AI model group, and an AI model group contains one or more AI models. The one or more AI models may differ in some characteristic, for example, one or more of the following characteristics: implementation complexity, performance, or similar. After the value of the first parameter is determined, the specific AI model to be selected from the corresponding AI model group may be predefined according to the protocol, or it may be determined based on system requirements or system settings. This is not limited to this case. When an AI model to be switched or updated is indicated, an AI model group may be indicated, or a specific AI model within an AI model group may be indicated. In this application, the indication of a correspondence between an AI model and the value of a first parameter includes the indication of a correspondence between an AI model group and the value of a first parameter, or the indication of a correspondence between a specific AI model and the value of a first parameter.

[0202]

[0216] In this embodiment of the present application, the value of the first parameter may represent the following: a specific value, for example, a specific quantized value, a range of values, or a level corresponding to a range of values.

[0203]

[0217] In this application, the correspondence information indicates the correspondence between the AI ​​model and the value of the first parameter. The estimated result of the first parameter, determined based on the channel measurement results, is the current value of the first parameter. Therefore, the AI ​​model corresponding to the estimated result of the first parameter can be determined based on the correspondence information. Optionally, different values ​​of the first parameter correspond to different AI models.

[0204]

[0218] Provided that it is possible to record the correspondence between the AI ​​model and the value of the first parameter, the correspondence information may be in the form of a table, function, string, or similar. Below, a table is used as an example to illustrate the correspondence information: Table 1 [Table 1]

[0205]

[0219] For example, suppose the AI ​​model currently in use by the first network element is AI Model 1, and the first network element has determined that the estimated value of the first parameter is 1. From the correspondence information, we know that value 1 corresponds to AI Model 1, and as a result, the first network element can determine that there is no need to switch or update the AI ​​model. If, however, the estimated value of the first parameter is determined to be 3, the first network element will determine that AI Model 1 needs to be switched to AI Model 3. In another example, if the first network element determines that the estimated value of the first parameter is 5, and based on the correspondence information, there is no AI model among the four existing AI models that matches value 5, the first network element will determine that the first AI model needs to be updated.

[0206]

[0220] In another example, M > N.

[0207]

[0221] It is known that higher complexity in an AI model indicates better robustness in adapting to more complex environments. Optionally, one value of the first parameter may correspond to multiple AI models with different complexity levels. The term "multiple" includes "two or more." Use Table 2 as an example for illustration: Table 2 [Table 2]

[0208]

[0222] In Table 2, the value of the first parameter, 4, corresponds to AI Model 4 and AI Model 5.

[0209]

[0223] For example, assuming that the complexity of AI model 5 is higher than that of AI model 4, AI model 5 has greater adaptability to the operating environment. Let's take CSI prediction as an example. In the same operating environment, the prediction accuracy of AI model 5 is higher than that of AI model 4. Assume that the first parameter is the migration speed of the network element to which the AI ​​model is deployed, and that the AI ​​model currently used by the first network element is AI model 2. If, based on correspondence information, it is determined that the estimated result of the first parameter corresponds to a value of 4, the first network element can further choose to switch to AI model 4 or AI model 5 based on the complexity of the AI ​​model. For example, if there is a high accuracy requirement for the AI ​​model at the current migration speed, the first network element may choose to switch to AI model 5; or, if there is a low accuracy requirement for the AI ​​model, the first network element may choose to switch to AI model 4.

[0210]

[0224] Optionally, method 200 may further include step 230, which may be performed before step 210.

[0211]

[0225] 230: The first network element retrieves the second piece of information, and the second piece of information indicates the first parameter.

[0212]

[0226] In Method 200, it will be understood that the first network element can make a timely decision to switch or update the AI ​​model by monitoring the input indicators of the AI ​​model, i.e., the first parameters. Therefore, the first parameters may alternatively be referred to as the monitoring parameters of the AI ​​model.

[0213]

[0227] Optionally, if the correspondence information indicates the correspondence between an AI model and several first parameters, the first network element determines the current monitoring parameters based on the acquired second information and decides whether to switch or update the AI ​​model, for example, deciding whether to switch or update the AI ​​model: Table 3 [Table 3]

[0214]

[0228] Table 3 is used as an example. The correspondences indicated by the correspondence information include the correspondence between the movement speed of the network element in which the AI ​​model is deployed and the AI ​​model, and the correspondence between the SINR and the AI ​​model. If the second information indicates that the first parameter is the movement speed of the network element in which the AI ​​model is deployed, the first network element decides whether to switch or update the AI ​​model based on the estimated movement speed of the network element in which the AI ​​model is deployed, for example, deciding whether to switch or update the AI ​​model.

[0215]

[0229] Optionally, in step 230, the first network element may obtain the second information from the network side. For example, the first network element may obtain the second information from the second or third network element. This is not limited to the first network element.

[0216]

[0230] As can be seen from the description of Method 200, the technical solution of this application allows for timely notification of changes in the operating environment of the AI ​​model by monitoring the estimation results of the first parameter, in other words, by monitoring the input to the AI ​​model. In this way, the AI ​​model can be switched or updated in a timely manner to adapt to changes in the operating environment of the AI ​​model. This helps to mitigate the problem of performance degradation of the network on which the AI ​​model is deployed after the operating environment of the AI ​​model changes.

[0217]

[0231] Below are some examples of the correspondence between the AI ​​model and the value of the first parameter.

[0218]

[0232] Example 1: The first parameter includes the movement speed of the UE: Table 4 [Table 4]

[0219]

[0233] Example 2: The first parameter includes the absolute number of paths whose energy is greater than k times the energy of the first path in the CIR. Hereafter, the absolute number of paths whose energy is greater than k times the energy of the first path in the CIR will be referred to simply as the number of paths. k is a number greater than 0. Optionally, k is a number greater than 0 and less than 1. For example, k is 0.5: Table 5 [Table 5]

[0220]

[0234] Example 3: The first parameter includes the relative ratio of paths whose energy is greater than k times the energy of the first path in the CIR, and is hereafter referred to as the relative ratio. k is a number greater than 0. Optionally, k is a number greater than 0 and less than 1. For example, k is 0.5: Table 6 [Table 6]

[0221]

[0235] Example 4: The first parameter includes SINR: Table 7 [Table 7]

[0222]

[0236] As mentioned above, the first parameter may, alternatively, be another parameter not mentioned in this case.

[0223]

[0237] Optionally, the first AI model is deployed to either the first or second network element.

[0224]

[0238] Optionally, in one example, if the first network element determines that the estimation result of the first parameter corresponds to the first value out of N values ​​for the first parameter, and that the first value corresponds to the second AI model out of M AI models, then the first network element decides to switch the first AI model to the second AI model.

[0225]

[0239] Optionally, in another example, if it is determined that the estimation result for the first parameter does not correspond to any of the N possible values ​​for the first parameter, the first network element decides to update the first AI model.

[0226]

[0240] The statement above that "the estimation result of the first parameter corresponds to the first value among the N values ​​of the first parameter" may be any of the following possible implementations.

[0227]

[0241] The N values ​​of the first parameter are N specific values, and the estimated result of the first parameter is equal to a specific value among the N specific values.

[0228]

[0242] Alternatively, the N values ​​of the first parameter are N value ranges, and the estimated result of the first parameter belongs to a specific value range among the N value ranges.

[0229]

[0243] Alternatively, the N values ​​of the first parameter are N levels, and the estimation result of the first parameter corresponds to a specific level among the N levels.

[0230]

[0244] In other words, any one of the aforementioned possible implementations is a "correspondence."

[0231]

[0245] In step 210, the method by which the first network element acquires the first information differs depending on whether the first AI model is deployed in the first network element or in the second network element.

[0232]

[0246] The following describes how to switch or update the first AI model using the first network element, in cases where the first AI model is deployed in either the first or second network element.

[0233]

[0247] (1) The first AI model is deployed to the first network element, and the first network element decides whether to switch or update the AI ​​model based on the corresponding information, for example, decides whether to switch or update the first AI model.

[0234]

[0248] If the first AI model is deployed to the first network element, the acquisition of the first information by the first network element in step 210 includes the following:

[0235]

[0249] The first network element measures a reference signal to obtain channel measurement results.

[0236]

[0250] The first network element obtains an estimate of the first parameter based on the channel measurement results, and the first information includes the estimate of the first parameter. In other words, in this case, the first information may be the estimate of the first parameter.

[0237]

[0251] For example, the first network element is a terminal device, and the second network element is an access network device. The first network element obtains channel measurement results by measuring a reference signal from the second network element. Based on the channel measurement results, the first network element can determine the estimated results of the first parameter. The reference signal includes, but is not limited to, one or more of the following: channel state information-reference signal (CSI-RS), synchronization signal and synchronization signal in a physical broadcast channel block (SSB), signal on a physical broadcast channel, and / or similar.

[0238]

[0252] When the first AI model is deployed to the first network element, the first network element can obtain channel measurement results by measuring a reference signal from the second network element. The estimation results of the first parameters may be obtained based on the channel measurement results. Furthermore, the first network element decides whether to switch or update the first AI model based on the correspondence information and the estimation results of the first parameters.

[0239]

[0253] Where possible, the first network element decides to switch the first AI model to the second AI model. In this case, the first network element sends first instruction information to the second network element, indicating that the first network element requests the first AI model to switch to the second AI model. Alternatively, the first instruction information may be called a switch request. It should be understood that the first network element's decision to switch the first AI model to the second AI model means that the second AI model belongs to one of the M AI models recorded in the correspondence information. After receiving the first instruction information from the first network element, the second network element sends the second AI model to the first network element. The first network element receives the second AI model from the second network element and switches the first AI model to the second AI model.

[0240]

[0254] In this case, for example, the first network element may be an inference network element for an AI model, and the second network element may be an AI model library, i.e., a storage network element for multiple AI models. When it is decided to switch from the first AI model to the second AI model based on the acquired correspondence information, the first network element sends first instruction information to the second network element. For example, the first instruction information indicates the identifier of the second AI model. Based on the first instruction information, the second network element sends the stored second AI model to the first network element.

[0241]

[0255] In all embodiments of this application, the statement that the first network element may be an inference network element of an AI model means that the first network element is at least an inference network element, and is not limited to cases where the first network element is not necessarily a storage network element or a training network element of an AI model library. For example, in an embodiment, the first network element may be an inference network element of an AI model, and may also be a storage network element and / or a training network element, although the functionality of the storage network element and / or training network element is not reflected in this embodiment. Alternatively, in an embodiment, the first network element is an inference network element of an AI model. However, in another embodiment, the first network element is a storage network element and / or a training network element of an AI model library. The relevant descriptions of the second and third network elements are similar and will not be described again below.

[0242]

[0256] In another possible case, the first network element decides to update the first AI model. In this case, it means that among the M AI models recorded in the correspondence information, there is no AI model that matches the estimation result of the first parameter currently determined by the first network element. The first network element can send second instruction information to the second network element, which is used to request that the first AI model be updated. Alternatively, the second instruction information may be called an update request. The second network element receives the second instruction information from the first network element and knows that the first network element is requesting that the first AI model be updated. In this case, the second network element obtains training data, trains the AI ​​model based on the training data, and obtains a third AI model. For example, after receiving the second instruction information, the second network element may instruct the first network element to send training data. Based on the instructions of the second network element, the first network element provides the training data to the second network element. Alternatively, the first network element may send second instruction information and training data to the second network element if it decides to update the first AI model. This is not limited to this. Furthermore, the second network element may, alternatively, obtain training data from the third network element. This is not limited to this.

[0243]

[0257] The second network element, after acquiring the third AI model through training, sends the third AI model to the first network element. The first network element receives the third AI model from the second network element and updates its own AI model to the third AI model.

[0244]

[0258] In this case, for example, the first network element may be an inference network element of the AI ​​model, and the second network element may be a training network element of the AI ​​model.

[0245]

[0259] Optionally, when the first AI model is deployed to the first network element, the correspondence information based on the first network element's decision to switch or update the first AI model may be obtained by the first network element from the second network element, or from the first network element from the third network element, or may be predefined and pre-stored in the protocol. Furthermore, alternatively, any correspondence indicated by the correspondence information may be obtained by the first network element from the second or third network element, and any correspondence may be predefined and pre-stored in the protocol. These implementations are not limited to those described herein.

[0246]

[0260] Referring to Figure 3, the following will explain, using specific examples, how the first AI model can be switched or updated by the first network element when the first AI model is deployed to the first network element.

[0247]

[0261] Figure 3 shows an example of a method for switching or updating an AI model according to this application.

[0248]

[0262] 301: The first network element measures the reference signal from the second network element to obtain the channel measurement result.

[0249]

[0263] 302: The first network element obtains the estimated result of the first parameter based on the channel measurement result.

[0250]

[0264] 303: The first network element decides whether to switch or update the first AI model based on the correspondence information and the estimation results of the first parameters.

[0251]

[0265] Where possible, the first network element decides to switch the first AI model to the second AI model based on the correspondence information and the estimation results of the first parameters. In this case, steps 304 to 306 are further included after step 303.

[0252]

[0266] 304: The first network element transmits first instruction information to the second network element, and the first instruction information is used to request a switch from the first AI model to the second AI model.

[0253]

[0267] The second network element transmits the second AI model to the first network element based on the first instruction information from the first network element.

[0254]

[0268] For example, in an implementation, correspondence information is predefined in the protocol. In this case, the first network element sends the first instruction information to the second network element, and the first instruction information carries the identifier of the second AI model. Based on the correspondence information predefined in the protocol and the identifier of the second AI model, the second network element sends the second AI model to the first network element.

[0255]

[0269] For example, in another implementation, the first network element transmits first instruction information and the estimation result of the first parameter to the second network element. Based on the first instruction information and the estimation result of the first parameter, the second network element decides to transmit the second AI model to the first network element. For example, based on the estimation result of the first parameter, the second network element determines that the estimation result of the first parameter corresponds to the second AI model in the model library stored in the second network element, and transmits the second AI model to the first network element. Optionally, in this implementation, the correspondence information may be predefined in the protocol or stored locally by the second network element. Furthermore, the first network element may obtain the correspondence information from the second or third network element, store the correspondence information in the first network element, and decide whether to switch or update the AI ​​model based on the correspondence information, as described in step 303 above.

[0256]

[0270] For example, in yet another implementation, the first network element sends the estimation result of the first parameter to the second network element, and the estimation result of the first parameter implicitly indicates that the first network element is requesting a switch to the first AI model. The second network element then sends the second AI model to the second network element based on the estimation result of the first parameter and the corresponding information.

[0257]

[0271] 305: The first network element receives the second AI model from the second network element.

[0258]

[0272] 306: The first network element switches the first AI model to the second AI model.

[0259]

[0273] In another possible case, the first network element decides to update the first AI model based on the correspondence information and the estimation results of the first parameters. In this case, steps 307 to 310 are further included after step 303.

[0260]

[0274] 307: The first network element transmits the second instruction information to the second network element, which is used to request that the first AI model be updated.

[0261]

[0275] Optionally, the implementation may be as follows: The first network element sends the estimation result of the first parameter to the second network element, and the second network element, based on the correspondence information and the estimation result of the first parameter, determines that the estimation result of the first parameter does not correspond to any AI model in the model library stored in the second network element. Therefore, the AI ​​model needs to be updated.

[0262]

[0276] Optionally, another implementation may be as follows: The first network element sends second instruction information and the estimation results of the first parameters to the second network element. Based on the second instruction information, the second network element knows that the first network element is requesting an update to the AI ​​model. For example, the estimation results of the first parameters may be used as training data, which the second network element uses to train the AI ​​model and obtain a new AI model.

[0263]

[0277] Optionally, the estimation result of the first parameter in step 302 is not limited to being obtained based on the channel measurement result of a single channel measurement, nor is it limited to being obtained based on the channel measurement results of multiple channel measurements. For example, the estimation result of the first parameter may be the average value of multiple channel measurement results. This is not limited.

[0264]

[0278] 308: The second network element executes AI model training based on the second instruction information to obtain a third AI model.

[0265]

[0279] For example, the second network element obtains training data from the first network element and executes AI model training.

[0266]

[0280] Optionally, the second network element obtaining training data and executing AI model training to obtain a third AI model may include one of the following implementations.

[0267]

[0281] The second network element obtains sufficient training data and trains the entire AI model to obtain a third AI model.

[0268]

[0282] Alternatively, the second network element obtains sufficient training data and trains a specific functional layer of the AI model to obtain a third AI model.

[0269]

[0283] Alternatively, the second network element obtains a small amount of training data and trains the entire AI model to obtain a third AI model.

[0270]

[0284] Alternatively, the second network element obtains a small amount of training data and trains a specific functional layer of the AI model to obtain a third AI model.

[0271]

[0285] 309: The second network element sends the third AI model to the first network element.

[0272]

[0286] The first network element receives the third AI model from the second network element.

[0273]

[0287] 310: The first network element updates the first AI model to the third AI model.

[0274]

[0288] (2) The first AI model is deployed to the second network element, and the first network element decides whether to switch or update the AI ​​model based on the corresponding information, for example, it decides whether to switch or update the first AI model deployed to the second network element.

[0275]

[0289] Referring to Figure 4, the following will explain, using specific examples, how the first network element can switch or update the first AI model when the first AI model is deployed to the second network element.

[0276]

[0290] Figure 4 shows an example of a method for switching or updating an AI model according to this application.

[0277]

[0291] 401: The second network element measures the reference signal from the first network element to obtain the channel measurement result.

[0278]

[0292] 402: The second network element obtains the estimated result of the first parameter based on the channel measurement result.

[0279]

[0293] 403: The second network element transmits the first information to the first network element.

[0280]

[0294] The first piece of information shows the estimation result for the first parameter.

[0281]

[0295] For example, the first piece of information may be the estimation result of the first parameter, or the first piece of information may be the information used to determine the estimation result of the first parameter.

[0282]

[0296] The first network element receives the first piece of information from the second network element.

[0283]

[0297] 404: The first network element determines whether to switch or update the first AI model based on the corresponding information and the first information.

[0284]

[0298] If possible, the first network element determines to switch the first AI model to the second AI model based on the corresponding information and the estimation result of the first parameter. In this case, after step 404, steps 405 and 406 are further included.

[0285]

[0299] 405: The first network element sends the second AI model to the second network element.

[0286]

[0300] The second network element receives the second AI model from the first network element.

[0287]

[0301] 406: The second network element switches the first AI model to the second AI model.

[0288]

[0302] In another possible case, the first network element determines to update the first AI model based on the corresponding information and the estimation result of the first parameter. In this case, after step 404, steps 407 to 409 are further included.

[0289]

[0303] 407: The first network element executes AI model training to obtain a third AI model.

[0290]

[0304] For example, the first network element may obtain training data from the second network element, execute AI model training based on the training data, and obtain a third AI model.

[0291]

[0305] Similar to step 308 above, the first network element obtaining the third AI model through training may include one of the following implementations.

[0292]

[0306] The first network element acquires sufficient training data, trains the entire AI model, and obtains the third AI model.

[0293]

[0307] Alternatively, the first network element acquires sufficient training data and trains a specific functional layer of the AI ​​model to obtain a third AI model.

[0294]

[0308] Alternatively, the first network element acquires a small amount of training data, trains the entire AI model, and obtains a third AI model.

[0295]

[0309] Alternatively, the first network element acquires a small amount of training data, trains a specific functional layer of the AI ​​model, and obtains a third AI model.

[0296]

[0310] 408: The first network element sends the third AI model to the second network element.

[0297]

[0311] The second network element receives the third AI model from the first network element.

[0298]

[0312] 409: The second network element updates the first AI model to the third AI model.

[0299]

[0313] In the examples shown in Figures 3 and 4, the first network element monitors the input to the AI ​​model and, based on corresponding information and monitoring results (e.g., estimation results for the first parameter), determines whether the AI ​​model needs to be switched or updated to adapt to changes in the environment in which the AI ​​model is used. This helps mitigate the problem of AI model performance degrading or becoming worse due to significant changes in the environment in which it is used.

[0300]

[0314] The AI ​​model switching or updating methods shown in Figures 2 to 4 may be applied to multiple scenarios, including but not limited to CSI prediction or beam management.

[0301]

[0315] The applications of the AI ​​model switching or updating method of the present application in CSI prediction or beam management will be described below using an example in which the first network element is a UE and the second network element is an access network device, or a device other than an access network device, such as a model library storage and / or training device. Optionally, the model library storage and / or training device may communicate with the UE via the access network device. The access network device may transparently transmit communication information, or process and then forward the communication information.

[0302]

[0316] Application Scenario 1

[0317] AI-based CSI prediction (i.e., AI-based CSI prediction or AI-CSI prediction)

[0318] Figure 5 shows the AI-CSI prediction diagram. The basic principle of AI-CSI prediction is to train a neural network on a training set containing a large amount of offline CSI using the feature extraction and fitting capabilities of the neural network. As a result, the neural network can learn channel change modes and fit a nonlinear mapping relationship between historical CSI and future CSI to replace the channel prediction closed-form expression based on mathematical models. In the inference phase, several historical CSI (e.g., CSI fed back by the receiving end and / or CSI predicted by the transmitting end at a past point in time) are input to the neural network used for channel prediction, and predicted values ​​of CSI at a future point in time are output. Since the transmitting end can predict the CSI, the receiving end does not need to feed back the estimated CSI at the prediction point in time. Therefore, the CSI feedback overhead on the air interface can be reduced.

[0303]

[0319] Figure 6 shows an example of the application of the AI ​​model switching or updating method described in this application.

[0304]

[0320] In the example in Figure 6, the first parameter being the movement speed of the UE is used for illustrative purposes.

[0305]

[0321] 601: Optionally, the UE retrieves correspondence information.

[0306]

[0322] For example, the UE obtains correspondence information from an access network device or a model library storage and / or training device. Optionally, the model library storage and / or training device may transmit correspondence information to the UE via the access network device.

[0307]

[0323] 602:UE obtains channel measurement results by measuring the channel status information-reference signal CSI-RS from the access network device.

[0308]

[0324] 603:UE obtains an estimated result for the first parameter based on the channel measurement results.

[0309]

[0325] The channel measurement results are input to the AI ​​model, and the estimation result of the first parameter can be determined based on the channel measurement results.

[0310]

[0326] The following sections will explain the switching and updating of AI models separately, using specific examples.

[0311]

[0327] A switch is used as an example. After step 603, steps 604 through 608 may be performed.

[0312]

[0328] 604:UE decides to switch from the first AI model to the second AI model based on the correspondence information and the estimation results of the first parameter.

[0313]

[0329] Table 2 mentioned above is used as an example. Assume that the first AI model currently used by UE is AI Model 2, and that UE's current movement speed is determined to be 90 km / h. Based on the correspondence information and UE's current movement speed, UE can determine that its current movement speed corresponds to value 4 out of the four values, and therefore decide to switch from AI Model 2 to AI Model 4.

[0314]

[0330] 605:UE transmits the first instruction information to an access network device or model library storage and / or training device, and the first instruction information is used to request a switch from the first AI model to the second AI model.

[0315]

[0331] The access network device or model library storage and / or training device receives first instruction information from the UE and transmits the AI ​​model 4 to the UE based on the first instruction information.

[0316]

[0332] Optionally, the model library storage and / or training device can receive first instruction information from the UE via the access network device and transmit the AI ​​model 4 to the UE via the access network device.

[0317]

[0333] 606:UE receives AI model 4 from an access network device or model library storage and / or training device.

[0318]

[0334] 607:UE switches AI Model 2 to AI Model 4.

[0319]

[0335] Optionally, step 608 is also included.

[0320]

[0336] 608:UE performs CSI prediction based on AI model 4.

[0321]

[0337] The update is used further as an example. After step 603, steps 609 through 614 may be performed.

[0322]

[0338] 609:UE decides to update the first AI model based on the correspondence information and the estimation results of the first parameter.

[0323]

[0339] Table 1 is used as an example. Assume that UE has determined, based on the channel measurement results, that its current speed is 120 km / h. It can be seen that UE's current speed does not correspond to any of the four values. In this case, UE decides to update the first AI model.

[0324]

[0340] 610:UE transmits second instruction information to an access network device or model library storage and / or training device, and the second instruction information is used to request an update to the first AI model.

[0325]

[0341] 611: An access network device or model library storage and / or training device performs AI model training based on the second instruction information to obtain a third AI model.

[0326]

[0342] In this example, we assume that the third AI model is AI Model 5.

[0327]

[0343] 612: The access network device or model library storage and / or training device transmits a third AI model (e.g., AI model 5) to the UE.

[0328]

[0344] 613:UE switches from AI Model 2 to AI Model 5.

[0329]

[0345] Optionally, step 614 is also included.

[0330]

[0346] 614:UE performs CSI predictions based on AI model 5.

[0331]

[0347] Application Scenario 2

[0348] AI-based beam management

[0349] Figure 7 shows an example of the application of the AI ​​model switching or updating method described in this application.

[0332]

[0350] In the example in Figure 7, the example where the first parameter is the movement speed of the UE is used again for illustrative purposes.

[0333]

[0351] 701: Optionally, the access network device obtains corresponding information.

[0334]

[0352] For example, an access network device may have pre-stored correspondence information or obtain it from a model library storage and / or training device.

[0335]

[0353] 702:UE obtains channel measurement results by measuring the channel status information-reference signal CSI-RS from the access network device.

[0336]

[0354] 703:UE obtains an estimated result for the first parameter based on the channel measurement results.

[0337]

[0355] 704:UE transmits the first information to an access network device or model library storage and / or training device, the first information representing the estimation result of the first parameter.

[0338]

[0356] A switch is used as an example. After step 704, steps 705 through 708 may be performed.

[0339]

[0357] 705: The access network device or model library storage and / or training device decides to switch the first AI model currently in use by the UE to the second AI model, based on the correspondence information and the estimation results of the first parameter.

[0340]

[0358] For an implementation where an access network device decides to switch from the first AI model to the second AI model, see step 604. For brevity, the details are not explained again here.

[0341]

[0359] 706: The access network device or model library storage and / or training device transmits the second AI model to the UE.

[0342]

[0360] The UE receives a second AI model from the access network device.

[0343]

[0361] 707:UE switches from the first AI model to the second AI model.

[0344]

[0362] Step 708 is also included as an option.

[0345]

[0363] 708:UE performs beam management based on a second AI model.

[0346]

[0364] An update is used as an example. After step 704, steps 709 through 713 may be performed.

[0347]

[0365] 709: The access network device or model library storage and / or training device decides to update the first AI model currently in use by the UE based on the correspondence information and the estimation results of the first parameter.

[0348]

[0366] 710: Access network devices or model library storage and / or training devices perform AI model training to obtain a third AI model.

[0349]

[0367] 711: The access network device sends the third AI model to the UE, and the UE receives the third AI model.

[0350]

[0368] 712:UE updates the first AI model to the third AI model.

[0351]

[0369] Optionally, step 713 is also included.

[0352]

[0370] 713:UE performs beam management based on a third AI model.

[0353]

[0371] In some of the examples above, the first AI model is deployed in a first network element, and after the first network element decides to switch or update the first AI model and sends a request to the second network element to switch or update the AI ​​model, the second network element provides the first network element with the second AI model to be used for the switch or a new AI model acquired through training, i.e., a third AI model. In some other examples, the first AI model is deployed in a second network element, and after the second network element decides to switch or update the first AI model, the first network element sends the second AI model to be used for the switch or a new AI model acquired through training to the second network element. These examples are illustrated by using examples where the network element to which the first AI model is deployed is not the storage network element of the AI ​​model library, nor is it the training network element of the AI ​​model. For example, the first AI model is deployed to the UE, which may be an inference network element of the AI ​​model, but not a storage network element of the AI ​​model library or a training network element of the AI ​​model. After deciding to switch the first AI model to the second AI model based on correspondence information, the UE can send a switch request to the storage network element of the second AI model library, retrieve the second AI model from the storage network element, and perform the switch. Alternatively, after deciding to update the first AI model based on correspondence information, the UE may send an update request to the training network element of the AI ​​model (e.g., an access network device or another network device), which will train the AI ​​model and deliver the new AI model to the UE.

[0354]

[0372] Furthermore, in the example described above, the access network device may function as both a storage network element and a training network element. In other words, the access network device has database storage and training capabilities. Optionally, the storage network element and training network element of the AI ​​model may be different network elements. This is not limited to the present application. For example, the first network element (e.g., UE) is the inference network element of the AI ​​model, the second network element (e.g., the access network device) is the training network element of the AI ​​model, and the M AI models recorded in the correspondence information are stored in the third network element, in other words, the third network element is the storage network element of the AI ​​model. In this scenario, we assume that the first network element decides whether to switch or update the AI ​​model, for example, deciding whether to switch or update the AI ​​model deployed in the first network element. If it decides to switch the currently deployed first AI model to the second AI model, the first network element retrieves the second AI model from the third network element. When it is decided to update the currently deployed first AI model, the first network element sends an update request to the second network element, and the second network element, after acquiring a new third AI model through training, provides the new third AI model to the first network element. Furthermore, the second network element may further send the third AI model to the third network element for storage. Thus, specific implementations of these are not particularly limited in this application.

[0355]

[0373] To simplify the explanation of the solution, the following embodiments will be illustrated by using an example in which the storage network element and the training network element are the same network element.

[0356]

[0374] Optionally, this application further provides an implementation where the inference network elements of the AI ​​model are both storage network elements and training network elements. For example, a UE can be used as an example, and at the same time, the UE may be an inference network element, a storage network element, and a training network element. Specifically, the UE can obtain channel measurement results by measuring a reference signal from an access network device, and further obtain estimation results of a first parameter. For example, the first parameter is the UE's movement speed. The UE monitors its movement speed and switches or updates the AI ​​model based on the corresponding information and the movement speed monitoring results.

[0357]

[0375] The above describes in detail the application of the method for switching or updating AI models provided in this application to CSI prediction and beam management. This application further provides the application of the method to auto-encoder (AE) models.

[0358]

[0376] The AE model will be explained below using AI-based CSI feedback (sometimes referred to as AI-CSI feedback) as an example.

[0359]

[0377] Figure 8 is a diagram of an AI-based CSI feedback process. In AI-CSI feedback, when the AI ​​model is deployed on the base station side, the base station obtains the estimated CSI-RS results from the UE side and uses the estimated results as labels (sometimes referred to as ground truth labels) for training. The AE model includes two sub-models: an encoder and a decoder. AE generally refers to a network structure containing two sub-models. The AE model may also be referred to as a bilateral model, dual-end model, or cooperative model. The encoder and decoder of the AE are usually trained together and may be used in a matching scheme. CSI feedback may be implemented based on the AI ​​model of the AE. For example, the UE side performs CSI compression and quantization by the encoder, and the base station performs CSI reconstruction by the decoder. As shown in Figure 8, for the base station, the input to the model is the CSI fed back by the UE, and the output is the reconstructed CSI. Model training requires using the CSI measured by the UE side as the ground truth label for the reconstructed CSI.

[0360]

[0378] Figure 9 is a schematic flowchart of the AI ​​model switching or updating method 800 according to the present application.

[0361]

[0379] As shown in Figure 8, in the dual-end model, the encoder on the UE side and the decoder on the base station side are used in a matching scheme. The encoder on the UE side may contain one or more AI models, and the decoder on the base station side that matches the encoder on the UE side may also contain one or more AI models. The number of AI models included in the encoder and decoder used in the matching scheme is the same, and there is a one-to-one correspondence between the AI ​​models. In the following embodiments, an example is used in which the encoder on the UE side contains one AI model and the decoder on the base station side contains one AI model for illustrative purposes.

[0362]

[0380] Assume that the first AI model is deployed to the first network element, the fourth AI model is deployed to the second network element, and that the fourth AI model and the first AI model are used in a matching scheme. It should be understood that the output of the first AI model is the input of the fourth AI model, and the ground truth label of the fourth AI model is the input of the first AI model.

[0363]

[0381] 801: The first network element retrieves the first piece of information.

[0364]

[0382] 802: The first network element decides whether to switch or update the first AI model based on the correspondence information and the first information.

[0365]

[0383] If, where possible, the first network element decides to switch from the first AI model to the second AI model, then steps 803 through 807 are further included after step 802.

[0366]

[0384] 803: If it is decided to switch from the first AI model to the second AI model, the first network element sends first instruction information to the second network element, and the first instruction information indicates that the first network element is requesting to switch from the first AI model to the second AI model.

[0367]

[0385] The second network element receives first instruction information from the first network element.

[0368]

[0386] For step 803, please refer to the alternative implementation of step 304 in method 300. Further details will not be provided here.

[0369]

[0387] 804: The second network element determines, based on the first instruction information, whether the W stored AI models include an AI model that matches the second AI model, and obtains the determination result, where W is an integer greater than or equal to 1.

[0370]

[0388] It should be understood that the correspondence information in the above-described embodiment indicates the correspondence between M AI models and N values ​​of the first parameter. In the dual-end model embodiment, the M AI models may be M encoders, and the W AI models in step 804 may be W decoders. M and W may or may not be equal; this is not limited. For example, a second network element knows, based on the first instruction information, that the first network element is requesting to switch the first encoder to the second encoder. The second network element determines whether the W stored decoders include a decoder that matches the second encoder and obtains a determination result.

[0371]

[0389] Optionally, the W decoders may be stored in a second network element, in other words, the second network element may be an AI model storage network element. Alternatively, the W decoders may be stored in a third network element (a third-party device), or in any other form. This is not limited to these.

[0372]

[0390] Optionally, correspondence information on the first network element side is predefined in the protocol. In this case, if the first network element decides to switch from the first AI model to the second AI model, it can include the identifier of the second AI model in the first instruction information to be sent to the second network element. Based on the correspondence information predefined in the protocol, the second network element can know that the first network element is requesting to switch to the second AI model.

[0373]

[0391] Optionally, in step 804, the second network element may determine, based on other correspondence information, whether the W AI models stored in the second network element include AI models that match the second AI model. The term "other correspondence information" is used to distinguish it from the "correspondence information" described above. For example, in step 210 of Figure 2, the correspondence information obtained by the first network element may be correspondence information 1. Here, the second network element determines, based on correspondence information 2, whether the W AI models include AI models that match the second AI model. Alternatively, the M AI models recorded in correspondence information 1 are an AI model library of AI models that can be used by the first network element, and the W AI models recorded in correspondence information 2 are an AI model library of AI models that can be used by the second network element.

[0374]

[0392] For example, the correspondences recorded in correspondence information 1 and correspondence information 2 are shown in the following tables: Table 8 [Table 8] Table 9 [Table 9]

[0375]

[0393] In one example, the first network element is currently using AI model 1, the second network element is currently using AI model a, and AI model 1 matches AI model a. If the estimation result of the first parameter corresponds to value 3 recorded in correspondence information 1, the first network element decides to switch from AI model 1 (e.g., the first AI model) to AI model 3 (e.g., the second AI model). The first network element sends the first instruction information and the estimation result of the first parameter to the second network element. The second network element determines the estimation result of the first parameter and the value 3 recorded in correspondence information 2, and value 3 corresponds to AI model c. In this case, the second network element determines that W AI models include an AI model that matches the second AI model, specifically AI model c in this example.

[0376]

[0394] 805: The second network element instructs the first network element whether to switch the first AI model based on the judgment result.

[0377]

[0395] In possible cases, if the determination result is that none of the W AI models match the second AI model, the second network element sends a third instruction to the first network element, and the third instruction instructs the first network element not to switch the first AI model. In this case, the first network element performs step 806.

[0378]

[0396] 806: The first network element skips switching the first AI model based on the third instruction information.

[0379]

[0397] In another possible case, if the determination result is that W AI models include a fifth AI model that matches the second AI model, the second network element sends a third instruction to the first network element, which instructs the first network element to switch the first AI model to the second AI model. In this case, the first network element performs step 807.

[0380]

[0398] 807: The first network element switches the first AI model to the second AI model based on the third instruction information. Furthermore, the second network element switches the fourth AI model to the fifth AI model. In other words, the second network element works in conjunction with the first network element to perform the AI ​​model switching.

[0381]

[0399] In the example in step 804, the first network element switches AI model 1 to AI model 3, and the second network element switches AI model a to AI model c.

[0382]

[0400] When Method 800 is applied to a CSI feedback application scenario, an example is used in which the first network element is the UE and the second network element is an access network device. The first AI model is deployed on the UE, and the fourth AI model is deployed on the access network device, with the first AI model matching the fourth AI model. The first AI model is used to compress or compress and quantize the CSI measured by the UE, and the output of the first AI model is the CSI to be fed back. The input of the fourth AI model is the output of the first AI model, and the output of the fourth AI model is the reconstructed CSI. The ground truth label of the fourth AI model is the CSI measured on the UE side. If the UE decides to switch the first AI model, it sends a switch request to the access network device. After receiving information from the access network device indicating that a switch should be performed (e.g., a possible third instruction), the UE switches from the first AI model to the second AI model, and the access network device switches from the fourth AI model to the fifth AI model. The second AI model matches the fifth AI model. If the UE receives information from the access network device indicating that a switch should not be performed (e.g., another possible third instruction), neither the UE nor the access network device switches the currently used AI model. The UE continues to use the first AI model, and the access network device continues to use the fourth AI model.

[0383]

[0401] Figure 10 is a schematic flowchart of the AI ​​model switching or updating method 900 according to the present application.

[0384]

[0402] 901: The first network element obtains the first piece of information.

[0385]

[0403] 902: The first network element decides whether to switch or update the first AI model based on the correspondence information and the first information.

[0386]

[0404] In cases where possible, if the first network element decides to update the first AI model, steps 903 through 907 are further included after step 902.

[0387]

[0405] 903: If it is decided to update the first AI model, the first network element sends second instruction information to the second network element, and the second instruction information indicates that the first network element is requesting that the first AI model be updated.

[0388]

[0406] The second network element receives second instruction information from the first network element.

[0389]

[0407] Optionally, the second network element further transmits the estimation results of the first parameter to the first network element.

[0390]

[0408] As an option, for step 903, please refer to the alternative implementation of step 307 in method 300. Further details are not provided here.

[0391]

[0409] 904: The second network element decides whether to update the first AI model based on the second instruction information.

[0392]

[0410] The second network element knows, based on the second instruction information, that the first network element is requesting an update to the first AI model. In one example, the training data used to update the AI ​​model is provided to the second network element by the first network element. If the first network element determines that it is able to provide training data that meets the requirements, the second network element decides to update the first AI model. The first network element is the UE, and the second network element is the network device. If the UE decides to update the first AI model, it provides the network device with more training data. For example, the training data is the estimation results of the first parameters. Alternatively, the UE may provide more training data to the network device. Optionally, the UE providing training data to the network device may be: after deciding to update the AI ​​model, the UE sends the training data to the network device. Alternatively, the network device instructs the UE to provide training data based on the second instruction information. This is not limited to this. In this example, the network device decides to update the first AI model. Subsequently, the network device trains the AI ​​model based on the training data provided by the UE to obtain a new AI model, for example, a third AI model. In another example, the training data used to update the AI ​​model is generated on the side of a second network element. In this case, if it is determined that no training data exists or that there is no training data that meets the requirements for training a new AI model, the second network element decides not to update the first AI model. For example, the second network element is the UE. If it is determined that no training data exists or that there is not enough training data, the UE decides not to train the first AI model.

[0393]

[0411] 905: The second network element instructs the first network element whether to update the first AI model based on the judgment result.

[0394]

[0412] In cases where it is possible, if it is decided not to update the first AI model, the second network element sends fourth instruction information to the first network element, which instructs the first network element not to update the first AI model. In this case, the first network element performs step 906.

[0395]

[0413] 906: The first network element skips updating the first AI model based on the fourth instruction information.

[0396]

[0414] In another possible case, if it is decided to update the first AI model, the second network element sends fourth instruction information to the first network element, which instructs the first network element to update the first AI model. In this case, the first network element performs step 907.

[0397]

[0415] 907: The first network element updates the first AI model to the third AI model based on the fourth instruction information.

[0398]

[0416] When it is decided to update the first AI model, and the first network element is instructed to update the first AI model, it should be understood that the second network element also updates the fourth AI model deployed in the second network element to match the updated AI model of the first network element. Specifically, the second network element updates the fourth AI model to the sixth AI model, and the sixth AI model matches the third AI model.

[0399]

[0417] When Method 900 is applied to a CSI feedback application scenario, an example is used in which the first network element is a UE and the second network element is an access network device. The first AI model is deployed on the UE, and the fourth AI model is deployed on the access network device, with the first AI model conforming to the fourth AI model. The first AI model is used to compress or compress and quantize the CSI measured by the UE, and the output of the first AI model is the feedbacked CSI. The input to the fourth AI model is the output of the first AI model, and the output of the fourth AI model is the reconstructed CSI. The ground truth label of the fourth AI model is the CSI measured on the UE side. When it is decided to update the first AI model, the UE sends an update request to the access network device. After receiving information from the access network device indicating that an update should be performed (e.g., a possible fourth instruction), the UE updates the first AI model to the third AI model, and the access network device updates the fourth AI model to the sixth AI model. The third AI model matches the sixth AI model. If the UE receives information from the access network device indicating that an update should not be performed (e.g., another possible fourth instruction), neither the UE nor the access network device updates the currently used AI model. The UE continues to use the first AI model, and the access network device continues to use the fourth AI model.

[0400]

[0418] The embodiments described above illustrate in detail the application of the method for switching or updating AI models provided in this application to a dual-end model. In the embodiments described above of the dual-end model, it can be understood that the UE determines, based on corresponding information, whether the AI ​​model currently in use by the UE needs to be switched or updated, and after determining that the AI ​​model currently in use by the UE needs to be switched or updated, the UE sends a switch request or update request to the access network device. The UE can switch or update the currently used first AI model based solely on acknowledgments fed back from the access network device to the switch request or update request, for example, based solely on third instruction information indicating to switch the first AI model, or fourth instruction information indicating to switch the first AI model.

[0401]

[0419] As an option, in an alternative implementation of switching or updating AI models in a dual-end model, assuming that the first AI model is deployed to a second network element, the determination of whether the first AI model needs to be switched or updated based on correspondence information may alternatively be performed by the first network element. For example, the second network element is the UE, the first AI model is deployed to the UE, and the first network element is an access network device. The access network device determines whether the first AI model on the UE side needs to be switched or updated based on first information reported by the UE (e.g., the estimated results of the first parameters). It should be noted that in a dual-end model, the decision by the access network device on whether to switch or update the first AI model on the UE side is, in practice, a determination of whether collaborative switching or collaborative updating is possible. If collaborative switching or collaborative updating is not possible, the access network device instructs the UE not to switch or update the first AI model. If it is determined that a collaborative switch or collaborative update is possible, the access network device instructs the UE to switch or update the first AI model.

[0402]

[0420] Figure 11 shows an example of the application of the method for switching or updating AI models in AI-CSI feedback according to this application.

[0403]

[0421] An example is used for explanation in which the first network element is the UE and the second network element is the access network device. The UE is deployed with the first AI model, and the access network device is deployed with the fourth AI model, and the first and fourth AI models are used in a matching scheme.

[0404]

[0422] 41: The UE and access network devices perform AI-CSI feedback based on the first AI model and the fourth AI model, respectively.

[0405]

[0423] 42:UE measures CSI-RS from access network devices to obtain channel measurement results.

[0406]

[0424] 43:UE obtains the estimated result of the first parameter based on the channel measurement results.

[0407]

[0425] 44:UE transmits the first piece of information to the access network device, which indicates the estimation result of the first parameter.

[0408]

[0426] 45: Optionally, access network devices can obtain compatibility information.

[0409]

[0427] 46: The access network device determines whether to switch or update the UE's first AI model based on the first information and corresponding information.

[0410]

[0428] 47: The access network device sends instruction information to the UE based on the determination result, and the instruction information instructs the UE to switch or update the first AI model.

[0411]

[0429] In one example, the access network device determines that the estimation result of the first parameter corresponds to the first value among N values, and the first value corresponds to the second AI model among M AI models. The access network device then determines whether the W stored AI models include an AI model that matches the second AI model.

[0412]

[0430] The access network device decides to switch the UE's first AI model if the determination result indicates that W AI models include a fifth AI model that matches the second AI model. In this case, the access network device sends information A to the UE, which instructs the UE to switch the first AI model to the second AI model. Furthermore, the access network device switches the fourth AI model to the fifth AI model. The access network device decides not to switch the UE's first AI model if the determination result indicates that W AI models do not include any AI models that match the second AI model.

[0413]

[0431] In another example, the access network device determines that the estimated result for the first parameter does not correspond to any of the N values. Based on the estimated result for the first parameter, the access network device determines whether to update the UE's first AI model. If the determination is that the UE's first AI model should be updated, the access network device sends information B to the UE, which instructs the UE to update the first AI model. In this case, the UE updates the first AI model to the third AI model, the access network device updates the fourth AI model to the sixth AI model, and the third AI model matches the sixth AI model.

[0414]

[0432] If the determination result is not to update the UE's first AI model, the access network device may not send any instructions to the UE until the access network device decides to switch or update the AI ​​model on the UE side based on the first information sent by the UE.

[0415]

[0433] 48: The UE switches or updates the first AI model based on instructions from the access network device.

[0416]

[0434] For example, the UE receives information A from the access network device. If information A indicates not to switch the first AI model, the UE does not switch the first AI model. If information A indicates to switch the first AI model, the UE switches the first AI model to the second AI model. In this case, the fourth AI model in the access network device is switched to the fifth AI model, and the fifth AI model matches the second AI model. In another example, the UE receives information B from the access network device. If information B indicates not to update the first AI model, the UE does not update the first AI model. For example, if information B indicates to update the first AI model, the UE updates the first AI model to the third AI model. In this case, the access network device updates the fourth AI model to the sixth AI model.

[0417]

[0435] In the embodiment shown in Figure 11, the access network device decides whether to switch or update the AI ​​model on the UE side and instructs the UE to do so. If the UE decides not to switch or update the AI ​​model, the access network device does not instruct the UE. The access network device sends information to the UE indicating that a switch or update should be performed only if it is necessary for the switch or update to be performed, and the UE switches or updates the AI ​​model on the UE side based on the information indicating that a switch or update should be performed. If the UE does not receive information from the access network device indicating that a switch or update should be performed, the UE does not switch or update its AI model.

[0418]

[0436] In the embodiment described above, the first network element can determine changes in the operating environment of the AI ​​model by monitoring the input of the AI ​​model, and can adapt to changes in the operating environment by determining in a timely manner whether the AI ​​model needs to be switched or updated.

[0419]

[0437] In this application, considering that the intermediate performance indicator of a dual-end model may also reflect the final network performance, e.g., network throughput performance, a solution is further proposed for monitoring the intermediate performance indicator of a dual-end model, and based on the monitoring results of the intermediate performance indicator, the network side and the UE side need to coordinately switch or update the AI ​​model. In this embodiment, network performance is the final performance, and the AI ​​model performance is the intermediate performance relative to the network performance. Therefore, the intermediate performance indicator may be applied to a unilateral AI model, e.g., CSI prediction or beam management, as described above, or to a bilateral model, e.g., CSI feedback. When the intermediate performance indicator is applied to a unilateral AI model, the input indicator of the AI ​​model is monitored to obtain the monitoring result. The input indicator of the AI ​​model is, for example, a first parameter, and the monitoring result is, for example, the estimation result of the first parameter. The estimation result of the first parameter is compared with the label of the AI ​​model to obtain the GCS. Whether to switch or update the AI ​​model is determined based on the GCS and a specified threshold T. Below, we will use a bilateral model as an example to illustrate the process of switching or updating AI models by monitoring intermediate performance indicators.

[0420]

[0438] Optional intermediate performance indicators may include, but are not limited to, one or more of the following: generalized cosine similarity (GCS), squared generalized cosine similarity (SGCS), cell throughput indicator, average / edge user perceived rate indicator, and similar indicators.

[0421]

[0439] Generally, higher complexity in an AI model indicates better robustness regarding the AI ​​model's adaptability to more complex environments. In this embodiment, a correspondence is established between the complexity of the AI ​​model and the AI ​​model itself, and as a result, the performance of the network on which the AI ​​model is deployed does not decrease or degrade with changes in the environment in which the AI ​​model is used. By monitoring intermediate performance indicators, if it is determined that the environment in which the AI ​​model is used has changed, an AI model of the corresponding complexity is selected based on the correspondence, ensuring that an AI model capable of meeting accuracy requirements is always used, further guaranteeing stable network performance.

[0422]

[0440] In this embodiment, the correspondence information indicates the correspondence between the AI ​​model and the complexity information of the AI ​​model. For example, the complexity information may be a complexity level. Different complexity levels correspond to different accuracies or accuracy ranges of the AI ​​model.

[0423]

[0441] An example is used where the intermediate performance indicator is SGCS. The correspondence information on the UE side is shown in Table 10, and the correspondence information on the access network device side is shown in Table 11. Tables 10 and 11 are examples showing the correspondence between the AI ​​model and the complexity of the AI ​​model. Using Table 10 as an example, AI model 1 corresponds to complexity level 1. AI model 2 corresponds to complexity level 2. If complexity level 2 is higher than complexity level 1, it indicates that AI model 2 is better suited to the usage environment. For example, in Table 10, complexity levels 1 through 3 gradually increase, and in Table 11, complexity levels 1 through 4 gradually increase: Table 10 [Table 10] Table 11 [Table 11]

[0424]

[0442] Figure 12 shows an example of how to switch between AI models in a dual-end model.

[0425]

[0443] For example, suppose the AI ​​model currently used by the UE is the first AI model, e.g., AI model 1 in Table 10, and the AI ​​model currently used on the network side is the fourth AI model, e.g., AI model a in Table 11. Both AI model 1 and AI model a correspond to complexity level 1; in other words, the first AI model matches the fourth AI model.

[0426]

[0444] Based on this, the process of switching or updating the AI ​​model will be explained with reference to Figure 12.

[0427]

[0445] 51: Optionally, the UE obtains correspondence information, which shows the correspondence between the AI ​​model on the UE side and the complexity information of the AI ​​model. Table 10 is used as an example. The correspondence information shows three complexity levels and the AI ​​models corresponding to the three complexity levels, specifically AI Model 1 to AI Model 3.

[0428]

[0446] 52:UE measures the reference signal from the access network device to obtain channel measurement results.

[0429]

[0447] 53:UE compresses the channel measurement results based on the first AI model, or compresses and quantizes the channel measurement results based on the first AI model to obtain the feedback channel measurement results.

[0430]

[0448] 54:UE sends the feedback channel measurement results to the access network device.

[0431]

[0449] The access network device receives channel measurement results fed back from the UE.

[0432]

[0450] 55: The access network device processes the feedbacked channel measurement results based on the fourth AI model to obtain the restored channel measurement results.

[0433]

[0451] 56: The access network device decides whether to switch or update the fourth AI model based on the recovered channel measurement results and validation dataset.

[0434]

[0452] The access network device reconstructs the feedbacked channel measurement results based on the fourth AI model to obtain the reconstructed channel measurement results. The validation dataset contains the labels of the fourth AI model, which are specifically the channel measurement results obtained through the measurements. The degree of difference between the reconstructed channel measurement results and the true values ​​may be determined by comparing the reconstructed channel measurement results with the true values. The degree of difference may be indicated by SGCS or GCS. Optionally, the degree of difference may be indicated by another error indicator. GCS is used as an example. A larger GCS indicates that the reconstructed channel measurement results are closer to the true values ​​and that the performance of the fourth AI model is better. Conversely, a smaller GCS indicates that the difference between the reconstructed channel measurement results and the true values ​​is large and that the performance of the fourth AI model is poor. In this way, it is possible to decide whether to switch or update the fourth AI model.

[0435]

[0453] For example, in an embodiment that monitors intermediate performance indicators, the switching conditions may be as follows: the monitoring result of the intermediate performance indicator is less than a specified threshold T, and the AI ​​model library on the access network device side includes an AI model with a higher complexity than the AI ​​model currently in use.

[0436]

[0454] The update conditions may be as follows: the monitoring result of the intermediate performance indicator is less than the specified threshold T, and the AI ​​model library on the access network device side does not contain any AI models with a complexity higher than the complexity of the AI ​​model currently in use.

[0437]

[0455] It should be understood that, in the aforementioned switching or update conditions, the fact that the monitoring result of the intermediate performance indicator is less than the specified threshold T is used merely as an example, and the specific setting of the switching or update conditions may depend on the intermediate performance indicator. For example, the switching or update condition may alternatively be that the monitoring result of the intermediate performance indicator is greater than or equal to the threshold T, or that the monitoring result of the intermediate performance indicator and the specified threshold T satisfy another specified relationship. This is not limited to the present application.

[0438]

[0456] In this embodiment, an example is used where the intermediate performance indicator is SGCS or GCS. If the value of SGCS or GCS is greater than the specified threshold T, it indicates that the accuracy error of the currently used AI model is within an acceptable range, and a switchover or update may not be performed. If the value of SGCS or GCS is less than or equal to the specified threshold T, it indicates that the accuracy error of the currently used AI model is beyond an acceptable range. In this case, the AI ​​model needs to be switched over or updated. The specific AI model switchover or update depends on whether an AI model for the switchover exists. If the model library on the access network device side contains an AI model for the switchover, it is determined that the AI ​​model should be switched over. If the model library on the access network device side does not contain an AI model for the switchover, it is determined that the AI ​​model should be updated. For example, suppose the intermediate performance indicator is SGCS and the set threshold T is 0.95. If the monitoring result of the intermediate performance indicator is less than the threshold T, it is determined that the switchover / update condition is met. Assume that the SGCS value of the validation dataset calculated by the access network device is 0.96, which is greater than the threshold 0.95. In this case, the access network device decides not to switch or update the fourth AI model. If the SGCS value calculated by the access network device is 0.91, which is less than the threshold T, the access network device decides to switch or update the fourth AI model, depending on whether the model library on the access network device side contains an AI model for switching.

[0439]

[0457] Figure 12 uses an example where the access network device decides to perform a switch. As mentioned above, the AI ​​model currently used by the access network device is AI model a, which corresponds to complexity level 1. If the SGCS value is determined to be less than threshold T, the access network device determines, based on correspondence information 1, that the model library on the access network device side contains AI models with a higher complexity than AI model a, and the access network device decides to switch to AI model a.

[0440]

[0458] 57: The access network device sends a failover request to the UE.

[0441]

[0459] A switch request indicates that an access network device is requesting a switch to a different AI model.

[0442]

[0460] 58:UE decides whether to switch AI models based on the switch request and corresponding information.

[0443]

[0461] Optionally, correspondence information is predefined in the protocol. For example, both correspondence information 1 and correspondence information 2 are predefined in the protocol. In this case, a switching request sent by the access network device may carry the identifier of the AI ​​model to which the access network device is requesting to switch. For example, the switching request carries the identifier of AI model b or the identifier of AI model 2. Based on the switching request and the identifier of AI model b or AI model 2, the UE can know that the access network device is requesting a switch to the currently used AI model.

[0444]

[0462] Optionally, correspondence information 2 is stored on the access network device side, and correspondence information 1 is stored on the UE side. The complexity levels and quantities of AI models used on the access network device side and the UE side are either predefined or negotiated. In this case, a switching request sent by the access network device can carry complexity information of the AI ​​model to which the access network device is requesting to switch, for example, information indicating complexity level 2. Based on the switching request and the AI ​​model complexity information, the UE side can know that the access network device is requesting to switch the current AI model, and can determine the corresponding AI model to be used for the switch on the UE side, for example, AI model 2 corresponding to complexity level 2.

[0445]

[0463] Optionally, correspondence information 2 is stored on the access network device side, and correspondence information 1 is stored on the UE side. The complexity levels and number of AI models used on the access network device side and the UE side are either predefined or determined by negotiation. For example, the AI ​​model switching rules in application scenarios (e.g., CSI feedback) on the UE side and the access network device side are either default or determined by negotiation. Here, the switching rules are as follows: when an AI model is switched, the complexity level of the AI ​​model is switched gradually, and the switch does not occur across multiple complexity levels. For example, if the currently used AI model a does not meet the accuracy requirements and AI model b corresponds to complexity level 2, AI model a will, by default, be switched to AI model b which corresponds to complexity level 2, and AI model a will not be switched to an AI model with a higher complexity level than complexity level 2. In this implementation, the access network device may only send switching requests to the UE. Based on the switch request, the UE can determine that the accuracy of the currently used first AI model does not meet the accuracy requirements and, by default, can switch AI model 1 to AI model 2.

[0446]

[0464] Naturally, the AI ​​model may be switched across multiple complexity levels as an alternative. For example, see the aforementioned example where the access network device displays the AI ​​model complexity information to the UE, or where the access network device displays the AI ​​model identifier to the UE, if the discrepancy between the monitoring results of the intermediate performance indicator and the accuracy requirements is excessively large. Further details will not be provided again.

[0447]

[0465] The above uses a switchover as an example, and the process for updating AI models is similar. If the access network device determines that none of the AI ​​models in the access network's AI model library meet the current accuracy requirements, it can send an update request to the UE. For example, if the access network device determines that the AI ​​model with the highest complexity level in the access network's model library does not meet the current accuracy requirements, it will decide to update the AI ​​model.

[0448]

[0466] Furthermore, the switching from low complexity to high complexity is used as an example in the aforementioned example. The complexity of the AI ​​model may be considered when switching from a complex environment scenario to a simpler environment scenario. A high-complexity AI model can be switched to a low-complexity AI model if it is possible to guarantee accuracy requirements. This is not limited to this case.

[0449]

[0467] If, in step 58, it is decided not to switch the first AI model, the UE sends an instruction to the access network device indicating that it will not perform the switch. In this case, neither the UE nor the access network device will perform the AI ​​model switch. Figure 12 does not show the case where the switch does not occur, but only the case where the switch is required, as shown in step 59.

[0450]

[0468] 59:UE sends information to access network devices indicating that a failover is to be performed.

[0451]

[0469] The access network device switches from the fourth AI model to the fifth AI model based on information indicating that a switchover is to be performed from the UE. Furthermore, the UE switches from the first AI model to the second AI model to complete the coordinated AI model switchover.

[0452]

[0470] In the example in Figure 12, timely switching or updating of the AI ​​model may be possible by monitoring the intermediate performance indicators of the dual-end model to adapt to changes in the environment in which the AI ​​model is used. This helps to reduce or avoid the impact on the performance of the network in which the AI ​​model is deployed.

[0453]

[0471] In the embodiment shown in Figure 12, the switching of AI models is primarily used as an illustrative example. Those skilled in the art will be able to understand, based on the aforementioned example in which the AI ​​model is updated by monitoring the AI ​​model's input, an implementation of updating the AI ​​model by monitoring intermediate performance indicators. Further details are not described here.

[0454]

[0472] The above describes in detail the method for switching or updating the AI ​​model provided in this application. The corresponding communication device is described below. Please refer to Figure 13. This application provides a communication device 1000.

[0455]

[0473] As shown in Figure 13, the communication device 1000 includes a processing module 1001 and a communication module 1002. The communication device 1000 may be a terminal device, or it may be a communication device such as a chip, chip system, or circuit that is used with a terminal device or used in a matching manner with a terminal device and is capable of implementing methods performed on the terminal device side. Alternatively, the communication device 1000 may be a network device, or it may be a communication device such as a chip, chip system, or circuit that is used with a network device or used in a matching manner with a network device and is capable of implementing methods performed on the network device side. For example, the network device may be an access network device in an embodiment of the method of the present application.

[0456]

[0474] A communication module may also be referred to as a transceiver module, transceiver, transceiver device, or transceiver apparatus. A processing module may also be referred to as a processor, processing board, processing unit, processing unit, or similar. Optionally, a communication module is configured to perform transmit and receive operations on the terminal device side or the network device side in the manner described above. A component configured to perform the receive function in a communication module may be considered a receive unit, and a component configured to perform the transmit function in a communication module may be considered a transmit unit. In other words, a communication module includes a receive unit and a transmit unit.

[0457]

[0475] When the communication device 1000 is used in a terminal device, the processing module 1001 may be configured to perform the processing functions of the terminal device in the embodiments shown in Figures 3 to 12, and the communication module 1002 may be configured to perform the receiving and transmitting functions of the terminal device in the embodiments shown in Figures 3 to 12.

[0458]

[0476] When the communication device 1000 is used in a network device, the processing module 1001 may be configured to perform the processing functions of the network device (e.g., an access network device) in the embodiments shown in Figures 3 to 12, and the communication module 1002 may be configured to perform the receiving and transmitting functions of the network device in the embodiments shown in Figures 3 to 12.

[0459]

[0477] The first or second network element shown in Figures 3, 4, 9, and 10 may specifically be a terminal device or a network device (e.g., an access network device), which should be noted as being described in detail in embodiments of the method described above. To understand that the first or second network element is a terminal device or a network device, it is possible to refer to specific embodiments. Further details are not described here.

[0460]

[0478] In addition, it should be noted that communication modules and / or processing modules may be implemented using virtual modules. For example, a processing module may be implemented using a software function unit or virtual device, and a communication module may be implemented using a software function or virtual device. Alternatively, a processing module or communication module may be implemented using an entity device. For example, if the device is implemented using a chip / chip circuit, the communication module may be an input / output circuit and / or a communication interface, performing input operations (corresponding to the receiving operations described above) and output operations (corresponding to the transmitting operations described above). The processing module is an integrated processor, microprocessor, or integrated circuit.

[0461]

[0479] The division into modules in this application is merely an example and represents a division into logical functions; other divisions are possible in actual implementation. Furthermore, the functional modules in the example of this application may be integrated into a single processor, and each module may exist physically independently, or two or more modules may be integrated into a single module. The integrated module may be implemented in hardware form or in the form of a software functional module.

[0462]

[0480] Please refer to Figure 14, based on the same technical concept. This application further provides a communication device 1100. Optionally, the communication device 1100 may be a chip or a chip system. Optionally, in this application, a chip system may include a chip, or include a chip and other individual components.

[0463]

[0481] The communication device 1100 may be configured to perform the functions of any network element in the communication system described in the above example. The communication device 1100 may include at least one processor 1110. Optionally, the processor 1110 is coupled to memory. The memory may be located within the device. Alternatively, the memory may be integrated with the processor. Or, the memory may be located outside the device. For example, the communication device 1100 may further include at least one memory 1120. The memory 1120 stores computer programs, computer programs or instructions, and / or data necessary to perform any of the above examples. The processor 1110 may execute the computer programs stored in the memory 1120 to complete any of the methods in the above examples.

[0464]

[0482] The communication device 1100 may further include a communication interface 1130, which may exchange information with other devices via the communication interface 1130. For example, the communication interface 1130 may be a transceiver, a circuit, a bus, a module, a pin, or another type of communication interface. If the communication device 1100 is a chip-type device or circuit, the communication interface 1130 within the device 1100 may alternatively be an input / output circuit that inputs information (or is referred to as receiving information) and outputs information (or is referred to as transmitting information). The processor may be an integrated processor, a microprocessor, an integrated circuit, or a logic circuit. The processor may determine output information based on input information.

[0465]

[0483] The coupling in this application may be an indirect coupling or communication connection between devices, units, or modules in an electrical, mechanical, or other form, and is used for information exchange between devices, units, or modules. The processor 1110 may work in cooperation with the memory 1120 and the communication interface 1130. The specific connection medium between the processor 1110, the memory 1120, and the communication interface 1130 is not limited in this application.

[0466]

[0484] Optionally, as shown in Figure 14, the processor 1110, memory 1120, and communication interface 1130 are connected to each other via bus 1140. Bus 1140 may be a peripheral component interconnect (PCI) bus, an extended industry standard architecture (EISA) bus, or similar. The bus can be divided into an address bus, a data bus, a control bus, and similar. For ease of representation, only one thick line is used to represent the bus in Figure 14, but this does not mean that there is only one bus or only one type of bus.

[0467]

[0485] In this application, the processor may be a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field-programmable gate array or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component, which may implement or carry out the methods, steps, and logic block diagrams disclosed in this application. The general-purpose processor may be a microprocessor, any conventional processor, or similar. The steps of the methods disclosed in connection with this application may be carried out directly by the hardware processor or by a combination of hardware and software modules within the processor.

[0468]

[0486] In this application, memory may be non-volatile memory, such as a hard disk drive (HDD) or solid-state drive (SSD), or volatile memory, such as random access memory (RAM). Memory may also be, but not limited to, any other medium that can be used to carry or store expected program code in the form of instructions or data structures and that is accessible by a computer. Alternatively, memory in this application may be a circuit or any other device capable of performing a storage function and configured to store program instructions and / or data.

[0469]

[0487] In possible implementations, the communication device 1100 may be used on the network device side, for example, an access network device, a core network device, or a host or cloud device in an OTT system, as in the embodiments of this application. Specifically, the communication device 1100 may be a network device, or a device capable of supporting a network device when performing the corresponding function on the network device side in any of the examples described above. Memory 1120 stores computer programs (or instructions) and / or data for performing the functions on the network device side in any of the examples described above. Processor 1110 can execute the computer programs stored in memory 1120 to complete the methods performed by the network device side in any of the examples described above. The communication interface within the communication device 1100 may be configured to interact with a terminal device and send information to or receive information from a terminal device. Furthermore, as an option, the communication interface within the communication device 1000 may be further configured to interact with another network element (e.g., a third network element) to, for example, obtain correspondence information from the third network element or obtain an AI model from the third network element.

[0470]

[0488] In another possible implementation, the communication device 1100 may be used as a terminal device. Specifically, the communication device 1100 may be a terminal device, or a device capable of supporting a terminal device when performing the functions of a terminal device in any of the examples described above. Memory 1120 stores computer programs (or instructions) and / or data for performing the functions of a terminal device in any of the examples described above. Processor 1110 can execute the computer programs stored in memory 1120 to complete the methods performed by the terminal device in any of the examples described above. The communication interface within the communication device 1100 may be configured to interact with a network device (e.g., an access network device) and to transmit information to or receive information from an access network device.

[0471]

[0489] The communication device 1100 provided in this example may be used on the network device side (e.g., an access network device) to complete the methods performed by the network device side, or it may be used on a terminal device to complete the methods performed by the terminal device. Therefore, for the technical effects that can be achieved by this embodiment, please refer to the embodiments of the method described above. Details will not be described again here.

[0472]

[0490] Based on the examples described above, this application provides a communication system. In one example, the communication system includes a first network element and a second network element. In another example, the communication system includes a first network element, a second network element, and a third network element. The communication system may implement a method for switching or updating AI models provided in the embodiments shown in Figures 3 to 12.

[0473]

[0491] All or part of the technical solutions provided in this application may be implemented by using software, hardware, firmware, or any combination thereof. If software is used to implement the technical solutions, all or part of the technical solutions may be implemented in the form of a computer program product. A computer program product includes one or more computer instructions. When computer program instructions are loaded onto a computer and executed, the procedures or functions of this application occur, in whole or in part. The computer may be a general-purpose computer, a dedicated computer, a computer network, a terminal device, an access network device, or another programmable device. Computer instructions may be stored on a computer-readable storage medium, or the transmission of computer instructions may be performed from one computer-readable storage medium to another. For example, the transmission of computer instructions may be performed from one website, computer, server, or data center to another website, computer, server, or data center by a wired (e.g., coaxial cable, optical fiber, or digital subscriber line (DSL)) or wireless (e.g., infrared, radio, or microwave) method. Computer-readable storage media may be any available media accessible by a computer, or a data storage device that integrates one or more available media, such as a server or data center. Available media may be magnetic media (e.g., floppy disks, hard disk drives, or magnetic tapes), optical media (e.g., digital video discs (DVDs)), semiconductor media, or similar.

[0474]

[0492] In this application, cross-referencing is permitted between specific examples, provided there is no logical inconsistency. For example, cross-referencing is permitted between methods and / or terms in embodiments of methods, between functions and / or terms in embodiments of apparatus, and between functions and / or terms in examples of apparatus and examples of methods.

[0475]

[0493] Units described as separate parts may or may not be physically separate, and parts illustrated as units may or may not be physical units, may be located in one place, or may be distributed across multiple network units. In order to achieve the objectives of the solutions of the embodiments, all or part of the units may be selected based on the actual requirements.

[0476]

[0494] Furthermore, the functional units in the embodiments of this application may be integrated into a single processing unit, or each unit may exist physically independently, or two or more units may be integrated into a single unit.

[0477]

[0495] When a function is implemented in the form of a software function unit and sold or used as an independent product, the function can be stored on a computer-readable storage medium. Based on such understanding, the technical solutions in this application, or the portion that contributes to the prior art, or a portion of the technical solutions, may be implemented in the form of a software product. A computer software product includes several instructions for instructing a computer device (which may be a personal computer, server, network device, or similar) to perform all or part of the steps of the method described in the embodiments of this application, which are stored on a storage medium. The aforementioned storage medium includes any medium capable of storing program code, such as a USB flash drive, removable hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0478]

[0496] A person skilled in the art can make various modifications and alterations to the present application without departing from its scope. The present application is intended to encompass these modifications and alterations, provided that they fall within the scope of protection defined by the claims of the present application and its equivalent art.

Claims

1. A method for switching or updating an artificial intelligence (AI) model, which is performed by a first network element or a chip of the first network element, wherein the method is: A step of obtaining first information, wherein the first information represents the estimation result of a first parameter, the estimation result of the first parameter is based on a channel measurement result, and the channel measurement result is the input to an AI model; and A step of determining whether to switch or update a first AI model based on correspondence information and the first information, wherein the first AI model is deployed in the first network element or the second network element, and the correspondence information indicates the correspondence between M AI models and N values ​​of the first parameter, where M is an integer greater than or equal to 1 and N is an integer greater than or equal to 1; A method that includes this.

2. The method according to claim 1, the step of determining whether to switch or update the first AI model based on the correspondence information and the first information is: A step of determining that the estimation result of the first parameter corresponds to a first value among the N values, wherein the first value corresponds to a second AI model among the M AI models; and The step of deciding to switch from the first AI model to the second AI model; A method that includes this.

3. The method according to claim 1, the step of determining whether to switch or update the first AI model based on the correspondence information and the first information is: The steps include determining that the estimation result of the first parameter does not correspond to any of the N values; and Steps include deciding to update the first AI model; A method that includes this.

4. A method according to any one of claims 1 to 3, wherein the first AI model is deployed on the first network element, and the first network element is a terminal device.

5. In the method of claim 4, after deciding to switch the first AI model to the second AI model based on the correspondence information and the first information, the method further: A step of transmitting the first instruction information to the second network element; A method comprising the first instruction information indicating that the first network element is requesting to switch the first AI model to the second AI model.

6. The method according to claim 5, further: The steps of receiving the second AI model from the second network element; and Steps to switch from the first AI model to the second AI model; A method that includes this.

7. In the method of claim 4, after deciding to update the first AI model based on the correspondence information and the first information, the method further: A step of transmitting second instruction information to the second network element; A method comprising the second instruction information being used to request the updating of the first AI model.

8. In the method according to any one of claims 4 to 7, the step of obtaining the first information is: A step of measuring a reference signal and obtaining the channel measurement result; and A step of obtaining an estimation result of the first parameter based on the channel measurement result; A method comprising, wherein the first information includes the estimation result of the first parameter.

9. In the method according to claim 8, prior to the step of deciding whether to switch or update the first AI model based on the correspondence information and the first information, the method further: A step of obtaining all or part of the correspondence information from the second or third network element; A method that includes this.

10. A method according to any one of claims 1 to 3, wherein the first network element is a network element, the first AI model is deployed on the second network element, and the second network element is a terminal device.

11. In the method according to claim 10, the step of obtaining the first information is: Steps include receiving the first information from the second network element; A method comprising, wherein the first information includes an estimated result of a first parameter, or the first information includes information used to determine the estimated result of the first parameter, and the estimated result of the first parameter is based on a channel measurement result obtained by measuring a reference signal on the side of the second network element.

12. In the method according to claim 10 or 11, if it is decided to switch the first AI model to the second AI model based on the correspondence information and the first information, the method further: The step of sending the second AI model to the second network element; Methods that include...

13. In the method according to claim 11 or 12, if it is decided to update the first AI model based on the correspondence information and the first information, the method further: Steps to obtain training data; A step of obtaining a third AI model by performing AI model training based on the aforementioned training data; and The third AI model is transmitted to the second network element; Methods that include...

14. In the method of claim 4, after deciding to switch the first AI model based on the correspondence information and the first information, the method further: A step of transmitting first instruction information to the second network element, wherein the first instruction information indicates that the first network element is requesting to switch the first AI model; A step of receiving third instruction information from the second network element, wherein the third instruction information instructs the first network element to switch the first AI model, or the third instruction information instructs the first network element not to switch the first AI model; and A step of switching the first AI model to the second AI model based on the third instruction information, or skipping the step of switching the first AI model based on the third instruction information; Methods that include...

15. In the method of claim 4, after deciding to update the first AI model based on the correspondence information and the first information, the method further: A step of transmitting a second instruction to the second network element, wherein the second instruction indicates that the first network element is requesting that the first AI model be updated; A step of receiving a fourth instruction information from the second network element, wherein the fourth instruction information instructs the first network element to update the first AI model, or the fourth instruction information instructs the first network element not to update the first AI model; and A step of switching the first AI model to the third AI model based on the fourth instruction information, or skipping the step of updating the first AI model based on the fourth instruction information; Methods that include...

16. A method according to any one of claims 1 to 13, wherein the AI ​​model is applied to CSI prediction or beam management.

17. A method according to claim 14 or 15, wherein the AI ​​model is applied to CSI feedback.

18. In the method according to any one of claims 4 to 17, the first parameter is: The movement speed of the aforementioned terminal device; Channel signal-to-interference ratio plus noise ratio (SINR); or A parameter that reflects the degree of non-line-of-sight (NLOS) in the channel; A method that includes one or more of the following.

19. A method according to any one of claims 8 to 11, wherein the reference signal is a channel state information-reference signal CSI-RS.

20. The method according to claim 11, wherein a fourth AI model is deployed on the first network element, the fourth AI model and the first AI model are used in a matching manner, and the step of deciding whether to switch the first AI model based on the correspondence information and the first information is: A step of determining that the estimation result of the first parameter corresponds to a first value among the N values, wherein the first value corresponds to a second AI model among the M AI models; The steps of determining whether the W stored AI models deployed in the first network element include an AI model that matches the second AI model; and A step of deciding whether to switch the first AI model based on the judgment result; Methods that include...

21. A method for switching or updating an AI model, which is performed by a second network element or a chip of the second network element, wherein the method is: Steps include obtaining channel measurement results fed back from the first network element; A step of restoring the feedbacked channel measurement results based on a fourth AI model to obtain the restored channel measurement results; and A step of determining whether to switch or update the fourth AI model based on the feedbacked channel measurement results and the restored channel measurement results; A method that includes this.

22. The method according to claim 21, wherein the feedback channel measurement result is obtained by the first network element by processing the channel measurement result obtained by measurement based on the first AI model.

23. The method according to claim 21 or 22, wherein the step of determining whether to switch or update the fourth AI model based on the feedbacked channel measurement results and the restored channel measurement results is: A step of determining the value of the error indicator of the validation dataset based on the feedbacked channel measurement results, the restored channel measurement results, and the validation dataset; and The step of deciding to switch or update the fourth AI model if the value of the error indicator meets the specified conditions; A method that includes this.

24. A method according to claim 23, wherein the specified condition is: the value of the error indicator is greater than or equal to a specified threshold T.

25. The method according to claim 23 or 24, the step of deciding to switch or update the fourth AI model if the value of the error indicator meets a specified condition is: The step of deciding to switch the fourth AI model if the value of the error indicator meets the specified conditions, and the Q AI models stored in the second network element include a fifth AI model, and the complexity of the fifth AI model is higher than the complexity of the fourth AI model currently in use; or The step of deciding to update the fourth AI model if the value of the error indicator meets the specified conditions and the Q AI models stored in the second network element do not have a higher complexity than the currently used fourth AI model; Methods that include...

26. In the method according to any one of claims 21 to 25, if it is decided to switch or update the fourth AI model, the method further: A step of sending a switching request to the first network element, wherein the switching request is indicating that it is requesting to switch the fourth AI model; The steps include receiving information from the first network element indicating that a switchover should be performed, and A step of switching the fourth AI model to the fifth AI model based on information indicating that the aforementioned switch should be performed; A method that includes this.

27. In the method according to any one of claims 21 to 25, if it is decided to update the fourth AI model, the method further: A step of sending an update request to the first network element, wherein the update request is used to request an update to the fourth AI model; The steps include receiving information from the first network element indicating that an update should be performed, and A step of updating the fourth AI model to the sixth AI model based on information indicating that the aforementioned update should be performed; Methods that further include the above.

28. A method for switching or updating an AI model, which is performed by a first network element or a chip of the first network element, wherein the method is: A step of receiving a switch request from a second network element, wherein the switch request indicates that the second network element is requesting a switch on the AI ​​model; and A step of determining whether to switch the currently used first AI model based on correspondence information and the switching request, wherein the correspondence information indicates the correspondence between Q AI models and Q complexity information of the AI ​​models, the complexity information corresponding to the first AI model indicates a first complexity level, and Q is an integer greater than or equal to 1; Methods that include...

29. The method according to claim 28, the step of determining whether to switch the currently used first AI model based on the correspondence information and the switching request is: A step of determining, based on the switching request, whether the Q AI models include an AI model corresponding to a second complexity level, wherein the second complexity level is higher than the first complexity level; and The step of deciding to switch the first AI model to the second AI model if the Q AI models include a second AI model corresponding to a second complexity level; and The method further includes: The steps include sending information to the second network element indicating that a switchover is to be performed; and Steps to switch from the first AI model to the second AI model; A method that includes this.

30. The method according to claim 28, the step of determining whether to switch the currently used first AI model based on the correspondence information and the switching request is: A step of determining, based on the switching request, whether the Q AI models include an AI model corresponding to a second complexity level, wherein the second complexity level is higher than the first complexity level; and If none of the Q AI models include a second AI model corresponding to a second complexity level, the step of deciding not to switch the first AI model; and The method further includes: A step of sending instruction information to the first network element indicating that the switch will not be performed; A method that includes this.

31. A method for switching or updating an AI model, which is performed by a second network element or a chip of the first network element, wherein the method is: Steps include sending correspondence information to the first network element; The correspondence information includes a correspondence between M AI models and N values ​​of a first parameter, where M is an integer greater than or equal to 1 and N is an integer greater than or equal to 1; or The correspondence information indicates the correspondence between Q AI models and R complexity information of the AI ​​models, where the R complexity information represents different complexity levels, and both Q and R are integers greater than or equal to 1.

32. The method according to claim 31, further: A method further comprising the step of transmitting second information to the first network element, wherein the second information includes the first parameter.

33. A communication device configured to perform the method described in any one of claims 1 to 32.

34. A communication device including a processor, wherein the processor is coupled to a memory, and the processor is configured to call computer program instructions stored in the memory to perform the method according to any one of claims 1 to 32.

35. A communication device comprising a processor and a communication interface, wherein the communication interface is configured to receive data and / or information and to transmit the received data and / or information to the processor; the processor processes the data and / or information; and the communication interface outputs the data and / or information processed by the processor, and the communication device is configured to perform the method according to any one of claims 1 to 32.

36. A computer-readable storage medium for storing instructions, wherein when the instructions are executed on the computer, the computer is able to perform the method according to any one of claims 1 to 32.

37. A computer program product wherein a computer-readable storage medium stores instructions, and when the instructions are executed on a computer, the computer becomes capable of performing the method according to any one of claims 1 to 32.

38. A communication system including a communication device as described in any one of claims 33 to 35.

Images