Communication method and related apparatus

Abstract

A communication method and a related apparatus. In the method, when different models deployed in different communication apparatuses are used for jointly completing the same AI task, on the basis of whether the performances of the different models meet conditions (or do not meet condition), a first communication apparatus may determine whether to adjust the different models, so as to implement model management of the models. In some implementations, the first communication apparatus may, on the basis of first information and second information, determine third information used for adjusting a first model and / or adjusting a second model, so that the foregoing solution can be applied to a scenario in which a plurality of distributed nodes cooperatively process the same AI task. In addition, a faulty node can be accurately determined on the basis of whether the performances of the different models meet conditions (or do not meet conditions), so that impacts on an entire distributed system are prevented, and a redundant operation caused by adjustment of a non-faulty device is prevented, thereby improving the efficiency of model management.

Description

Communication method and related apparatus

[0001] This application claims priority from the Chinese patent application No. 202410895725.5, filed on July 4, 2024, and entitled "A communication method and related apparatus", the content of which is incorporated herein by reference in its entirety. TECHNICAL FIELD

[0002] The present application relates to the field of communication, and in particular to a communication method and related apparatus. BACKGROUND

[0003] With the development of communication technology, in a communication system, in addition to the traditional communication service, the service performed by the communication device can also include other new services, such as artificial intelligence (AI) service. Generally, the communication system capable of processing the AI service can also be referred to as an AI system.

[0004] Currently, the communication device can be used as a participating node of the AI system, and the computing power of the communication device is applied to a certain link of the AI system. Generally, the AI function introduced in the communication network needs to rely on a model to be implemented. However, in the above process, how to manage the model, there is currently no related scheme to solve it. SUMMARY

[0005] The present application provides a communication method and related apparatus for implementing model management.

[0006] The first aspect of the present application provides a communication method, which is performed by a first communication device. The first communication device can be a communication apparatus (e.g., a terminal device or a network device), or the first communication device can be a part of the communication apparatus (e.g., a circuit or a chip responsible for communication functions (e.g., a Modem chip, also known as a baseband chip, or a system on chip (SoC) chip or a system in package (SIP) chip containing a modem core), etc.), or the first communication device can also be a logic module or software capable of implementing all or part of the functions of the communication apparatus. In the method, the first communication device obtains first information, which is used to determine whether the performance of a first model deployed on the first communication device meets a first condition; the first communication device receives second information, which is used to determine whether the performance of a second model deployed on a second communication device meets a second condition; wherein the second model and the first model are used to jointly complete an AI task; and the first communication device determines third information based on the first information and the second information, the third information being used to indicate adjustment of the first model and / or adjustment of the second model.

[0007] Based on the above scheme, the first communication device can determine, through the obtained first information, whether the performance of the first model deployed on the first communication device meets the first condition, and determine, through the received second information, whether the performance of the second model deployed on the second communication device meets the second condition. Thereafter, the first communication device can determine, based on the first information and the second information, third information used to adjust the first model and / or adjust the second model. Wherein the second model and the first model are used to jointly complete an AI task. In this way, in the case that different models deployed on different communication devices are used to jointly complete the same AI task, the first communication device can determine whether to adjust the different models based on the condition that the performance of the different models meets (or does not meet) the condition, so as to achieve model management of the models.

[0008] In addition, in the above scheme, the first communication device can determine, based on the first information and the second information, third information used to adjust the first model deployed on the first communication device and / or adjust the second model deployed on the second communication device. To this end, the above scheme can be applied to a scenario in which a plurality of nodes cooperatively process the same AI task in a distributed manner, the plurality of nodes at least including the first communication device and the second communication device, and the condition that the performance of the different models meets (or does not meet) the condition can be used to accurately determine the node that fails, so as to avoid affecting the entire distributed system, and avoid redundant operations caused by adjusting the fault-free device, so as to improve the efficiency of model management.

[0009] In this application, the AI task can be replaced by other terms, such as task, AI use case, AI function, AI-enabled function, or AI-enabled feature, etc.

[0010] In this application, the model can include a mathematical model, an AI model, a neural network model, an AI neural network model, a machine learning model, or an AI processing model, etc.

[0011] In this application, one model is deployed in one communication device (for example, the first model is deployed in the first communication device, and the second model is deployed in the second communication device, etc.), which can be understood as that the one communication device obtains the model parameters of the one model, and then obtains / generates / constructs the one model based on the model parameters of the one model. Subsequently, the one communication device can perform model adjustment on the one model.

[0012] Optionally, the model parameters can include one or more of the hyperparameters of the model, the dataset of the model (including the input data of the model and the label data corresponding to the input data, etc.), the structure of the model, the weight of the model, and the parameters of the model.

[0013] It should be understood that two or more models are used to jointly complete an AI task, which can be understood as that these models complete the same AI task through cooperation. The processing of these models can be serial (or chain), parallel, or a combination of serial and parallel. For example, the execution result of the AI task can be obtained based on the outputs of these models.

[0014] Optionally, adjusting the model (for example, adjusting the first model and / or adjusting the second model) can include one or more of resetting the model, retraining the model, initializing the model, training the model, updating the model, fine-tuning the model, or fine-tuning the model.

[0015] Optionally, the first model and the second model can have different association relationships (or cooperation relationships). Some examples will be provided below, taking the first model and the second model as examples for illustrative description.

[0016] Implementation example one, the first model and the second model are different parts of the models used to jointly complete the AI task.

[0017] In implementation example one, the model for jointly completing the AI task can include a first model and a second model. For example, the output of the first model can be part or all of the input of the second model, the output of the second model can be obtained through processing of the second model, and the execution result of the AI task can include the output of the second model (optionally, the output of the first model can also be included). For another example, the output of the second model can be part or all of the input of the first model, the output of the first model can be obtained through processing of the first model, and the execution result of the AI task can include the output of the first model (optionally, the output of the second model can also be included).

[0018] In implementation example two, the first model and the second model are partially the same.

[0019] In implementation example two, the first model can include the model for jointly completing the AI task, and the second model can include a part of the model for jointly completing the AI task. Alternatively, the second model can include the model for jointly completing the AI task, and the first model can include a part of the model for jointly completing the AI task.

[0020] For example, the first model can include one or more sub-models, and the second model is one of the one or more sub-models, i.e., the second model is a sub-module (or a subset) of the first model; in this case, the first model can include the model for jointly completing the AI task, and the second model can include a part of the model for jointly completing the AI task. In other words, the first model can be referred to as a complete model (or a complete model for jointly completing the AI task), and the second model can be referred to as a partial model (or a partial model for jointly completing the AI task).

[0021] For another example, the second model can include one or more sub-models, and the first model is one of the one or more sub-models, i.e., the first model is a sub-module (or a subset) of the second model; in this case, the second model can include the model for jointly completing the AI task, and the second model can include a part of the model for jointly completing the AI task. In other words, the second model can be referred to as a complete model (or a complete model for jointly completing the AI task), and the first model can be referred to as a partial model (or a partial model for jointly completing the AI task).

[0022] In implementation example three, the first model and the second model are the same.

[0023] In an implementation example three, the first model and the second model can each comprise a model for jointly completing an AI task. In a performance of a certain joint AI task, the first model can participate in the performance partially or entirely, and the second model can also participate in the performance partially or entirely. Wherein, the first model and the second model can each complete the AI task independently. In other words, the first model and the second model can be referred to as complete models (or complete models for jointly completing the AI task).

[0024] It should be noted that the performance of a model (e.g., the performance of the first model, the performance of the second model, etc.) can be characterized by a plurality of parameters, and correspondingly, the condition corresponding to the performance of the model (e.g., the first condition corresponding to the first model, the second condition corresponding to the second model, etc.) can be a condition adapted to the parameter. Wherein, the condition can be implemented in a plurality of ways. For example, in way A, satisfying the condition can be understood as the performance of the model being better, and not satisfying the condition can be understood as the performance of the model being worse. For another example, in way B, satisfying the condition can be understood as the performance of the model being worse, and not satisfying the condition can be understood as the performance of the model being better.

[0025] As an example, the performance of a model can be determined based on a difference between an output of the model and an expected output (or a target output), and correspondingly, the condition corresponding to the performance of the model can indicate whether the difference is greater than a threshold value. Wherein, the greater the difference, the lower the performance of the model; on the contrary, the smaller the difference, the higher the performance of the model.

[0026] As another example, the performance of a model can be determined based on a communication performance of an output of the model applied to a communication process, and correspondingly, the condition corresponding to the performance of the model can be a threshold value of the communication performance. Wherein, the higher the communication performance, the higher the performance of the model; on the contrary, the lower the communication performance, the lower the performance of the model.

[0027] For example, the performance of the above-mentioned model is determined based on a difference between an output of the model and an expected output (or a target output). In the above-mentioned way B, satisfying the condition can be understood as the performance of the model being lower than (or equal to) the threshold value indicated by the condition, and not satisfying the condition can be understood as the performance of the model being higher than (or equal to) the threshold value indicated by the condition. In the above-mentioned way A, satisfying the condition can be understood as the performance of the model being higher than (or equal to) the threshold value indicated by the condition, and not satisfying the condition can be understood as the performance of the model being lower than (or equal to) the threshold value indicated by the condition.

[0028] Optionally, the difference can be represented by a mathematical calculation result of the different outputs. For example, the mathematical calculation result can include one or more of a difference, a mean square error (MSE), a normalized mean square error (NMSE), or a cosine similarity.

[0029] Optionally, the communication performance can be represented by a communication parameter. For example, the communication parameter can include one or more of a reference signal received power (RSRP), a reference signal received power quality (RSRQ), or a signal and interference plus noise ratio (SINR).

[0030] In a possible implementation of the first aspect, the first information is used to indicate a performance of the first model using a first parameter for data processing; and the first communication device obtaining the first information comprises: in a case where a performance of the first model using a second parameter for data processing is lower than (or equal to) a threshold, the first communication device obtaining the first information; wherein the first parameter is a parameter corresponding to the first model after a model adjustment operation is completed, and the second parameter is a parameter corresponding to the first model before the model adjustment operation is performed (or completed).

[0031] Based on the above scheme, in a case where a performance of the first model using a parameter corresponding to the first model before the model adjustment operation is performed (i.e., the second parameter) for data processing is lower than a threshold, the first communication device can be triggered to obtain the first information, and the first information is used to indicate a performance of the first model using a parameter corresponding to the first model after the model adjustment operation is completed for data processing. In this way, the first communication device can trigger performance detection of the parameter corresponding to the first model after the model adjustment operation based on the performance of the parameter corresponding to the first model before the model adjustment operation, and determine the first information based on the performance of the parameter corresponding to the first model after the model adjustment operation.

[0032] In addition, in a case where joint processing of the first model and the second model fails (for example, the performance of the joint processing is low), the above scheme can exclude an influence caused by the model adjustment of the first model, and through the first information, it can be represented whether the failure is caused by degradation of the model performance of the first model, so as to realize accurate positioning of the failure, and improve the efficiency of model management.

[0033] It should be understood that the performance of data processing of a model using a certain parameter (for example, the performance of data processing of the first model using the first parameter, the performance of data processing of the first model using the second parameter, and the like) can be understood as the performance of the output data corresponding to the model processing of the input data based on the model after the model parameter of the model is updated (or set) to the parameter. For example, the performance of the output data can be characterized by the aforementioned difference, communication performance, and the like.

[0034] Optionally, the model adjustment can include one or more of model resetting, model initializing, model retraining, model training, model updating, model fine-tuning, or model fine-tuning. For example, in the above scheme, taking the model adjustment as model resetting as an example, the first parameter can be the model parameter after resetting, and the second parameter can be the model parameter before resetting (for example, the model parameter used by the first model in the process of performing a certain AI task).

[0035] In a possible implementation form of the first aspect, the second information is obtained based on processing of the first model processing result by the second model, and the first model processing result is obtained based on processing of the first parameter.

[0036] Based on the above scheme, the first model and the second model can be used to jointly complete the same AI task, that is, the AI task can be completed based on the processing of the first model and the processing of the second model. The second information can be obtained based on processing of the first processing result by the second model, and the first processing result can be obtained based on processing of the first parameter before the model adjustment of the first model, so that the first model participates in the joint processing of the first model and the second model through the parameter before the model adjustment, avoiding the influence of the model adjustment of the first model on the performance of the second model.

[0037] In addition, in the case where the joint processing of the first model and the second model fails (for example, the performance of the joint processing is low), the above scheme can exclude the influence caused by the model adjustment of the first model, and through the second information, it can be indicated whether the failure is caused by the performance degradation of the second model, so as to realize accurate positioning of the failure and improve the efficiency of model management.

[0038] In a possible implementation form of the first aspect, the second information is used to indicate the performance of data processing of the second model using a third parameter, and the second information is obtained in a case where the performance of data processing of the second model using a fourth parameter is lower than a threshold value; the third parameter is a parameter after the model adjustment operation is performed on the second model, and the fourth parameter is a parameter before the model adjustment operation is performed on the second model.

[0039] Based on the above scheme, the second information can be used to indicate the performance of the second model obtained by using the third parameter for data processing, the third parameter being a parameter after performing a model adjustment operation on the second model, so that the second model participates in the joint processing of the first model and the second model through the parameter before the model adjustment, avoiding the influence of the model adjustment operation on the performance of the second model.

[0040] In addition, in the case where the joint processing of the first model and the second model fails (for example, the performance of the joint processing is low), the above scheme can exclude the influence caused by the model adjustment of the second model, and through the second information, it can be indicated whether the failure is caused by the performance degradation of the second model, so that the failure can be accurately located, and the efficiency of model management can be improved.

[0041] Optionally, the second information is obtained in a case where the performance obtained by using the fourth parameter based on the second model and the first processing result for data processing is lower than a threshold.

[0042] In a possible implementation manner of the first aspect, the method further includes: in a case where it is determined based on the first information that the performance of the first model satisfies the first condition, the first communication device sends request information for requesting the second information.

[0043] Based on the above scheme, in a case where it is determined based on the first information that the performance of the first model satisfies the first condition, the first communication device can determine that the first model is normal (or has no failure), and accordingly, in a case where the joint processing of the first model and the second model fails (for example, the performance of the joint processing is low), the first communication device can determine the cause of the failure based on the above process and determine that the failure is not caused by the first model. To this end, the first communication device can send the request information, so as to determine through the subsequently received second information whether the cause of the failure is caused by the second model.

[0044] In addition, the first communication device determines whether the first model deployed on itself has a failure through the first information, which can be understood as a process of model monitoring of the first communication device on the model of itself, and this process can be referred to as a process of model self-checking (process 1 for short); the first communication device determines whether the second model deployed on the second communication device has a failure through the second information, which can be understood as a process of model monitoring of the first communication device on the model deployed on the opposite end (process 2 for short). In the two processes, the former can generally be executed locally, and the latter generally needs to be determined based on the interaction information (for example, the second information) of the opposite end. To this end, in the above scheme, the first communication device executes process 1 first and then executes process 2, which can avoid unnecessary overhead, and improve the efficiency of model management.

[0045] It should be noted that in the implementation of the process 1 and the process 2, the process 2 can be triggered based on the process 1, that is, the first communication apparatus triggers the process 2 in the case that the first communication apparatus determines that the model deployed at the local end is normal (or fault-free) through the process 1. Accordingly, the first communication apparatus can determine the third information based on the second information (that is, the first information can not be a basis for determining the third information). In this case, the first communication apparatus can determine the third information based on the second information, and the third information is used to indicate adjusting the second model, or the third information is used to indicate adjusting the first model and the second model.

[0046] In a possible implementation of the first aspect, the first communication apparatus acquires the first information, including: in the case that the performance of the second model satisfies the second condition based on the second information, the first communication apparatus acquires the first information.

[0047] Based on the above scheme, in the case that the performance of the second model satisfies the second condition based on the second information, the first communication apparatus can determine that the second model is normal (or fault-free), and accordingly, in the case that the joint processing of the first model and the second model fails (for example, the performance of the joint processing is low), the first communication apparatus can determine the cause of the failure based on the above process is not caused by the second model. To this end, the first communication apparatus can acquire the first information to determine whether the cause of the failure is caused by the first model through the first information, and unnecessary model monitoring of the first model can be avoided to improve the efficiency of model management.

[0048] In a possible implementation of the first aspect, the third information satisfies any one of the following conditions:

[0049] In the case that the performance of the first model does not satisfy the first condition based on the first information and the performance of the second model satisfies the second condition based on the second information (denoted as case 1), the third information is used to indicate performing a model adjustment operation on the first model; or,

[0050] In the case that the performance of the first model satisfies the first condition based on the first information and the performance of the second model does not satisfy the second condition based on the second information (denoted as case 2), the third information is used to indicate performing a model adjustment operation on the second model; or,

[0051] In the case that the performance of the first model does not satisfy the first condition based on the first information and the performance of the second model does not satisfy the second condition based on the second information (denoted as case 3), the third information is used to indicate performing a model adjustment operation on the first model and the second model.

[0052] In a case (denoted as Case 4) that the performance of the first model satisfies the first condition based on the first information and the performance of the second model satisfies the second condition based on the second information, and the performance of the AI task is lower than a threshold, the third information is used to indicate a model adjustment operation on the first model and the second model.

[0053] Based on the above scheme, the third information can be implemented in the above-mentioned multiple ways, and in a case that the joint processing of the first model and the second model fails (for example, the performance of the joint processing is lower), the cause of the failure can be identified by the first information and the second information, and the cause of the failure can be troubleshooted by the third information.

[0054] For example, in a case that the performance of the AI task completed by the first model and the second model jointly is poor (for example, the performance of the first model does not satisfy the first condition, the performance of the second model does not satisfy the second condition, the performance of the output data of the AI task is lower than a threshold, the performance of the AI task is lower than a threshold, etc.), the first communication device can determine that the first model and / or the second model may fail. The specific implementation process corresponding to the three cases will be described below in conjunction with some examples.

[0055] In the above-mentioned Case 1, the first communication device can determine that the failure is caused by the model performance degradation of the first model (and the failure is very likely not caused by the model performance degradation of the second model), and for this purpose, the third information determined by the first communication device based on the first information and the second information can indicate a model adjustment operation on the first model to troubleshoot the failure.

[0056] In the above-mentioned Case 2, the first communication device can determine that the failure is caused by the model performance degradation of the second model (and the failure is very likely not caused by the model performance degradation of the first model), and for this purpose, the third information determined by the first communication device based on the first information and the second information can indicate a model adjustment operation on the second model to troubleshoot the failure.

[0057] In the above-mentioned Case 3, the first communication device can determine that the failure is caused by the model performance degradation of the first model and the second model, and for this purpose, the third information determined by the first communication device based on the first information and the second information can indicate a model adjustment operation on the first model and the second model to troubleshoot the failure.

[0058] In the case 4, the first communication device can determine that the failure is not caused by the performance degradation of the first model and the second model. As described above, in some cases, the first information can represent whether the failure is caused by the performance degradation of the first model, and the second information can represent whether the failure is caused by the performance degradation of the second model. Therefore, the first communication device can determine that the failure is likely caused by the inadaptability of the data used by the model (for example, data drift), and determine third information indicating the model adjustment operation on the first model and the second model.

[0059] In a possible implementation of the first aspect, in the case that the third information indicates the model adjustment operation on the second model, the method further includes: the first communication device sending the third information.

[0060] Based on the above scheme, the first communication device can indicate the model adjustment operation on the second model to the second communication device through the sent third information, and in the case that the joint processing of the first model and the second model fails (for example, the performance of the joint processing is low), and the cause of the failure is the second model, the second model deployed by the second communication device is adjusted through the third information to eliminate the failure.

[0061] In a possible implementation of the first aspect, in the case that the third information indicates the model adjustment operation on the first model and the second model, the method further includes: the first communication device sending fourth information indicating the model adjustment operation on the second model.

[0062] Based on the above scheme, the first communication device can indicate the model adjustment operation on the second model to the second communication device through the sent fourth information, and in the case that the joint processing of the first model and the second model fails (for example, the performance of the joint processing is low), the second model deployed by the second communication device is adjusted through the fourth information to eliminate the failure.

[0063] Optionally, in a case where the performance of the AI task completed jointly by the first model and the second model is poor (for example, the performance of the first model does not meet the first condition, the performance of the second model does not meet the second condition, the performance of the output data of the AI task is lower than a threshold, the performance of the AI task is lower than a threshold, etc.), the first communication apparatus can determine that the first model and / or the second model may have a fault. As described above, in some cases, the first information can represent whether the fault is caused by the model performance degradation of the first model, and the second information can also represent whether the fault is caused by the model performance degradation of the second model. In a case where the first information and the second information represent that the fault is not caused by the model performance degradation of the first model and the second model (i.e., case 4), the cause of the fault is not caused by the first model and the second model, but may be caused by the inadaptability of the data used by the model (for example, data drift). Therefore, the third information determined by the first communication apparatus is used to indicate a model adjustment operation on the first model and the second model, and accordingly, the first communication apparatus can perform a model adjustment operation on the second model deployed by the second communication apparatus through the fourth information to eliminate the fault.

[0064] In a possible implementation of the first aspect, the first information includes at least one of the following:

[0065] first indication information used to indicate the output data of the first model;

[0066] second indication information used to indicate the difference between the output data of the first model and the expected output;

[0067] third indication information used to indicate the difference between the second processing result and the third processing result; the second processing result indicates a processing result obtained by processing the output data of the second model by the first model, and the third processing result is used to indicate a processing result obtained by processing the second processing result by the first model; wherein the processing of the first model and the processing of the second model are partially or entirely the same (for example, refer to the processing of H' and H" below);

[0068] fourth indication information used to indicate whether the performance of the first model meets the first condition.

[0069] Based on the above scheme, the first information used to determine whether the performance of the first model deployed in the first communication apparatus meets the first condition can be implemented in the above-mentioned multiple ways, so as to improve the flexibility of the implementation of the scheme.

[0070] In a possible implementation of the first aspect, the second information includes at least one of the following:

[0071] fifth indication information used to indicate the output data of the second model;

[0072] sixth indication information, used for indicating a difference between the output data of the second model and an expected output;

[0073] seventh indication information, used for indicating a difference between a third processing result and a fourth processing result; the third processing result is a processing result obtained by processing the output data of the first model by using the second model, and the fourth processing result is a processing result obtained by processing the third processing result by using the second model;

[0074] eighth indication information, used for indicating whether the performance of the second model meets the first condition.

[0075] Based on the above scheme, the second information used for determining whether the performance of the second model deployed in the second communication device meets the second condition can be implemented in the above-mentioned multiple ways, so as to improve the flexibility of the scheme implementation.

[0076] The second aspect of the present application provides a communication method, which is performed by a second communication device. The second communication device can be a communication device (such as a terminal device or a network device), or the second communication device can be a part of the communication device (for example, a circuit or a chip responsible for communication function (such as a Modem chip, also known as a baseband chip, or a SoC chip or a SIP chip containing a modem core, etc.), or the second communication device can also be a logic module or software capable of realizing all or part of the communication device function. In the method, the second communication device determines second information, the second information is used for determining whether the performance of a second model deployed in the second communication device meets a second condition; wherein the second model and a first model deployed in a first communication device are used to jointly complete an AI task, the second information is used for determining third information, the third information is used for indicating adjusting the first model and / or adjusting the second model; and the second communication device sends the second information.

[0077] Based on the above scheme, the second information sent by the second communication device to the first communication device can be used to determine whether the performance of the second model deployed in the second communication device meets the second condition. Thereafter, the first communication device can determine the third information used for adjusting the first model and / or adjusting the second model based on the second information. Wherein the second model and the first model are used to jointly complete an AI task. In this way, in the case that different models deployed in different communication devices are used to jointly complete the same AI task, the first communication device can determine whether to adjust the different model based on the condition that the performance of the second model meets (or does not meet) the condition, so as to realize the model management of the model.

[0078] In addition, in the above scheme, the first communication device can determine, based on the first information and the second information, third information for adjusting the first model deployed in the first communication device and / or adjusting the second model deployed in the second communication device. To this end, the above scheme can be applied to a scenario where a plurality of nodes distributedly process the same AI task, the plurality of nodes at least including the first communication device and the second communication device, and the performance of different models can be used to accurately determine the node that fails, avoid affecting the entire distributed system, and avoid redundant operations caused by adjusting the non-faulty device, thereby improving the efficiency of model management.

[0079] In a possible implementation of the second aspect, the second information and the first information are used to determine the third information, and the first information is used to determine whether the performance of the first model deployed in the first communication device meets a first condition.

[0080] Based on the above scheme, the determination of the third information is based on the first information in addition to the second information, and the first information is used to determine whether the performance of the first model deployed in the first communication device meets a first condition. In this way, in the case where different models deployed in different communication devices are used to jointly complete the same AI task, the first communication device can determine whether to adjust the different models based on whether the performance of the different models meets a condition, thereby achieving model management of the models.

[0081] In addition, in the above scheme, the first communication device can determine, based on the first information and the second information, third information for adjusting the first model and / or adjusting the second model. To this end, the above scheme can be applied to a scenario where a plurality of nodes distributedly process the same AI task, the plurality of nodes at least including the first communication device and the second communication device, and the performance of different models can be used to accurately determine the node that fails, avoid affecting the entire distributed system, and avoid redundant operations caused by adjusting the non-faulty device, thereby improving the efficiency of model management.

[0082] In a possible implementation of the second aspect, the first information is used to indicate the performance of the first model using a first parameter for data processing; the first information is obtained in a case where the performance of the first model using a second parameter for data processing is lower than a threshold; the first parameter is a parameter corresponding to the first model after a model adjustment operation is performed on the first model, and the second parameter is a parameter corresponding to the first model before the model adjustment operation is performed on the first model.

[0083] Based on the above scheme, in the case that the performance of the processing process of the second parameter corresponding to the model adjustment operation (i.e., the second parameter) performed (or completed) by the first model deployed by the first communication device is lower than a threshold value, the first communication device can be triggered to obtain the first information, and the first information is used to indicate the performance obtained by the processing of the parameter corresponding to the model adjustment operation completed by the first model. In this way, the first communication device can trigger the performance detection of the parameter after the model adjustment based on the processing of the parameter before the model adjustment, and determine the first information based on the processing of the parameter after the model adjustment.

[0084] In a possible implementation of the second aspect, the second information is obtained based on processing of the first processing result by the second model, and the first processing result is obtained by processing of the first model by the second parameter.

[0085] Based on the above scheme, the first model and the second model can be used to jointly complete the same AI task, i.e., the AI task can be completed based on the processing of the first model and the processing of the second model. The second information can be obtained based on the processing of the first processing result by the second model, and the first processing result can be obtained by the processing of the first model by the parameter before the model adjustment, so that the first model participates in the joint processing of the first model and the second model by the parameter before the model adjustment, avoiding the influence of the model adjustment operation on the performance of the second model.

[0086] In addition, in the case that the joint processing of the first model and the second model fails (for example, the performance of the joint processing is lower), the above scheme can exclude the influence of the model adjustment of the first model, and the second information can represent whether the failure is caused by the performance degradation of the second model, so that the failure can be accurately located to improve the efficiency of model management.

[0087] In a possible implementation of the second aspect, the second information is used to indicate the performance obtained by processing of the third parameter by the second model, and the second information is obtained in the case that the performance obtained by data processing of the second model using the fourth parameter is lower than a threshold value; the third parameter is a parameter of the second model after the model adjustment operation, and the fourth parameter is a parameter of the second model before the model adjustment operation.

[0088] Based on the above scheme, the second information can be used to indicate the performance obtained by data processing of the third parameter by the second model, and the third parameter is a parameter of the second model after the model adjustment operation, so that the second model participates in the joint processing of the first model and the second model by the parameter before the model adjustment, avoiding the influence of the model adjustment operation on the performance of the second model.

[0089] In addition, in a case where the joint processing of the first model and the second model fails (for example, the joint processing has a low performance), the above scheme can exclude the influence caused by the model adjustment of the second model, can represent whether the failure is caused by the performance degradation of the second model based on the second information, can accurately locate the failure, and can improve the efficiency of model management.

[0090] Optionally, the second information is obtained in a case where a performance obtained by performing data processing based on the second model, the fourth parameter, and the first processing result is lower than a threshold.

[0091] In a possible implementation of the second aspect, the first information is obtained in a case where it is determined based on the second information that the performance of the second model satisfies the second condition.

[0092] Based on the above scheme, in a case where it is determined based on the second information that the performance of the second model satisfies the second condition, the first communication device can determine that the second model is normal (or does not fail), and accordingly, in a case where the joint processing of the first model and the second model fails (for example, the joint processing has a low performance), the first communication device can determine, based on the above process, that the cause of the failure is not caused by the second model. To this end, the first communication device can obtain the first information to determine, based on the first information, whether the cause of the failure is caused by the first model, and can avoid unnecessary model monitoring of the first model, thereby improving the efficiency of model management.

[0093] In a possible implementation of the second aspect, the method further includes: the second communication device receives request information used to request the second information.

[0094] Based on the above scheme, in a case where it is determined based on the first information that the performance of the first model satisfies the first condition, the first communication device can determine that the first model is normal (or does not fail), and accordingly, in a case where the joint processing of the first model and the second model fails (for example, the joint processing has a low performance), the first communication device can determine, based on the above process, that the cause of the failure is not caused by the first model. To this end, the first communication device can send the above request information to determine, based on subsequently received second information, whether the cause of the failure is caused by the second model.

[0095] Further, the first communication device determines whether the first model deployed on itself is faulty through the first information, which can be understood as a process of model monitoring of the first communication device on the model of itself, and this process can be referred to as a process of model self-checking (process 1 for short); and the first communication device determines whether the second model deployed on the second communication device is faulty through the second information, which can be understood as a process of model monitoring of the first communication device on the model deployed on the opposite end (process 2 for short). In the two processes, the former can generally be executed locally, and the latter generally needs to determine the information (for example, the second information) of the interaction of the opposite end, and for this purpose, in the above scheme, the manner of executing process 1 first and then executing process 2 can avoid unnecessary overhead, so as to improve the efficiency of model management.

[0096] In a possible implementation manner of the second aspect, the third information satisfies any one of the following conditions:

[0097] In a case where it is determined based on the first information that the performance of the first model does not satisfy the first condition and it is determined based on the second information that the performance of the second model satisfies the second condition, the third information is used to instruct a model adjustment operation on the first model; or,

[0098] In a case where it is determined based on the first information that the performance of the first model satisfies the first condition and it is determined based on the second information that the performance of the second model does not satisfy the second condition, the third information is used to instruct a model adjustment operation on the second model; or,

[0099] In a case where it is determined based on the first information that the performance of the first model does not satisfy the first condition and it is determined based on the second information that the performance of the second model does not satisfy the second condition, the third information is used to instruct a model adjustment operation on the first model and the second model.

[0100] In a case where it is determined based on the first information that the performance of the first model satisfies the first condition and it is determined based on the second information that the performance of the second model satisfies the second condition, and the performance of the AI task is lower than a threshold, the third information is used to instruct a model adjustment operation on the first model and the second model.

[0101] Based on the above scheme, the third information can be implemented in the above-mentioned multiple manners, and in a case where the joint processing of the first model and the second model fails (for example, the performance of the joint processing is lower), the cause of the failure can be identified through the first information and the second information, and the cause of the failure can be addressed through the third information.

[0102] In a possible implementation manner of the second aspect, in a case where the third information is used to instruct a model adjustment operation on the second model, the method further includes: receiving, by the second communication device, the third information.

[0103] Based on the above scheme, the first communication device can indicate to the second communication device, through the third information sent, that the second model is subjected to the model adjustment operation, and in the case that the joint processing of the first model and the second model fails (for example, the performance of the joint processing is low), and the cause of the failure is the second model, the second model deployed by the second communication device is subjected to the model adjustment operation through the third information to eliminate the failure.

[0104] In a possible implementation of the second aspect, in the case that the third information is used to indicate that the first model and the second model are subjected to the model adjustment operation, the method further includes: the second communication device receiving fourth information, the fourth information being used to indicate that the second model is subjected to the model adjustment operation.

[0105] Based on the above scheme, the first communication device can indicate to the second communication device, through the fourth information sent, that the second model is subjected to the model adjustment operation, and in the case that the joint processing of the first model and the second model fails (for example, the performance of the joint processing is low), the second model deployed by the second communication device is subjected to the model adjustment operation through the fourth information to eliminate the failure.

[0106] Optionally, in the case that the performance of the AI task completed by the first model and the second model jointly is poor (for example, the performance of the first model does not meet the first condition, the performance of the second model does not meet the second condition, the performance of the output data of the AI task is lower than a threshold, the performance of the AI task is lower than a threshold, etc.), the first communication device can determine that the first model and / or the second model is likely to fail. As described above, in some cases, the first information can represent whether the failure is caused by the model performance degradation of the first model, and the second information can also represent whether the failure is caused by the model performance degradation of the second model. In the case that the first information and the second information represent that the failure is not caused by the model performance degradation of the first model and the second model (i.e., case 4), the cause of the failure is not caused by the first model and the second model, but is likely caused by the inadaptability of the data used by the model (for example, data drift), and therefore, the third information determined by the first communication device is used to indicate that the first model and the second model are subjected to the model adjustment operation, and accordingly, the first communication device can subject the second model deployed by the second communication device to the model adjustment operation through the fourth information to eliminate the failure.

[0107] In a possible implementation of the second aspect, the first information includes at least one of:

[0108] first indication information, used to indicate the output data of the first model;

[0109] second indication information, used for indicating a difference between the output data of the first model and an expected output;

[0110] third indication information, used for indicating a difference between a second processing result and a third processing result; the second processing result is used for indicating a processing result obtained by processing, by the first model, the output data of the second model, and the third processing result is used for indicating a processing result obtained by processing, by the first model, the second processing result; wherein the processing of the first model and the processing of the second model are partially same or totally same;

[0111] fourth indication information, used for indicating whether the performance of the first model meets the first condition.

[0112] Based on the above scheme, the first information used for determining whether the performance of the first model deployed in the first communication device meets the first condition can be implemented in the above-mentioned multiple ways, so as to improve the flexibility of the implementation of the scheme.

[0113] In a possible implementation manner of the second aspect, the second information includes at least one of the following:

[0114] fifth indication information, used for indicating the output data of the second model;

[0115] sixth indication information, used for indicating a difference between the output data of the second model and an expected output;

[0116] seventh indication information, used for indicating a difference between a third processing result and a fourth processing result; the third processing result is used for indicating a processing result obtained by processing, by the second model, the output data of the first model, and the fourth processing result is used for indicating a processing result obtained by processing, by the second model, the third processing result;

[0117] eighth indication information, used for indicating whether the performance of the second model meets the first condition.

[0118] Based on the above scheme, the second information used for determining whether the performance of the second model deployed in the second communication device meets the second condition can be implemented in the above-mentioned multiple ways, so as to improve the flexibility of the implementation of the scheme.

[0119] The third aspect of the present application provides a communication device, which is a first communication device, comprising a transceiver unit and a processing unit; the processing unit is configured to obtain first information, the first information being used to determine whether a performance of a first model deployed in the first communication device meets a first condition; the transceiver unit is configured to receive second information, the second information being used to determine whether a performance of a second model deployed in a second communication device meets a second condition; wherein the second model and the first model are used to jointly complete an AI task; the processing unit is further configured to determine third information based on the first information and the second information, the third information being used to indicate adjustment of the first model and / or adjustment of the second model.

[0120] In the third aspect of the present application, the constituent modules of the communication device can also be configured to perform the steps performed in the various possible implementation manners of the first aspect and achieve the corresponding technical effects, which can be referred to the first aspect for details and will not be described here.

[0121] The fourth aspect of the present application provides a communication device, which is a second communication device, comprising a transceiver unit and a processing unit; the processing unit is configured to determine second information, the second information being used to determine whether a performance of a second model deployed in the second communication device meets a second condition; wherein the second model and a first model deployed in a first communication device are used to jointly complete an AI task, the second information is used to determine third information, the third information being used to indicate adjustment of the first model and / or adjustment of the second model; the transceiver unit is configured to send the second information.

[0122] In the fourth aspect of the present application, the constituent modules of the communication device can also be configured to perform the steps performed in the various possible implementation manners of the second aspect and achieve the corresponding technical effects, which can be referred to the second aspect for details and will not be described here.

[0123] The fifth aspect of the present application provides a communication device, comprising at least one processor coupled with a memory; the memory is configured to store programs or instructions; the at least one processor is configured to execute the programs or instructions to enable the device to implement the method in any one of the possible implementation manners of any one of the preceding first aspect to second aspect. Optionally, the communication device can comprise the memory.

[0124] The sixth aspect of the present application provides a communication device, comprising at least one logic circuit and an input-output interface; the logic circuit is configured to execute the method in any one of the possible implementation manners of any one of the preceding first aspect to second aspect.

[0125] The seventh aspect of the present application provides a communication system, comprising the first communication device and the second communication device.

[0126] The eighth aspect of the present application provides a computer readable storage medium for storing one or more computer-executable instructions, which, when executed by a processor, cause the processor to perform the method of any possible implementation of any one of the first aspect to the second aspect.

[0127] The ninth aspect of the present application provides a computer program product (or computer program), which, when executed by a processor, causes the processor to perform the method of any possible implementation of any one of the first aspect to the second aspect.

[0128] The tenth aspect of the present application provides a chip or chip system, which includes at least one processor for supporting a communication apparatus to perform the method of any possible implementation of any one of the first aspect to the second aspect. For example, the chip can be a baseband chip, a modem chip, an SoC chip (such as an SoC chip including a modem core), a SIP chip, or a communication module, etc.

[0129] In a possible design, the chip or chip system can further include a memory for storing necessary program instructions and data of the communication apparatus. The chip system can be composed of a chip, or can include a chip and other discrete devices. Optionally, the chip system further includes an interface circuit for providing program instructions and / or data for the at least one processor.

[0130] The technical effects brought by any one of the third aspect to the tenth aspect can be referred to the technical effects brought by different design manners of the first aspect to the second aspect, which will not be repeated here. BRIEF DESCRIPTION OF DRAWINGS

[0131] FIGS. 1a to 1c are schematic diagrams of a communication system provided by the present application;

[0132] FIGS. 2a to 2g are schematic diagrams of an AI processing procedure related to the present application;

[0133] FIG. 3 is an interaction schematic diagram of a communication method provided by the present application;

[0134] FIGS. 4a to 4c are some schematic diagrams of a model provided by the present application;

[0135] FIGS. 4d to 4g are some interaction schematic diagrams of a communication method provided by the present application;

[0136] FIGS. 5 to 9 are schematic diagrams of a communication apparatus provided by the present application. DETAILED DESCRIPTION

[0137] First, some terms in the embodiments of the present application are explained and described, so as to facilitate the understanding of those skilled in the art.

[0138] (1) Terminal device: can be a wireless terminal device capable of receiving network device scheduling and indication information, the wireless terminal device can be a device that provides voice and / or data connectivity to a user, or a handheld device with wireless connection function, or other processing devices connected to a wireless modem.

[0139] The terminal device can communicate with one or more core networks or the Internet through a radio access network (RAN), and the terminal device can be a mobile terminal device, such as a mobile phone (or called "cellular" phone, mobile phone), computer and data card, for example, it can be a portable, pocket-sized, handheld, computer built-in or vehicle-mounted mobile device, which exchanges voice and / or data with the radio access network. For example, personal communication service (PCS) phone, cordless phone, session initiation protocol (SIP) phone, wireless local loop (WLL) station, personal digital assistant (PDA), tablet or pad, computer with wireless transceiver function and the like. The wireless terminal device can also be called system, subscriber unit, subscriber station, mobile station (MS), remote station, access point (AP), remote terminal, access terminal, user terminal, user agent, subscriber station (SS), customer premises equipment (CPE), terminal, user equipment (UE), mobile terminal (MT) and the like.

[0140] By way of example and not limitation, in embodiments of the present application, the terminal device can also be a wearable device. The wearable device can also be referred to as a smart wearable device or a smart wearable device, etc., which is a general term for devices that apply wearable technology to the intelligent design and development of daily wear, such as glasses, gloves, watches, clothing, and shoes, etc. The wearable device is a portable device that can be directly worn on the body or integrated into the user's clothes or accessories. The wearable device is not just a hardware device, but also has powerful functions through software support and data interaction, cloud interaction. The broad sense of wearable smart devices includes devices with full functions, large sizes, and the ability to realize complete or partial functions without relying on smart phones, such as smart watches or smart glasses, etc., and devices that focus only on a certain application function and need to be used with other devices such as smart phones, such as various smart wristbands, smart helmets, smart jewelry, etc.

[0141] The terminal can also be a drone, a robot, a terminal in device-to-device (D2D) communication, a terminal in vehicle to everything (V2X), a virtual reality (VR) terminal device, an augmented reality (AR) terminal device, a wireless terminal in industrial control, a wireless terminal in self driving, a wireless terminal in telemedicine or telehealth services, a wireless terminal in smart grid, a wireless terminal in transportation safety, a wireless terminal in smart city, a wireless terminal in smart home, etc.

[0142] In addition, the terminal device can also be a terminal device in a future communication system (such as a 5G Advanced communication system, etc.) after the 5th generation (5G) communication system or a terminal device in a future evolved public land mobile network (PLMN), etc. For example, the 5G Advanced network can further expand the form and function of the 5G communication terminal, and the 5G Advanced terminal includes but is not limited to vehicles, cellular network terminals (with satellite terminal functions), drones, and internet of things (IoT) devices.

[0143] In the embodiments of the present application, the terminal device can also obtain an artificial intelligence (AI) service provided by the network device. Optionally, the terminal device can also have AI processing capability.

[0144] (2) Network device: can be a device in a wireless network, for example, the network device can be a RAN node (or device) for accessing the terminal device to the wireless network, which can also be referred to as a base station. At present, some examples of RAN devices are: base station, evolved NodeB (eNodeB), base station gNB (gNodeB) in 5G communication system, transmission reception point (TRP), evolved Node B (eNB), radio network controller (RNC), Node B (NB), home base station (for example, home evolved Node B, or home Node B, HNB), base band unit (BBU) or wireless fidelity (Wi-Fi) access point (AP) and the like. In addition, in one network structure, the network device can include a central unit (CU) node, or a distributed unit (DU) node, or a RAN device including a CU node and a DU node.

[0145] Optionally, the RAN node can also be a macro base station, a micro base station or an indoor station, a relay node or a donor node, or a wireless controller in a cloud radio access network (CRAN) scenario. The RAN node can also be a server, a wearable device, a vehicle or a vehicle-mounted device, etc. For example, the access network device in vehicle external connection (V2X) technology can be a road side unit (RSU).

[0146] In another possible scenario, a terminal is assisted by multiple RAN nodes to implement wireless access, and different RAN nodes respectively implement part of functions of a base station. For example, a RAN node can be a CU, a DU, a CU-control plane (CP), a CU-user plane (UP), or a radio unit (RU), etc. The CU and the DU can be separately configured, or can be included in the same network element, for example, a baseband unit (BBU). The RU can be included in a radio frequency device or a radio frequency unit, for example, included in a remote radio unit (RRU), an active antenna unit (AAU), a radio head (RH), or a remote radio head (RRH).

[0147] In different systems, the CU (or CU-CP and CU-UP), the DU, or the RU can also have different names, but those skilled in the art can understand their meanings. For example, in an open RAN (O-RAN or ORAN) system, the CU can also be referred to as an O-CU (open CU), the DU can also be referred to as an O-DU, the CU-CP can also be referred to as an O-CU-CP, the CU-UP can also be referred to as an O-CU-UP, and the RU can also be referred to as an O-RU. For the convenience of description, the CU, the CU-CP, the CU-UP, the DU, and the RU are taken as examples for description in this application. Any one of the CU (or the CU-CP, the CU-UP), the DU, and the RU in this application can be implemented by a software module, a hardware module, or a combination of a software module and a hardware module.

[0148] The communication between the access network device and the terminal device complies with a certain protocol layer structure. The protocol layer can include a control plane protocol layer and a user plane protocol layer. The control plane protocol layer can include at least one of the following: a radio resource control (RRC) layer, a packet data convergence protocol (PDCP) layer, a radio link control (RLC) layer, a media access control (MAC) layer, or a physical (PHY) layer, etc. The user plane protocol layer can include at least one of the following: a service data adaptation protocol (SDAP) layer, a PDCP layer, an RLC layer, a MAC layer, or a physical layer, etc.

[0149] For the correspondence between the network elements in the ORAN system and the protocol layer functions that can be implemented by the network elements, refer to Table 1 below.

[0150] Table 1

[0151] The network device can be another device that provides a wireless communication function for the terminal device. Embodiments of the present application do not limit the specific technology and specific device form adopted by the network device. For the convenience of description, embodiments of the present application do not limit.

[0152] The network device can also include a core network device, for example, a mobility management entity (MME) in a fourth generation (4G) network, a home subscriber server (HSS), a serving gateway (S-GW), a policy and charging rules function (PCRF), a public data network gateway (PDN gateway or P-GW), a network element such as an access and mobility management function (AMF) in a 5G network, a user plane function (UPF), or a session management function (SMF). In addition, the core network device can also include other core network devices in the 5G network and the next generation network of the 5G network.

[0153] In the embodiments of the present application, the network device mentioned above can also be an AI-capable network node, which can provide AI services for terminals or other network devices, for example, AI nodes, computing power nodes, AI-capable RAN nodes, AI-capable core network elements, etc. on the network side (access network or core network).

[0154] In the embodiments of the present application, the device for implementing the function of the network device can be a network device or a device capable of supporting the network device to implement the function, such as a chip system, which can be arranged in the network device. In the technical solutions provided in the embodiments of the present application, the device for implementing the function of the network device is taken as an example to describe the technical solutions provided in the embodiments of the present application.

[0155] (3) Configuration and pre-configuration: in this application, both configuration and pre-configuration will be used. Among them, configuration refers to that the network device / server sends some parameter configuration information or parameter values to the terminal through messages or signaling, so that the terminal determines the communication parameters or resource in transmission according to the values or information. Pre-configuration is similar to configuration, which can be parameter information or parameter values agreed by the network device / server and the terminal device in advance, or parameter information or parameter values adopted by the base station / network device or the terminal device according to the standard protocol, or parameter information or parameter values pre-stored in the base station / server or the terminal device. This application does not limit it.

[0156] Further, these values and parameters can be changed or updated.

[0157] (4) The terms "system" and "network" in the embodiments of the present application can be used interchangeably. "Multiple" means two or more. "And / or" describes the association between the associated objects, which means that there can be three relationships, for example, A and / or B, which means that A exists alone, A and B exist together, and B exists alone, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects. "At least one of the following" or similar expressions means any combination of these items, including any combination of single item or multiple items. For example, "at least one of A, B and C" includes A, B, C, AB, AC, BC or ABC. In addition, unless otherwise specified, the ordinal numbers "first", "second", etc. mentioned in the embodiments of the present application are used to distinguish multiple objects, and are not used to limit the order, time sequence, priority or importance of the multiple objects.

[0158] (5) In the embodiments of the present application, "sending" and "receiving" represent the direction of signal transmission. For example, "sending information to XX" can be understood as that the destination of the information is XX, which can include direct sending through the air interface, or indirect sending through the air interface by other units or modules. "Receiving information from YY" can be understood as that the source of the information is YY, which can include direct receiving from YY through the air interface, or indirect receiving from YY through the air interface by other units or modules. "Sending" can also be understood as the "output" of the chip interface, and "receiving" can also be understood as the "input" of the chip interface.

[0159] In other words, sending and receiving can be carried out between devices, such as between network devices and terminal devices, or within devices, such as between components, modules, chips, software modules or hardware modules within a device through buses, wires or interfaces.

[0160] It can be understood that the information can be processed, such as encoding and modulation, between the source end and the destination end of the information transmission, but the destination end can understand the effective information from the source end. Similar expressions in this application can be similarly understood, and will not be repeated here.

[0161] (6) In the embodiments of the present application, “indication” can include direct indication and indirect indication, and can also include explicit indication and implicit indication. The information indicated by certain information (indication information described below) is referred to as to-be-indicated information. In the implementation process, there are many ways to indicate the to-be-indicated information, for example, but not limited to, the to-be-indicated information can be directly indicated, such as the to-be-indicated information itself or the index of the to-be-indicated information. The to-be-indicated information can also be indirectly indicated by indicating other information, where the other information and the to-be-indicated information have an association relationship. The to-be-indicated information can also be indicated only by a part, and the other part of the to-be-indicated information is known or agreed in advance. For example, the indication of a specific information can be achieved by means of the arrangement order of each information agreed in advance (for example, protocol predefined), thereby reducing the indication overhead to a certain extent. The present application does not limit the specific manner of indication. It can be understood that for the sender of the indication information, the indication information can be used to indicate the to-be-indicated information, and for the receiver of the indication information, the indication information can be used to determine the to-be-indicated information.

[0162] In the present application, the same or similar parts of each embodiment can be mutually referred to, unless otherwise specified. In the various embodiments of the present application, and the various methods / designs / implementation manners in each embodiment, the terms and / or descriptions of different embodiments, and the various methods / designs / implementation manners in each embodiment are consistent and can be mutually referred to, unless otherwise specified and logically conflicted. The technical features of different embodiments, and the various methods / designs / implementation manners in each embodiment can be combined to form new embodiments, methods, or implementation manners according to their inherent logical relationship. The implementation manners of the present application described below do not constitute a limitation on the protection scope of the present application.

[0163] The present application can be applied to a long term evolution (LTE) system, a new radio (NR) system, or a future communication system after 5G. The communication system includes at least one network device and / or at least one terminal device.

[0164] Please refer to FIG. 1a, which is a schematic diagram of a communication system in the present application. In FIG. 1a, a network device and six terminal devices are exemplarily shown, which are terminal device 1, terminal device 2, terminal device 3, terminal device 4, terminal device 5 and terminal device 6. In the example shown in FIG. 1a, the terminal device 1 is exemplarily taken as a smart tea cup, the terminal device 2 is exemplarily taken as a smart air conditioner, the terminal device 3 is exemplarily taken as a smart gas station, the terminal device 4 is exemplarily taken as a vehicle, the terminal device 5 is exemplarily taken as a mobile phone, and the terminal device 6 is exemplarily taken as a printer.

[0165] As shown in FIG. 1a, the sending entity of the AI configuration information can be the network device. The receiving entity of the AI configuration information can be the terminal devices 1-6. In this case, the network device and the terminal devices 1-6 form a communication system, in which the terminal devices 1-6 can send data to the network device, and the network device receives the data sent by the terminal devices 1-6. The network device can send configuration information to the terminal devices 1-6.

[0166] Exemplarily, in FIG. 1a, the terminal devices 4-6 can also form a communication system. Among them, the terminal device 5 acts as a network device, i.e., the sending entity of the AI configuration information; the terminal devices 4 and 6 act as terminal devices, i.e., the receiving entity of the AI configuration information. For example, in a vehicle-to-everything system, the terminal device 5 sends AI configuration information to the terminal devices 4 and 6, and receives data sent by the terminal devices 4 and 6; correspondingly, the terminal devices 4 and 6 receive the AI configuration information sent by the terminal device 5, and send data to the terminal device 5.

[0167] Taking the communication system shown in FIG. 1a as an example, in addition to performing communication-related services, different devices (including network devices, network devices and terminal devices, and / or terminal devices and terminal devices) can also perform AI-related services.

[0168] As shown in FIG. 1b, taking a network device as a base station as an example, the base station can perform communication-related services and AI-related services with one or more terminal devices, and different terminal devices can also perform communication-related services and AI-related services.

[0169] As shown in FIG. 1c, taking a terminal device including a television and a mobile phone as an example, the television and the mobile phone can also perform communication-related services and AI-related services.

[0170] The technical solutions provided in the present application can be applied to a wireless communication system (for example, the system shown in FIG. 1a, FIG. 1b or FIG. 1c), and for example, an AI network element can be introduced in the communication system provided in the present application to implement part or all of the AI-related operations. The AI network element can also be referred to as an AI node, an AI device, an AI entity, an AI module, an AI model, or an AI unit, etc. The AI network element can be built-in in a network element of the communication system. For example, the AI network element can be an AI module built-in in an access network device, a core network device, a cloud server, or an operation, administration and maintenance (OAM), to implement AI-related functions. The OAM can serve as a network management of the core network device and / or a network management of the access network device. Alternatively, the AI network element can also be a network element independently arranged in the communication system. Optionally, an AI entity can also be included in a terminal or a chip built-in in the terminal, to implement AI-related functions.

[0171] Optionally, in the communication system, the AI application use cases can include but are not limited to: CSI feedback enhancement, or end-to-end transceiver, etc. The following will be described taking the CSI feedback enhancement as an example.

[0172] The CSI is the channel property of the communication link, and is the channel quality information reported by the terminal device to the network device. The terminal device reports the channel quality information to the network device, so as to select a suitable modulation and coding scheme (MCS) for the terminal device, so that the wireless channel can be adapted to the change. For example, the terminal device performs channel estimation according to the received channel state information-reference signal (CSI-RS), and then feeds back the channel quality information to the network device. The information is used as the input of the model of the network device, so that the network device can implement AI model training. By applying AI to the CSI feedback enhancement, the overhead can be reduced, the accuracy can be improved, and the prediction can be realized, etc.

[0173] The CSI-RS feedback enhancement can include at least one sub-function, such as: CSI compression, CSI prediction, and CSI-RS configuration signaling reduction. The CSI compression can be further divided into CSI compression in at least one of the spatial domain, the time domain and the frequency domain.

[0174] It should be understood that the definition of each of the technical terms above is only an example. For example, as technology continues to evolve, the scope of the above definitions can also change, and the embodiments of the present application are not limited.

[0175] For example, an AI function can include multiple AI sub-functions.

[0176] Optionally, the AI application case is also referred to as an AI application scenario or an AI function.

[0177] From the above description of the AI application case, it can be seen that AI can be widely used in CSI feedback enhancement, beam management, positioning accuracy enhancement, energy saving, mobility enhancement, and load balancing to improve network performance. AI models can usually be deployed on the network side and / or the terminal device side, and the training of AI models depends on the collection of training data, which can come from the measurement and feedback of terminal devices.

[0178] The concepts that can be involved in the present application will be briefly introduced below.

[0179] AI can give machines human-like intelligence, for example, machines can use computer hardware and software to simulate some intelligent behavior of humans. To achieve artificial intelligence, machine learning methods can be used. In machine learning methods, machines learn (or train) models using training data. The model represents the mapping between input and output. The learned model can be used for inference (or prediction), i.e., the model can be used to predict the output corresponding to a given input. The output can also be referred to as an inference result (or prediction result).

[0180] Machine learning can include supervised learning, unsupervised learning, and reinforcement learning. Unsupervised learning can also be referred to as non-supervised learning.

[0181] Supervised learning uses collected sample values and sample labels to learn the mapping relationship from sample values to sample labels using machine learning algorithms, and uses an AI model to express the learned mapping relationship. The process of training a machine learning model is the process of learning this mapping relationship. In the training process, the sample value is input into the model to obtain the predicted value of the model, and the error between the predicted value of the model and the sample label (ideal value) is calculated to optimize the model parameters. After the mapping relationship is learned, the learned mapping can be used to predict new sample labels. The learned mapping relationship can include linear mapping or nonlinear mapping. According to the type of label, the learned task can be divided into classification tasks and regression tasks.

[0182] Unsupervised learning relies on collected sample values ​​to discover inherent patterns within the samples themselves. One type of unsupervised learning algorithm uses the samples themselves as supervisory signals, meaning the model learns the mapping relationship from sample to sample; this is called self-supervised learning. During training, model parameters are optimized by calculating the error between the model's predictions and the samples themselves. Self-supervised learning can be used for signal compression and decompression recovery applications; common algorithms include autoencoders and generative adversarial networks.

[0183] Reinforcement learning, unlike supervised learning, is a type of algorithm that learns problem-solving strategies through interaction with the environment. Unlike supervised and unsupervised learning, reinforcement learning problems do not have explicit "correct" action labels. The algorithm needs to interact with the environment to obtain reward signals from the environment, and then adjust its decision actions to obtain a larger reward signal value. For example, in downlink power control, the reinforcement learning model adjusts the downlink transmission power of each user based on the total system throughput feedback from the wireless network, aiming to achieve a higher system throughput. The goal of reinforcement learning is also to learn the mapping relationship between the environment state and a better (e.g., optimal) decision action. However, because the label of the "correct action" cannot be obtained in advance, the network cannot be optimized by calculating the error between the action and the "correct action." Reinforcement learning training is achieved through iterative interaction with the environment.

[0184] Neural networks (NNs) are a specific model in machine learning techniques. According to the general approximation theorem, neural networks can theoretically approximate any continuous function, thus enabling them to learn arbitrary mappings. Traditional communication systems rely on extensive expert knowledge to design communication modules, while deep learning communication systems based on neural networks can automatically discover hidden pattern structures from large datasets, establish mapping relationships between data, and achieve performance superior to traditional modeling methods.

[0185] The idea behind neural networks comes from the neuronal structure of the brain. For example, each neuron performs a weighted summation of its input values ​​and outputs the result through an activation function.

[0186] Figure 2a shows a schematic diagram of a neuron structure. Assume the input to the neuron is x = [x0, x1, ..., x...]. n The weights corresponding to each input are w = [w0, w1, ..., w] n ], where n is a positive integer, w i and x i It can be any possible type, such as a decimal, an integer (e.g., 0, a positive integer, or a negative integer), or a complex number. i As x i The weights are used to assign weights to x. iThe inputs are weighted. The bias for the weighted sum of the inputs according to the weights is, for example, b. The form of the activation function can be various. Assuming that the activation function of a neuron is y = f(z) = max(0, z), the output of the neuron is: For example, the activation function of a neuron is y = f(z) = z, the output of the neuron is: where b can be various possible types such as a decimal number, an integer (for example, 0, a positive integer, or a negative integer), or a complex number. The activation functions of different neurons in a neural network can be the same or different.

[0187] In addition, a neural network generally includes multiple layers, and each layer can include one or more neurons. By increasing the depth and / or width of the neural network, the expressive ability of the neural network can be improved, and the neural network can provide stronger information extraction and abstract modeling capabilities for complex systems. The depth of the neural network can refer to the number of layers included in the neural network, and the number of neurons included in each layer can be referred to as the width of the layer. In an implementation manner, the neural network includes an input layer and an output layer. The input layer of the neural network processes the received input information through neurons, and transmits the processing result to the output layer, and the output layer obtains the output result of the neural network. In another implementation manner, the neural network includes an input layer, a hidden layer, and an output layer. The input layer of the neural network processes the received input information through neurons, and transmits the processing result to the intermediate hidden layer. The hidden layer calculates the received processing result to obtain a calculation result, and transmits the calculation result to the output layer or the next adjacent hidden layer, and finally the output layer obtains the output result of the neural network. The neural network can include one hidden layer, or include multiple sequentially connected hidden layers, which is not limited.

[0188] The neural network is, for example, a deep neural network (DNN). According to the construction manner of the network, the DNN can include a feedforward neural network (FNN), a convolutional neural network (CNN), and a recurrent neural network (RNN).

[0189] FIG. 2b is a schematic diagram of a FNN network. The characteristic of the FNN network is that the neurons of adjacent layers are completely connected two by two. This characteristic makes the FNN usually need a large amount of storage space, and leads to a high calculation complexity.

[0190] CNN is a kind of neural network specially designed to deal with data with similar grid structure. For example, time series data (e.g. time axis discrete sampling) and image data (e.g. two-dimensional discrete sampling) can be considered as similar grid structure data. CNN does not use all input information at once for operation, but uses a fixed size window to extract part of the information for convolution operation, which greatly reduces the calculation of model parameters. In addition, according to the different types of window extraction information (such as people and objects in the same picture are different types of information), each window can use different convolution kernel operation, which makes CNN better extract the features of input data.

[0191] RNN is a kind of neural network that uses feedback time series information. The input of RNN includes the new input value at the current time and the output value of itself at the previous time. RNN is suitable for obtaining sequence characteristics with temporal correlation, such as speech recognition, channel coding and decoding applications.

[0192] In the above model training process of machine learning, a loss function can be defined. The loss function describes the gap or difference between the output value of the model and the ideal target value. The loss function can be embodied in various forms, and the specific form of the loss function is not limited. The model training process can be regarded as the following process: by adjusting part or all of the parameters of the model, the value of the loss function is less than the threshold value or meets the target demand.

[0193] The model can also be called an AI model, a rule or other names. The AI model can be considered as a specific method to realize the AI function. The AI model represents the mapping relationship or function between the input and output of the model. The AI function can include one or more of the following: data collection, model training (or model learning), model information publishing, model inference (or model reasoning, reasoning, or prediction, etc.), model monitoring or model verification, or inference result publishing, etc. The AI function can also be called AI (related) operation, or AI related function.

[0194] The implementation process of neural network will be described below with reference to the accompanying drawings.

[0195] 1. Fully connected neural network, also known as multilayer perceptron (MLP).

[0196] As shown in Figure 2c, an MLP includes an input layer (left side), an output layer (right side), and multiple hidden layers (middle). Each layer of the MLP includes a number of nodes, called neurons. Among them, the neurons of adjacent two layers are connected to each other.

[0197] Optionally, considering the neurons of two adjacent layers, the output h of the neuron of the next layer is the weighted sum of all the neurons x of the previous layer connected thereto and is processed by an activation function, which can be expressed as: h = f(w x + b).

[0198] wherein w is a weight matrix, b is a bias vector, and f is an activation function.

[0199] Further optionally, the output of the neural network can be recursively expressed as: y = f z (w z f z-1 (…)+b z ).

[0200] wherein z is the index of the layer of the neural network, z is greater than or equal to 1, and z is less than or equal to Z, wherein Z is the total number of layers of the neural network.

[0201] In other words, the neural network can be understood as a mapping relationship from a set of input data to a set of output data. Usually, the neural network is randomly initialized, and the process of obtaining this mapping relationship from the random w and b using the existing data is called training of the neural network.

[0202] Optionally, the specific way of training is to evaluate the output result of the neural network by using a loss function.

[0203] As shown in FIG. 2d, the error can be back-propagated, and the neural network parameters (including w and b) can be iteratively optimized by the gradient descent method until the output of the loss function reaches a minimum value, i.e., the “better point (e.g., optimal point)” in FIG. 2d. It can be understood that the neural network parameters corresponding to the “better point (e.g., optimal point)” in FIG. 2d can be used as the neural network parameters in the trained AI model information.

[0204] Further optionally, the process of gradient descent can be expressed as:

[0205] wherein θ is the parameter to be optimized (including w and b), L is the loss function, η is the learning rate, which controls the step size of gradient descent, represents the derivation operation, represents the derivative of L with respect to θ.

[0206] Further optionally, the process of back-propagation utilizes the chain rule of partial derivative.

[0207] As shown in FIG. 2e, the gradient of the parameters of the previous layer can be recursively calculated from the gradient of the parameters of the next layer, which can be expressed as:

[0208] wherein w ij is the weight of node j connected to node i, and si The weighted sum of the inputs at node i.

[0209] 2. Federated Learning (FL).

[0210] The concept of federated learning effectively addresses the current challenges in the development of artificial intelligence. While fully protecting user data privacy and security, it enables various edge devices and central servers to collaborate efficiently to complete the model's learning task.

[0211] As shown in Figure 2f, the FL architecture is the most widely used training architecture in the current FL field, and the FedAvg algorithm is the basic algorithm of FL. The FedAvg algorithm flow is roughly as follows:

[0212] (1) Initialize the model to be trained at the center end. And broadcast it to all clients.

[0213] (2) In the t∈[1,T] round, the client k∈[1,K] is based on the local dataset. For the received global model Perform E epochs of training to obtain the local training results. This is then reported to the central node. In the example shown in Figure 2f, the local training results sent by distributed nodes n, k, and m are denoted as G, respectively. n G k G m .

[0214] (3) The central node collects local training results from all (or some) clients. Assume the set of clients uploading local models in round t is... The central server will use the number of samples from the corresponding client as weights to calculate the new global model. The specific update rule is as follows: Then the central end will send the latest version of the global model. The broadcast is sent to all clients for a new round of training.

[0215] (4) Repeat steps (2) and (3) until the model finally converges or the number of training rounds reaches the upper limit.

[0216] Optionally, in addition to reporting the local model, the client can also... It can also train local gradients The central node averages the local gradients reported by all clients and updates the global model based on this average gradient.

[0217] It can be seen that in the FL framework, the dataset exists at the distributed nodes (such as clients), that is, the distributed nodes collect local datasets and perform local training, and report the local results (models or gradients) obtained by training to the center node. The center node itself can have no dataset, and can be responsible for fusing the training results of the distributed nodes to obtain a global model and issuing the global model to the distributed nodes.

[0218] 3. Decentralized learning.

[0219] As shown in FIG. 2g, it is a completely distributed system without a center node. The design goal f(x) of the decentralized learning system is generally the average of the goals f i (x) of each node, that is, where n is the number of distributed nodes, and x is the parameter to be optimized in machine learning, that is, the parameter of the machine learning (such as neural network) model. Each node calculates the local gradient i (x) using local data and local goal f Then it is sent to the neighbor nodes that are communicable. After receiving the gradient information sent by the neighbor nodes, any node can update the parameter x of the local model according to the following formula:

[0220] wherein, represents the parameter of the local model of the i-th node after the k+1 (k is a natural number) time of updating, represents the parameter of the local model of the i-th node after the k time of updating (if k is 0, it represents is the parameter of the local model of the i-th node that does not participate in updating), a k represents the tuning coefficient, N i is the neighbor node set of node i, and |N i | represents the number of elements in the neighbor node set of node i, that is, the number of neighbor nodes of node i. Through the information interaction between nodes, the decentralized learning system will finally learn a unified model.

[0221] The technical scheme provided in the application can be applied to a communication system (such as the system shown in FIG. 1a or FIG. 1b or FIG. 1c). In the communication system, the communication nodes generally have signal transceiving capability and computing capability. Taking a network device with computing capability as an example, the computing capability of the network device is mainly to provide computing power support for the signal transceiving capability (for example: to perform sending processing and receiving processing on the signal) to realize the communication task of the network device and other communication nodes.

[0222] With the development of communication technology, in a communication system, in addition to the traditional communication service, the service performed by the communication device can also include other new services, such as artificial intelligence (AI) services. Generally, a system capable of processing AI services, such as a communication system, can also be referred to as an AI system. At present, the communication device can act as a participating node of the AI system, and the computing power of the communication device is applied to a certain link of the AI system. Generally speaking, the AI function introduced in the communication network needs to rely on a model to be implemented. However, in the above process, how to manage the model has not yet been solved by the current related solutions.

[0223] To solve the above problems, the present application provides a communication method and related devices, which will be described in detail below in conjunction with the accompanying drawings.

[0224] Please refer to FIG. 3, which is an implementation schematic diagram of the communication method provided by the present application. The method includes the following steps.

[0225] It should be noted that in the following, the first communication device and other communication devices (such as the second communication device, or the third communication device, etc.) are taken as an example to illustrate the execution subject of the interaction in FIGS. 3-5, but the present application does not limit the execution subject of the interaction. For example, the communication device can be a communication device, or a chip, a baseband chip, a modem chip, a system on chip (SoC) chip containing a modem core, a system in package (SIP) chip, a communication module, a chip system, a processor, a logic module or software, etc. in the communication device. Optionally, the communication device can be a terminal device or a network device (such as an access network device, an access network element, a core network element, or a core network device, etc.).

[0226] S301. The first communication device obtains first information. Wherein the first information is used to determine whether the performance of the first model deployed in the first communication device meets the first condition.

[0227] S302. The second communication device sends second information, and correspondingly, the first communication device receives the second information. Wherein the second information is used to determine whether the performance of the second model deployed in the second communication device meets the second condition. Wherein the second model and the first model are used to jointly complete the AI task.

[0228] S303. The first communication device determines third information based on the first information and the second information, the third information being used to indicate adjusting the first model and / or adjusting the second model.

[0229] In this application, the AI task can be replaced by other terms, such as task, AI use case, AI function, AI-enabled function, or AI-enabled feature, etc.

[0230] In this application, the model can include a mathematical model, an AI model, a neural network model, an AI neural network model, a machine learning model, or an AI processing model, etc. For example, the model can be used for CSI feedback enhancement, end-to-end transceiver, etc. to improve network performance. For another example, the model can be used for other applications, such as transmission and / or enhancement of audio / video, data processing related to large language models, etc.

[0231] In this application, one model is deployed in one communication device (for example, the first model is deployed in the first communication device, and the second model is deployed in the second communication device, etc.), which can be understood as that the one communication device obtains the model parameters of the one model, and then obtains / generates / constructs the one model based on the model parameters of the one model. Subsequently, the one communication device can perform model adjustment on the one model.

[0232] Optionally, the model parameters can include one or more of the hyperparameters of the model, the dataset of the model (including the input data of the model and the label data corresponding to the input data, etc.), the structure of the model, the weight of the model, and the parameters of the model.

[0233] It should be understood that two or more models are used to jointly complete an AI task, which can be understood as that these models complete the same AI task through cooperation. The processing of these models can be serial (or chain), parallel, or a combination of serial and parallel. For example, the execution result of the AI task can be obtained based on the output of these models. For example, in the case where the number of models jointly completing the AI task is 2, these models can be referred to as a dual-end model; wherein the dual-end model includes the first model and the second model involved in this application. For another example, in the case where the number of models jointly completing the AI task is greater than 2, these models can be referred to as a multi-end model; wherein the multi-end model includes at least the first model and the second model involved in this application.

[0234] Optionally, adjusting the model (for example, adjusting the first model and / or adjusting the second model) can include one or more of resetting the model, retraining the model, initializing the model, training the model, updating the model, fine-tuning the model, or fine-tuning the model.

[0235] Optionally, the first model and the second model can have different association relationships (or cooperation relationships). Some examples will be provided below, taking the first model and the second model as examples for illustrative purposes.

[0236] Implementation Example 1: The first model and the second model are different parts of the models used to jointly complete the AI task, respectively.

[0237] In an implementation example one, the model for jointly completing the AI task can include a first model and a second model. For example, the output of the first model can be part or all of the input of the second model, and the output of the second model can be obtained by processing the second model, and the execution result of the AI task can include the output of the second model (optionally, the output of the first model can also be included). For another example, the output of the second model can be part or all of the input of the first model, and the output of the first model can be obtained by processing the first model, and the execution result of the AI task can include the output of the first model (optionally, the output of the second model can also be included).

[0238] In an implementation example two, the first model and the second model are partially the same.

[0239] In the implementation example two, the first model can include the model for jointly completing the AI task, and the second model can include a part of the model for jointly completing the AI task. Alternatively, the second model can include the model for jointly completing the AI task, and the first model can include a part of the model for jointly completing the AI task.

[0240] For example, the first model can include one or more sub-models, and the second model is one of the one or more sub-models, i.e., the second model is a sub-module (or a subset) of the first model; in this case, the first model can include the model for jointly completing the AI task, and the second model can include a part of the model for jointly completing the AI task. In other words, the first model can be referred to as a complete model (or a complete model for jointly completing the AI task), and the second model can be referred to as a partial model (or a partial model for jointly completing the AI task).

[0241] For another example, the second model can include one or more sub-models, and the first model is one of the one or more sub-models, i.e., the first model is a sub-module (or a subset) of the second model; in this case, the second model can include the model for jointly completing the AI task, and the second model can include a part of the model for jointly completing the AI task. In other words, the second model can be referred to as a complete model (or a complete model for jointly completing the AI task), and the first model can be referred to as a partial model (or a partial model for jointly completing the AI task).

[0242] In an implementation example three, the first model and the second model are the same.

[0243] In the third implementation example, the first model and the second model can each include a model for jointly completing an AI task. In a certain execution of the joint AI task, the first model can participate in the execution partially or entirely, and the second model can also participate in the execution partially or entirely. The first model and the second model can each independently complete the AI task. In other words, the first model and the second model can be referred to as complete models (or complete models for jointly completing an AI task).

[0244] The above implementation examples will be described below with reference to FIGS. 4a-4c, taking a two-end model as an example of the first model and the second model. The two-end model can be used in a CSI compression feedback process in CSI feedback enhancement. For example, the two-end model is deployed on different devices (such as a terminal device and a network device). The first communication apparatus can be the terminal device and the second communication apparatus can be the network device, or the second communication apparatus can be the terminal device and the first communication apparatus can be the network device. The CSI compression feedback process is a typical scenario using a two-end model. Generally, after the terminal device performs channel estimation, the terminal device uses a CSI compression neural network (also referred to as an encoder (E) neural network) on the local side to compress the CSI to obtain compressed CSI with a smaller data volume, and sends the compressed CSI to the network device. The network device inputs the received compressed CSI into a CSI reconstruction neural network (also referred to as a decoder (D) neural network) to reconstruct the CSI, and performs precoding for downlink transmission based on the reconstructed CSI.

[0245] As shown in FIG. 4a, in the first implementation example, the first model and the second model are different parts of the model for jointly completing the AI task. The first model is model “E” and the second model is model “D”. The first communication apparatus can be the network device in FIG. 4a and the second communication apparatus can be the terminal device in FIG. 4a. Alternatively, the first model is model “D” and the second model is model “E”. The first communication apparatus can be the terminal device in FIG. 4a and the second communication apparatus can be the network device in FIG. 4a. The two devices can transmit processing data (including input data and / or output data) of model “E” and / or model “D” through a communication process to implement the CSI compression feedback process.

[0246] As shown in FIG. 4b, in the above-mentioned implementation example two, the first model and the second model are the same. Among them, the first model and the second model both include the model “E” and the model “D”, i.e., the first communication device can be the network equipment in FIG. 4b and the second communication device can be the terminal equipment in FIG. 4b, or the second communication device can be the network equipment in FIG. 4b and the first communication device can be the terminal equipment in FIG. 4b. The two devices can transmit the processing data (including the input data and / or the output data) of the model “E” and / or the model “D” through a communication process to implement the CSI compressed feedback process.

[0247] As shown in FIG. 4c, in the above-mentioned implementation example three, the first model and the second model are partially the same. Among them, the first model includes the model “E” and the model “D”, and the second model is the model “D”, i.e., the first communication device can be the network equipment in FIG. 4c and the second communication device can be the terminal equipment in FIG. 4c. Or, the first model is the model “D” and the second model includes the model “E” and the model “D”, i.e., the first communication device can be the terminal equipment in FIG. 4c and the second communication device can be the network equipment in FIG. 4c. The two devices can transmit the processing data (including the input data and / or the output data) of the model “E” and / or the model “D” through a communication process to implement the CSI compressed feedback process.

[0248] Optionally, when the two-side devices come from different manufacturers, a matching method is generally needed to align the two-side models. In a commonly used model alignment method, a complete neural network (i.e., an encoder and a decoder) is trained on one side (such as a network equipment), and then the information (using an encoder itself, or a data set composed of the input and output of the encoder) required by the partial neural network (such as an encoder required by a terminal equipment) of the other side is sent to the terminal equipment. So that the terminal equipment obtains the encoder based on the received information (the encoder itself or the data set required for training the encoder).

[0249] It should be noted that the performance of the model (such as the performance of the first model, the performance of the second model, etc.) can be characterized by a plurality of parameters, and correspondingly, the condition corresponding to the performance of the model (such as the first condition corresponding to the first model, the second condition corresponding to the second model, etc.) can be a condition adapted to the parameter. Among them, the condition can be implemented in a plurality of ways. For example, in way A, satisfying the condition can be understood as that the performance of the model is better, and not satisfying the condition can be understood as that the performance of the model is worse. For another example, in way B, satisfying the condition can be understood as that the performance of the model is worse, and not satisfying the condition can be understood as that the performance of the model is better (denoted as way B).

[0250] As an example, the performance of the model can be determined based on a difference between the output of the model and an expected output (or target output), and accordingly, the condition corresponding to the performance of the model can indicate whether the difference is greater than a threshold. Wherein, the greater the difference, the lower the performance of the model; on the contrary, the smaller the difference, the higher the performance of the model.

[0251] As another example, the performance of the model can be determined based on a communication performance of a communication process to which the output of the model is applied, and accordingly, the condition corresponding to the performance of the model can be a threshold of the communication performance. Wherein, the higher the communication performance, the higher the performance of the model; on the contrary, the lower the communication performance, the lower the performance of the model.

[0252] For example, the performance of the model is determined based on a difference between the output of the model and an expected output (or target output) as described above. In the above manner B, the condition is satisfied can be understood as the performance of the model is lower than (or equal to) the threshold indicated by the condition, and the condition is not satisfied can be understood as the performance of the model is higher than (or equal to) the threshold indicated by the condition. In the above manner A, the condition is satisfied can be understood as the performance of the model is higher than (or equal to) the threshold indicated by the condition, and the condition is not satisfied can be understood as the performance of the model is lower than (or equal to) the threshold indicated by the condition.

[0253] Optionally, the difference can be represented by a mathematical calculation result of different outputs. For example, the mathematical calculation result can include one or more of a difference value, a mean square error (MSE), a normalized mean square error (NMSE), or a cosine similarity.

[0254] Optionally, the communication performance can be represented by a communication parameter. For example, the communication parameter can include one or more of a reference signal received power (RSRP), a reference signal received power quality (RSRQ), or a signal and interference plus noise ratio (SINR).

[0255] For example, the first communication device can be the network device or the terminal device in FIG. 4c, taking the scenario shown in FIG. 4c as an example. Wherein, the first communication device can obtain the first information (i.e., determine whether the performance of the first model deployed locally satisfies the first condition) in a variety of ways.

[0256] For example, the first communication device can input the reference data into the local encoder to obtain an output, and then input the output of the encoder into the local decoder to obtain an output. Thereafter, the first communication device calculates a correlation and / or a similarity index based on the output of the decoder and the reference data, and if the index is less than a preset threshold, it indicates that the performance of the local model is abnormal (i.e., the first condition is not met); otherwise, it indicates that the performance of the local model is normal (i.e., the first condition is met).

[0257] For another example, the first communication device can input the encoder output (denoted as Z) received from the opposite side before starting the fault detection into the local decoder to obtain an output of estimated channel information (denoted as H'). Thereafter, the first communication device inputs H' into the local encoder to obtain an output denoted as Z', and inputs Z' into the local decoder to obtain an output denoted as H". The first communication device can calculate a correlation and / or a similarity index based on H' and H", and if the index is less than a preset threshold, it indicates that the performance of the local model is abnormal (i.e., the first condition is not met); otherwise, it indicates that the performance of the local model is normal (i.e., the first condition is met).

[0258] In a possible implementation, the first information obtained by the first communication device in step S301 includes at least one of the following:

[0259] First indication information for indicating the output data of the first model;

[0260] Second indication information for indicating the difference between the output data of the first model and the expected output;

[0261] Third indication information for indicating the difference between a second processing result and a third processing result; the second processing result indicates a processing result obtained by processing the output data of the second model by the first model, and the third processing result indicates a processing result obtained by processing the second processing result by the first model; wherein the processing of the first model and the processing of the second model are partially or wholly the same (for reference to the processing process of H' and H" described above);

[0262] Fourth indication information for indicating whether the performance of the first model meets the first condition.

[0263] Therefore, the first information for determining whether the performance of the first model deployed in the first communication device meets the first condition can be implemented in the above-mentioned multiple ways, so as to improve the flexibility of the implementation scheme.

[0264] In a possible implementation, the second information received by the first communication device in step S302 includes at least one of the following:

[0265] The fifth indication information is used to indicate the output data of the second model.

[0266] The sixth indication information is used to indicate the difference between the output data of the second model and the expected output.

[0267] The seventh indication information is used to indicate the difference between the third processing result and the fourth processing result; the third processing result indicates the processing result of the second model on the output data of the first model, and the fourth processing result indicates the processing result of the second model on the third processing result.

[0268] The eighth indication information is used to indicate whether the performance of the second model meets the first condition.

[0269] Therefore, the second information used to determine whether the performance of the second model deployed in the second communication device meets the second condition can be implemented in the above-mentioned multiple ways, so as to improve the flexibility of the scheme implementation.

[0270] Based on the scheme shown in FIG. 3, the first communication device can determine whether the performance of the first model deployed in the first communication device meets the first condition through the first information obtained in step S301, and determine whether the performance of the second model deployed in the second communication device meets the second condition through the second information received in step S302. Thereafter, the first communication device can determine the third information used to adjust the first model and / or adjust the second model based on the first information and the second information in step S303. Wherein, the second model and the first model are used to jointly complete the AI task. In this way, in the case that different models deployed in different communication devices are used to jointly complete the same AI task, the first communication device can determine whether to adjust the different models based on the condition that the performance of the different models meets the condition (or does not meet the condition), so as to realize the model management of the models.

[0271] In addition, in the above-mentioned scheme, the first communication device can determine the third information used to adjust the first model deployed in the first communication device and / or adjust the second model deployed in the second communication device based on the first information and the second information. For this purpose, the above-mentioned scheme can be applied to the scene that multiple nodes in a distributed manner cooperatively process the same AI task, the multiple nodes at least include the first communication device and the second communication device, and the performance of the different models meeting the condition (or not meeting the condition) can be used to accurately determine the node that appears to be faulty, so as to avoid the influence on the entire distributed system, and also avoid the redundant operation caused by the adjustment of the non-faulty device, so as to improve the efficiency of the model management.

[0272] In a possible implementation manner of the method shown in FIG. 3, taking the previous manner A as an example, the third information determined by the first communication device in step S303 meets any one of the following conditions:

[0273] In a case (denoted as Case 1) that the performance of the first model determined based on the first information does not satisfy the first condition and the performance of the second model determined based on the second information satisfies the second condition, the first communication device can determine that the performance of the first model is poor and the performance of the second model is optimal, and the third information is used to indicate that the model adjustment operation is performed on the first model; or,

[0274] In a case (denoted as Case 2) that the performance of the first model determined based on the first information satisfies the first condition and the performance of the second model determined based on the second information does not satisfy the second condition, the first communication device can determine that the performance of the first model is optimal and the performance of the second model is poor, and the third information is used to indicate that the model adjustment operation is performed on the second model; or,

[0275] In a case (denoted as Case 3) that the performance of the first model determined based on the first information does not satisfy the first condition and the performance of the second model determined based on the second information does not satisfy the second condition, the first communication device can determine that the performance of the first model and the performance of the second model are poor, and the third information is used to indicate that the model adjustment operation is performed on the first model and the second model.

[0276] In a case (denoted as Case 4) that the performance of the first model determined based on the first information satisfies the first condition and the performance of the second model determined based on the second information satisfies the second condition, and the performance of the AI task is lower than a threshold value, the third information is used to indicate that the model adjustment operation is performed on the first model and the second model.

[0277] Optionally, taking the foregoing manner B as an example, the third information determined by the first communication device in step S303 satisfies any one of the following:

[0278] In a case that the performance of the first model determined based on the first information does not satisfy the first condition and the performance of the second model determined based on the second information satisfies the second condition, the first communication device can determine that the performance of the first model is optimal and the performance of the second model is poor, and the third information is used to indicate that the model adjustment operation is performed on the second model; or,

[0279] In a case that the performance of the first model determined based on the first information satisfies the first condition and the performance of the second model determined based on the second information does not satisfy the second condition, the first communication device can determine that the performance of the first model is poor and the performance of the second model is optimal, and the third information is used to indicate that the model adjustment operation is performed on the second model; or,

[0280] In a case where it is determined based on the first information that the performance of the first model satisfies the first condition and based on the second information that the performance of the second model satisfies the second condition, the first communication apparatus can determine that the performance of the first model and the performance of the second model are poor, and the third information is used to instruct a model adjustment operation on the first model and the second model.

[0281] In a case where it is determined based on the first information that the performance of the first model does not satisfy the first condition and based on the second information that the performance of the second model satisfies the second condition, and the performance of the AI task is lower than a threshold value, the third information is used to instruct a model adjustment operation on the first model and the second model.

[0282] Thus, the third information can be implemented in the above-mentioned multiple ways, and in a case where the joint processing of the first model and the second model fails (for example, the performance of the joint processing is poor), the cause of the failure can be identified by the first information and the second information, and the cause of the failure can be troubleshooted by the third information.

[0283] For example, in a case where the performance of the AI task completed by the first model and the second model in combination is poor (for example, the performance of the first model does not satisfy the first condition, the performance of the second model does not satisfy the second condition, the performance of the output data of the AI task is lower than a threshold value, the performance of the AI task is lower than a threshold value, etc.), the first communication apparatus can determine that the first model and / or the second model is likely to fail. Taking the above-mentioned way B as an example, the following will describe the specific implementation process corresponding to three cases with some examples.

[0284] In the above-mentioned case 1, the first communication apparatus can determine that the failure is caused by the model performance degradation of the first model (and the failure is very likely not caused by the model performance degradation of the second model), and for this purpose, the third information determined by the first communication apparatus based on the first information and the second information can instruct a model adjustment operation on the first model to troubleshoot the failure.

[0285] In the above-mentioned case 2, the first communication apparatus can determine that the failure is caused by the model performance degradation of the second model (and the failure is very likely not caused by the model performance degradation of the first model), and for this purpose, the third information determined by the first communication apparatus based on the first information and the second information can instruct a model adjustment operation on the second model to troubleshoot the failure.

[0286] In the above-mentioned case 3, the first communication apparatus can determine that the failure is caused by the model performance degradation of the first model and the second model, and for this purpose, the third information determined by the first communication apparatus based on the first information and the second information can instruct a model adjustment operation on the first model and the second model to troubleshoot the failure.

[0287] In the case 4, the first communication device can determine that the failure is not caused by the performance degradation of the first model and the second model. As described above, in some cases, the first information can represent whether the failure is caused by the performance degradation of the first model, and the second information can represent whether the failure is caused by the performance degradation of the second model. Therefore, the first communication device can determine that the failure is likely caused by the inadaptability of the data used by the models (e.g., data drift), and the third information determined by the first communication device is used to indicate the model adjustment operation on the first model and the second model.

[0288] Optionally, in the case that the third information is used to indicate the model adjustment operation on the second model, the method shown in FIG. 3 further includes that the first communication device sends the third information, and correspondingly, the second communication device can receive the third information. In other words, the first communication device can indicate the model adjustment operation on the second model to the second communication device through the sent third information, and in the case that the joint processing of the first model and the second model fails (e.g., the performance of the joint processing is low), and the cause of the failure is the second model, the second model deployed by the second communication device can be adjusted through the third information to eliminate the failure.

[0289] Optionally, in the case that the third information is used to indicate the model adjustment operation on the first model and the second model, the method shown in FIG. 3 further includes that the first communication device sends fourth information, and the fourth information is used to indicate the model adjustment operation on the second model. In other words, the first communication device can indicate the model adjustment operation on the second model to the second communication device through the sent fourth information, and in the case that the joint processing of the first model and the second model fails (e.g., the performance of the joint processing is low), the second model deployed by the second communication device can be adjusted through the fourth information to eliminate the failure.

[0290] Optionally, in a case where the performance of the AI task completed by the first model in combination with the second model is poor (e.g., the performance of the first model does not satisfy the first condition, the performance of the second model does not satisfy the second condition, the performance of the output data of the AI task is lower than a threshold, the performance of the AI task is lower than a threshold, etc.), the first communication device can determine that the first model and / or the second model may have a fault. As described above, in some cases, the first information can represent whether the fault is caused by the model performance degradation of the first model, and the second information can also represent whether the fault is caused by the model performance degradation of the second model. In a case where the first information and the second information represent that the fault is not caused by the model performance degradation of the first model and the second model (i.e., case 4), the cause of the fault is not caused by the first model and the second model, but is likely caused by the inadaptation of the data used by the model (e.g., data drift). Therefore, the third information determined by the first communication device is used to indicate the model adjustment operation on the first model and the second model, and accordingly, the first communication device can perform the model adjustment operation on the second model deployed by the second communication device through the fourth information to eliminate the fault.

[0291] It should be noted that in the scheme shown in FIG. 3, the first communication device can trigger the acquisition of the first information in step S301 in a plurality of ways, which will be described below in combination with some possible implementation manners.

[0292] In implementation manner one, in step S301, the first communication device triggers the acquisition of the first information based on the processing process of the first model deployed by itself. For example, the first information can be used to indicate the performance obtained by the first model using the first parameter for data processing; and in step S301, the process of the first communication device acquiring the first information includes: in a case where the performance obtained by the first model using the second parameter for data processing is lower than (or equal to) a threshold, the first communication device acquires the first information; wherein the first parameter is the parameter corresponding to the first model after the model adjustment operation is performed on the first model, and the second parameter is the parameter corresponding to the first model before the model adjustment operation is performed (or completed) on the first model.

[0293] In implementation manner one, for the first model deployed by the first communication device, in a case where the performance of the processing process of the parameter corresponding to the first model before the model adjustment operation is performed (or completed) is lower than a threshold, the first communication device can be triggered to acquire the first information, and the first information is used to indicate the performance obtained by the first model using the parameter corresponding to the first model after the model adjustment operation is completed for data processing. In this way, the first communication device can trigger the performance detection of the parameter after the model adjustment based on the processing of the parameter before the model adjustment, and determine the first information based on the processing of the parameter after the model adjustment.

[0294] It should be understood that the performance of data processing using a certain parameter by a model (e.g., the performance of data processing using a first parameter by a first model, the performance of data processing using a second parameter by the first model, etc.) can be understood as the performance of output data obtained by model processing of input data based on the model after the model parameter of the model is updated (or set) to the parameter. For example, the performance of the output data can be characterized by the aforementioned difference, communication performance, etc.

[0295] Optionally, the model adjustment can include one or more of model resetting, model initializing, model retraining, model training, model updating, model fine-tuning, or model fine-tuning. For example, in the above scheme, taking the model adjustment as model resetting as an example, the first parameter can be the model parameter after resetting, and the second parameter can be the model parameter before resetting (e.g., the model parameter used by the first model in performing a certain AI task).

[0296] Optionally, in the first implementation, the second information sent by the second communication device in step S302 can be obtained by processing the first processing result based on the second model, and the first processing result is obtained by processing the first model based on the second parameter. Specifically, the first model and the second model can be used to jointly complete the same AI task, that is, the AI task can be completed based on the processing of the first model and the processing of the second model. Wherein, the second information can be obtained by processing the first processing result based on the second model, and the first processing result can be obtained by processing the first model based on the parameter before the model adjustment, so that the first model participates in the joint processing of the first model and the second model through the parameter before the model adjustment, avoiding the influence of the operation of the model adjustment of the first model on the performance of the second model. In addition, in the case that the joint processing of the first model and the second model fails (e.g., the performance of the joint processing is low), the above scheme can exclude the influence caused by the model adjustment of the first model, and through the second information, it can be determined whether the failure is caused by the performance degradation of the second model, so as to realize accurate positioning of the failure and improve the efficiency of model management.

[0297] Optionally, in the first implementation, the second information sent by the second communication device in step S302 can be used to indicate the performance of the second model using the third parameter for data processing, and the second information is obtained in a case where the performance of the second model using the fourth parameter for data processing is lower than a threshold; the third parameter is a parameter after the model adjustment operation is performed on the second model, and the fourth parameter is a parameter before the model adjustment operation is performed on the second model. Specifically, the second information can be used to indicate the performance of the second model using the third parameter for data processing, and the third parameter is a parameter after the model adjustment operation is performed on the second model, so that the second model uses the parameter before the model adjustment to participate in the joint processing of the first model and the second model, avoiding the influence of the model adjustment operation on the performance of the second model. In addition, in a case where the joint processing of the first model and the second model fails (for example, the performance of the joint processing is low), the above scheme can exclude the influence caused by the model adjustment of the second model, and through the second information, it can be indicated whether the failure is caused by the degradation of the model performance of the second model, so as to realize accurate positioning of the failure and improve the efficiency of model management.

[0298] Optionally, the second information is obtained in a case where the performance of the second model using the fourth parameter and the first processing result for data processing is lower than a threshold. Thus, in a case where the joint processing of the first model and the second model fails (for example, the performance of the joint processing is low), the above scheme can exclude the influence caused by the model adjustment of the first model and the model adjustment of the second model, and through the second information, it can be indicated whether the failure is caused by the degradation of the model performance of the second model, so as to realize accurate positioning of the failure and improve the efficiency of model management.

[0299] In the first implementation, before step S302, the method shown in FIG. 3 can further include: in a case where the performance of the first model meets the first condition based on the first information, the first communication device sends request information used to request the second information. Specifically, in a case where the performance of the first model meets the first condition based on the first information, the first communication device can determine that the first model is normal (or has no failure), and accordingly, in a case where the joint processing of the first model and the second model fails (for example, the performance of the joint processing is low), the first communication device can determine the cause of the failure based on the above process and the cause is not the first model. Therefore, the first communication device can send the above request information to determine whether the cause of the failure is the second model based on the second information received subsequently.

[0300] Further, the first communication device determines whether the first model deployed on itself is faulty through the first information, which can be understood as a process of model monitoring of the first communication device on the model of itself, which can be referred to as a process of model self-checking (process 1 for short); and the first communication device determines whether the second model deployed on the second communication device is faulty through the second information, which can be understood as a process of model monitoring of the first communication device on the model deployed on the opposite end (process 2 for short). In the two processes, the former can generally be executed locally, and the latter generally needs to determine the information (for example, the second information) of the interaction of the opposite end, and for this purpose, in the above scheme, the manner of executing process 1 first and then executing process 2 can avoid unnecessary overhead, so as to improve the efficiency of model management.

[0301] It should be noted that process 1 can be understood as an implementation example of the foregoing step S301, and process 2 can be understood as an implementation example of the foregoing step S302. In the implementation of the above process 1 and process 2, process 2 can be triggered and executed based on process 1, that is, the first communication device triggers process 2 in the case that it is determined through process 1 that the model deployed on the local end is normal (or not faulty). Correspondingly, the first communication device can determine the third information based on the second information in step S303 (that is, the first information can not be a basis for determining the third information). In this case, step S303 shown in FIG. 3 can be expressed as:

[0302] S303. The first communication device determines the third information based on the second information, and the third information is used to indicate adjustment of the second model, or the third information is used to indicate adjustment of the first model and the second model.

[0303] Through the above process, the first communication device can execute the local model self-checking through step S301 to determine the case that the local model is normal (or not faulty), and execute the opposite end model self-checking through step S302.

[0304] In the case that the self-checking result of the opposite end model in step S302 (that is, the second information) indicates that the second model is abnormal (or faulty), the third information determined by the first communication device in step S303 can be used to determine adjustment of the second model, so as to exclude the fault by model adjustment (for example, model parameter reset) on the model parameter of the second model.

[0305] In the case that the self-checking result of the opposite end model in step S302 (that is, the second information) indicates that the second model is normal (or not faulty), the third information determined by the first communication device in step S303 can be used to determine adjustment of the first model and adjustment of the second model, so as to exclude the fault by model adjustment (for example, model parameter reset) on the model parameters of the two models.

[0306] In implementation two, in step S301, the first communication apparatus triggers the acquisition of the first information based on the second information. For example, in the process of acquiring the first information in step S301, the first communication apparatus acquires the first information in a case where it is determined based on the second information that the performance of the second model satisfies the second condition.

[0307] In implementation two, in a case where it is determined based on the second information that the performance of the second model satisfies the second condition, the first communication apparatus can determine that the second model is not abnormal (or not faulty), and accordingly, in a case where the joint processing of the first model and the second model is faulty (for example, the performance of the joint processing is low), the first communication apparatus can determine the cause of the fault based on the above process and determine that the cause of the fault is not caused by the second model. To this end, the first communication apparatus can acquire the first information to determine whether the cause of the fault is caused by the first model through the first information, and unnecessary model monitoring of the first model can be avoided to improve the efficiency of model management.

[0308] It can be understood that in implementation two, the first communication apparatus performs step S302 before performing step S301, that is, the second information received by the first communication apparatus in step S302 can be used to trigger the first communication apparatus to acquire the first information in step S301.

[0309] It should be noted that in step S303, the first communication apparatus can trigger the determination of the third information through the above-mentioned implementation one or implementation two. For example, in step S303, the process of determining the third information by the first communication apparatus based on the first information and the second information includes: in a case where the performance of the first model processed by the second parameter is lower than (or equal to) the threshold, or in a case where it is determined based on the second information that the performance of the second model satisfies the second condition, the first communication apparatus determines the third information based on the first information and the second information. The implementation of these triggering processes can refer to the above-mentioned implementation one or implementation two.

[0310] As an application example, implementation one can be applied to the scenario of the double-end model shown in FIGS. 4b and 4c, in which the first communication apparatus can be a device with a complete model, such as the network device in FIG. 4b or the network device or terminal device in FIG. 4c. In a case where the double-end model is faulty, the first communication apparatus can perform fault detection through a fault detection process, which can include the following three steps:

[0311] Step 1. Full model side self-check: if a fault is found, the full model on the side is reset. After resetting, if the fault is eliminated, the process ends, otherwise, go to Step 2. For example, in Step 1, the first communication device can perform a full model side self-check, and the self-check result can be the first information in the aforementioned step S301 (i.e., the first communication device can obtain the first information through the process of full model side self-check here).

[0312] Optionally, in Step 1, if the first communication device determines that the full model on the side is faulty based on the first information, the first communication device can perform a model adjustment operation on the first model deployed by the first communication device.

[0313] Step 2. Opposite side model self-check: if a fault is found, the opposite side model is reset. After resetting, if the fault is eliminated, the process ends, otherwise, go to Step 3. For example, in Step 2, the first communication device can determine the opposite side model self-check result through the second information received in step S302.

[0314] Optionally, in Step 2, if the first communication device determines that the opposite side model is faulty based on the second information, the first communication device can perform a model adjustment operation on the second model deployed by the second communication device.

[0315] Step 3. Double side model reset: if neither the full model on the side nor the opposite side model has a problem or the fault is not eliminated after resetting, data drift may have occurred, i.e., the current model is not suitable for the data in the current scenario, so the double side model needs to be reset to complete the fault detection and elimination process.

[0316] For example, in Step 3, if the first communication device performs a full model self-check and determines that no fault occurs, and the first communication device performs an opposite side model self-check and determines that no fault occurs, the third information determined by the first communication device in step S303 is used to indicate that a model adjustment operation is performed on the first model deployed by the first communication device and the second model deployed by the second communication device.

[0317] The following will take the first communication device as the network device in FIG. 4b (with a full model) and the second communication device as the terminal device in FIG. 4b (with a partial model) as an example to illustrate the above three steps in combination with more drawings.

[0318] As shown in FIG. 4d, it is an implementation example of the full model side self-check process of the above-mentioned Step 1, which includes the following steps.

[0319] A1. Double side CSI compression feedback reconstruction, i.e., the terminal device performs CSI compression, feeds back the compressed CSI, and the network device performs CSI reconstruction. The network device performs model performance monitoring, finds performance abnormalities, and starts the fault detection process.

[0320] A2. The complete model side performs self-checking.

[0321] A3. If the complete model side finds that the performance of the model on this side is abnormal, it sends a pairing configuration reporting indication to the opposite side, indicating that the opposite side reports the pairing configuration. The pairing configuration contains the model information currently used by the opposite side (such as the encoder information of the terminal device in the CSI compression feedback), such as the model ID.

[0322] A4. The complete model side receives the pairing configuration information reported by the opposite side, and resets the model on this side (i.e., the decoder in the CSI compression feedback) based on the information, so that the model on this side can match the model on the opposite side.

[0323] A5. Continue to monitor the performance of the model to see if the fault is eliminated. If the fault is eliminated, the fault detection process ends. If the fault is not eliminated, the complete model side reverts the model on this side to the model used before step A3, and continues the following model self-checking process on the opposite side.

[0324] The above process completes the first step of the fault detection process, i.e., self-checking of the complete model side. If the fault is still not eliminated, the model self-checking on the opposite side is performed.

[0325] As shown in FIG. 4e, it is an example of the implementation of the above-mentioned second step of the model self-checking process on the opposite side, which includes the following steps.

[0326] B1. The complete model side (such as the network device in the CSI compression feedback) is successfully self-checked, and the model on the complete model side is fault-free, or the fault is still not eliminated after the model is reset.

[0327] B2. The complete model side sends a fault detection indication to the opposite side, indicating that the opposite side starts fault detection.

[0328] B3. The opposite side inputs reference data into the model on this side to obtain the inference result of the model on this side, and sends it to the complete model side.

[0329] B4. The complete model side receives the inference result of the opposite side, continues to perform inference of the model on this side to obtain the final inference result, and compares it with the label of the reference data to calculate the correlation and similarity indicators. If the indicator is less than a preset threshold, it indicates that the performance of the model on the opposite side is abnormal; otherwise, it indicates that the performance of the model on the opposite side is normal.

[0330] B5. If the performance of the model on the opposite side is abnormal, the complete model side sends a model configuration indication to indicate that the opposite side performs model reconfiguration. The model configuration indication contains the model information of the complete model side, such as the model ID.

[0331] B6. Based on the model configuration indication, the model on this side is reconfigured so as to match the model on the complete model side.

[0332] B7. Continue to monitor the model performance, if the fault is eliminated, the fault detection process is ended, if the fault is not eliminated, the opposite side will rollback the model of the side to the model used before step B6, and continue the following process.

[0333] The above process completes the second step of the fault detection process, i.e. the self-check of the opposite side, if the fault is still not eliminated, it is considered that the current scene has changed, the data part has drifted, and then the bilateral model resetting process is performed.

[0334] As shown in FIG. 4f, it is an implementation example of the above-mentioned third step of the bilateral model resetting process, including the following steps.

[0335] C1. Both the complete model side and the opposite side complete the fault self-check, and both the bilateral models are fault-free.

[0336] C2. The complete model side judges that data drift has occurred, determines the data distribution of the current scene, and re-acquires the model suitable for the scene or data distribution.

[0337] C3. The complete model side performs the model reconfiguration of the side, and instructs the opposite side to perform the model reconfiguration.

[0338] C4. Continue to monitor the model performance, if the fault is eliminated, the fault detection process is ended, otherwise, it can be selected to rollback to the non-AI method.

[0339] The above implementation process of FIG. 4d to FIG. 4f introduces the fault detection process under the condition that only one side has a complete model. When both sides have complete models, the performance self-check of the model of the side can be performed independently. The following will be introduced in combination with the scene shown in FIG. 4g.

[0340] In the scene shown in FIG. 4g, the first communication device and the second communication device both have complete models, i.e. the first communication device can be the network device in FIG. 4b and the second communication device can be the terminal device in FIG. 4c, or, the first communication device can be the network device in FIG. 4c and the second communication device can be the terminal device in FIG. 4c, the fault detection process performed in this scene includes the steps shown in FIG. 4g.

[0341] D1. Both sides perform CSI compression feedback reconstruction, i.e. the terminal device performs CSI compression, feeds back the compressed CSI, and the network device performs CSI reconstruction. The network device performs model performance monitoring and finds performance abnormalities.

[0342] D2. The network device starts the fault detection process, including the model fault self-check this time, and instructs the opposite side to perform the model self-check. Since both sides have complete models.

[0343] D3. The terminal device reports the self-check result.

[0344] D4. The network device determines the fault cause according to the self-check result of the local side and the self-check result reported by the opposite side, and performs corresponding next fault elimination operation, as follows:

[0345] The network device model is faulty, the terminal device model is not faulty, and the decoder is configured.

[0346] The network device model is not faulty, the terminal device model is faulty, and the terminal device is instructed to configure the encoder.

[0347] Both sides are not faulty, and the two sides are instructed to configure the model.

[0348] In the implementation processes of FIGS. 4d-4f, the node with a complete model initiates the fault detection process, and first completes the fault self-check using the complete model; the complete model node successfully performs the fault self-check, and performs fault detection on other nodes (opposite side nodes) based on reference data; after the fault of the other nodes is eliminated, it is determined that the fault is caused by data drift, and model training and redeployment are performed. In the implementation process of FIG. 4g, when both sides have complete models, the network device initiates the fault detection process. Thus, the fault detection in the model system of both sides is implemented, the node with a complete model is used to gradually locate the fault position, so as to implement model management.

[0349] Referring to FIG. 5, an embodiment of the present application provides a communication apparatus 500, which can implement the functions of the first communication apparatus (or the second communication apparatus) in the above method embodiments, and thus can also implement the beneficial effects possessed by the above method embodiments. In the embodiment of the present application, the communication apparatus 500 can be the first communication apparatus (or the second communication apparatus), or an integrated circuit or element etc. inside the first communication apparatus (or the second communication apparatus), such as a chip, a baseband chip, a modem chip, an SoC chip (such as an SoC chip containing a modem core), a SIP chip, a communication module, a chip system, a processor, etc.

[0350] It should be noted that the transceiver unit 502 can include a sending unit and a receiving unit, which are respectively used for performing sending and receiving.

[0351] In a possible implementation, when the apparatus 500 is configured to perform the method performed by the first communication apparatus in FIG. 3 and related embodiments, the apparatus 500 includes a processing unit 501 and a transceiver unit 502. The processing unit 501 is configured to obtain first information, where the first information is used to determine whether a performance of a first model deployed on a first communication apparatus meets a first condition. The transceiver unit 502 is configured to receive second information, where the second information is used to determine whether a performance of a second model deployed on a second communication apparatus meets a second condition. The second model is used to jointly complete an AI task with the first model. The processing unit 501 is further configured to determine third information based on the first information and the second information, where the third information is used to indicate adjustment of the first model and / or adjustment of the second model.

[0352] In a possible implementation, when the apparatus 500 is configured to perform the method performed by the third communication apparatus in FIG. 3 and related embodiments, the apparatus 500 includes a processing unit 501 and a transceiver unit 502. The processing unit 501 is configured to determine second information, where the second information is used to determine whether a performance of a second model deployed on a second communication apparatus meets a second condition. The second model is used to jointly complete an AI task with a first model deployed on a first communication apparatus. The second information is used to determine third information, where the third information is used to indicate adjustment of the first model and / or adjustment of the second model. The transceiver unit 502 is configured to send the second information.

[0353] In a possible design, when the communication apparatus 500 is a terminal device or a communication module in a terminal, the function of the processing unit 501 can be implemented by one or more processors. Specifically, the processor can include a modem chip, a SoC chip (such as a SoC chip including a modem core), or a SIP chip. The function of the transceiver unit 502 can be implemented by a transceiver circuit.

[0354] In a possible design, when the communication apparatus 500 is a circuit or chip responsible for communication functions in a terminal, such as a modem chip or a SoC chip or a SoC chip including a modem core or a SIP chip, the function of the processing unit 501 can be implemented by a circuit system including one or more processors or processor cores in the chip. The function of the transceiver unit 502 can be implemented by an interface circuit or a data transceiver circuit on the chip.

[0355] It should be noted that the information execution process and the like of the units of the communication apparatus 500 described above can be specifically refer to the descriptions in the method embodiments described above, which will not be repeated here.

[0356] Please refer to FIG. 6, which is another schematic structural diagram of a communication apparatus 600 provided in the present application, the communication apparatus 600 includes a logic circuit 601 and an input-output interface 602. Wherein, the communication apparatus 600 can be a chip or an integrated circuit.

[0357] Wherein, the transceiver unit 502 shown in FIG. 5 can be a communication interface, which can be the input-output interface 602 in FIG. 6, and the input-output interface 602 can include an input interface and an output interface. Alternatively, the communication interface can also be a transceiver circuit, which can include an input interface circuit and an output interface circuit.

[0358] In a possible implementation, when the apparatus 600 is used to perform the method performed by the first communication apparatus in FIG. 3 and related embodiments, the logic circuit 601 is configured to obtain first information, the first information being used to determine whether the performance of a first model deployed in the first communication apparatus meets a first condition; the input-output interface 602 is configured to receive second information, the second information being used to determine whether the performance of a second model deployed in a second communication apparatus meets a second condition; wherein the second model and the first model are used to jointly complete an AI task; and the logic circuit 601 is further configured to determine third information based on the first information and the second information, the third information being used to indicate adjusting the first model and / or adjusting the second model.

[0359] In a possible implementation, when the apparatus 600 is used to perform the method performed by the second communication apparatus in FIG. 3 and related embodiments, the logic circuit 601 is configured to determine second information, the second information being used to determine whether the performance of a second model deployed in the second communication apparatus meets a second condition; wherein the second model and a first model deployed in a first communication apparatus are used to jointly complete an AI task, the second information is used to determine third information, the third information being used to indicate adjusting the first model and / or adjusting the second model; and the input-output interface 602 is configured to send the second information.

[0360] Wherein, the logic circuit 601 and the input-output interface 602 can also perform other steps performed by the first communication apparatus or the second communication apparatus in any of the embodiments and achieve the corresponding beneficial effects, which will not be described here.

[0361] In a possible implementation, the processing unit 501 shown in FIG. 5 can be the logic circuit 601 in FIG. 6.

[0362] Optionally, the logic circuit 601 can be a processing apparatus, and the functions of the processing apparatus can be partially or entirely implemented by software. Wherein, the functions of the processing apparatus can be partially or entirely implemented by software.

[0363] Optionally, the processing apparatus can include a memory and a processor, wherein the memory is configured to store a computer program, and the processor is configured to read and execute the computer program stored in the memory to perform the corresponding processing and / or steps in any one of the method embodiments.

[0364] Optionally, the processing apparatus can only include the processor. The memory for storing the computer program is located outside the processing apparatus, and the processor is connected with the memory through a circuit / wire to read and execute the computer program stored in the memory. The memory and the processor can be integrated together, or can be physically independent of each other.

[0365] Optionally, the processing apparatus can be one or more chips, or one or more integrated circuits. For example, the processing apparatus can be one or more field-programmable gate arrays (FPGA), application specific integrated circuits (ASIC), system on chips (SoC), central processing units (CPU), network processors (NP), digital signal processors (DSP), micro controller units (MCU), programmable logic devices (PLD) or other integrated chips, or any combination of the above chips or processors, etc.

[0366] Please refer to FIG. 7, which shows a communication apparatus 700 involved in the above embodiments provided by the embodiments of the present application. The communication apparatus 700 can be specifically a communication apparatus as a terminal device in the above embodiments, and the example shown in FIG. 7 is implemented by a terminal device (or a component in the terminal device).

[0367] Optionally, the communication apparatus 700 can include but is not limited to at least one processor 701 and a communication port 702.

[0368] Optionally, the transceiver unit 502 shown in FIG. 5 can be a communication interface, which can be the communication port 702 in FIG. 7. The communication port 702 can include an input interface and an output interface. Alternatively, the communication port 702 can also be a transceiver circuit, which can include an input interface circuit and an output interface circuit.

[0369] Further, the apparatus can further include at least one of a memory 703, and a bus 704, in an embodiment of the present application, the at least one processor 701 is configured to control processing of actions of the communication apparatus 700.

[0370] Further, the processor 701 can be a central processing unit, a general purpose processor, a digital signal processor, an application specific integrated circuit, a field programmable gate array, or other programmable logic device, transistor logic, hardware component, or any combination thereof. It can implement or execute various example logical blocks, modules, and circuits described in connection with the present disclosure. The processor can also be a combination of computing functionality, such as a combination of one or more microprocessors, a combination of a digital signal processor and a microprocessor, or the like. For the sake of brevity and conciseness, the specific processes performed by the system, apparatus, and units described above can be referred to the corresponding processes in the method embodiments described above, and will not be described here again.

[0371] It should be noted that the communication apparatus 700 shown in FIG. 7 can be specifically used to implement the steps implemented by the terminal device in the foregoing method embodiments, and achieve the corresponding technical effects of the terminal device. The specific implementation of the communication apparatus shown in FIG. 7 can be referred to the description in the foregoing method embodiments, and will not be described here again.

[0372] Please refer to FIG. 8, which is a structural schematic diagram of a communication apparatus 800 involved in the foregoing embodiments provided by an embodiment of the present application. The communication apparatus 800 can be specifically the communication apparatus as the network device in the foregoing embodiments, and the example shown in FIG. 8 is implemented by the network device (or components in the network device). The structure of the communication apparatus can refer to the structure shown in FIG. 8.

[0373] The communication apparatus 800 includes at least one processor 811 and at least one network interface 814. Further, the communication apparatus can further include at least one memory 812, at least one transceiver 813, and one or more antennas 815. The processor 811, the memory 812, the transceiver 813, and the network interface 814 are connected, for example, through a bus. In an embodiment of the present application, the connection can include various interfaces, transmission lines, or buses, etc., and the present embodiment is not limited thereto. The antenna 815 is connected to the transceiver 813. The network interface 814 is configured to enable the communication apparatus to communicate with other communication devices through a communication link. For example, the network interface 814 can include a network interface between the communication apparatus and a core network device, such as an S1 interface. The network interface can include a network interface between the communication apparatus and other communication apparatuses (such as other network devices or core network devices), such as an X2 or Xn interface.

[0374] The transceiving unit 502 shown in FIG. 5 can be a communication interface, which can be a network interface 814 in FIG. 8, and can include an input interface and an output interface. Alternatively, the network interface 814 can also be a transceiving circuit, which can include an input interface circuit and an output interface circuit.

[0375] The processor 811 is mainly used for processing communication protocols and communication data, and controlling the entire communication device, executing software programs, and processing data of the software programs, for example, for supporting the communication device to perform the actions described in the embodiments. The communication device can include a baseband processor and a central processor, the baseband processor is mainly used for processing communication protocols and communication data, and the central processor is mainly used for controlling the entire terminal device, executing software programs, and processing data of the software programs. The processor 811 in FIG. 8 can integrate the functions of the baseband processor and the central processor, and those skilled in the art can understand that the baseband processor and the central processor can also be independent processors interconnected by a bus or the like. Those skilled in the art can understand that the terminal device can include multiple baseband processors to adapt to different network modes, and the terminal device can include multiple central processors to enhance its processing capability, and various components of the terminal device can be connected by various buses. The baseband processor can also be referred to as a baseband processing circuit or a baseband processing chip. The central processor can also be referred to as a central processing circuit or a central processing chip. The function of processing communication protocols and communication data can be built into the processor, or stored in the memory in the form of a software program, and the processor executes the software program to realize the baseband processing function.

[0376] The memory is mainly used for storing software programs and data. The memory 812 can exist independently and be connected to the processor 811. Alternatively, the memory 812 can be integrated with the processor 811, for example, integrated in a chip. The memory 812 can store program codes for executing the technical solutions of the embodiments of the present application, and the processor 811 controls the execution. Various computer programs executed can also be regarded as a driver of the processor 811.

[0377] FIG. 8 only shows one memory and one processor. In actual terminal devices, there can be multiple processors and multiple memories. The memory can also be referred to as a storage medium or a storage device, etc. The memory can be a storage element on the same chip as the processor, that is, an on-chip storage element, or an independent storage element, and the embodiments of the present application do not limit this.

[0378] The transceiver 813 can be configured to support the receiving or transmitting of radio frequency signals between the communication device and a terminal. The transceiver 813 can be connected to the antenna 815. The transceiver 813 includes a transmitter Tx and a receiver Rx. Specifically, the one or more antennas 815 can receive radio frequency signals, and the receiver Rx of the transceiver 813 is configured to receive the radio frequency signals from the antenna and convert the radio frequency signals into digital baseband signals or digital intermediate frequency signals, and provide the digital baseband signals or digital intermediate frequency signals to the processor 811 for further processing, such as demodulation and decoding, by the processor 811. In addition, the transmitter Tx of the transceiver 813 is also configured to receive modulated digital baseband signals or digital intermediate frequency signals from the processor 811, and convert the modulated digital baseband signals or digital intermediate frequency signals into radio frequency signals, and transmit the radio frequency signals through the one or more antennas 815. Specifically, the receiver Rx can selectively perform one or more levels of down-mixing and analog-to-digital conversion to obtain the digital baseband signals or digital intermediate frequency signals, and the order of the down-mixing and analog-to-digital conversion can be adjustable. The transmitter Tx can selectively perform one or more levels of up-mixing and digital-to-analog conversion to obtain the radio frequency signals, and the order of the up-mixing and digital-to-analog conversion can be adjustable. The digital baseband signals and the digital intermediate frequency signals can be collectively referred to as digital signals.

[0379] The transceiver 813 can also be referred to as a transceiving unit, a transceiver, a transceiving device, etc. Optionally, the devices in the transceiving unit for implementing the receiving function can be regarded as a receiving unit, and the devices in the transceiving unit for implementing the transmitting function can be regarded as a transmitting unit, i.e., the transceiving unit includes the receiving unit and the transmitting unit. The receiving unit can also be referred to as a receiver, an input port, a receiving circuit, etc. The transmitting unit can be referred to as a transmitter, a transmitter, or a transmitting circuit, etc.

[0380] It should be noted that the communication device 800 shown in FIG. 8 can be specifically configured to implement the steps implemented by the network device in the foregoing method embodiments, and achieve the corresponding technical effects of the network device. The specific implementation manner of the communication device 800 shown in FIG. 8 can be referred to the description in the foregoing method embodiments, which will not be described here one by one.

[0381] Please refer to FIG. 9, which is a structural schematic diagram of a communication device involved in the foregoing embodiments provided by the embodiments of the present application.

[0382] It can be understood that the communication apparatus 900 includes, for example, modules, units, elements, circuits, or interfaces, and the like, which are appropriately configured together to perform the technical solutions provided in the present application. The communication apparatus 900 can be a terminal device or a network device described above, or can be a component (for example, a chip) of the devices, to implement the methods described in the following method embodiments. The communication apparatus 900 includes one or more processors 901. The processor 901 can be a general processor or a special-purpose processor, and the like. For example, it can be a baseband processor or a central processing unit. The baseband processor can be used to process communication protocols and communication data, and the central processing unit can be used to control the communication apparatus (such as a RAN node, a terminal, or a chip, and the like), execute software programs, and process data of the software programs.

[0383] Optionally, in one design, the processor 901 can include a program 903 (which can also be referred to as code or instructions at times) that can be run on the processor 901, so that the communication apparatus 900 performs the methods described in the following embodiments. In yet another possible design, the communication apparatus 900 includes a circuit (not shown in FIG. 9).

[0384] Optionally, the communication apparatus 900 can include one or more memories 902 having a program 904 (which can also be referred to as code or instructions at times) stored thereon, which can be run on the processor 901, so that the communication apparatus 900 performs the methods described in the above method embodiments.

[0385] Optionally, the processor 901 and / or the memory 902 can include an AI module 907, 908, which is used to implement AI-related functions. The AI module can be implemented in software, hardware, or a combination of software and hardware. For example, the AI module can include a radio intelligence control (RIC) module. For example, the AI module can be a near-real-time RIC or a non-real-time RIC.

[0386] Optionally, the processor 901 and / or the memory 902 can also store data. The processor and the memory can be separately arranged, or can be integrated together.

[0387] Optionally, the communication apparatus 900 can also include a transceiver 905 and / or an antenna 906. The processor 901 can also be referred to as a processing unit, which controls the communication apparatus (such as a RAN node or a terminal). The transceiver 905 can also be referred to as a transceiving unit, a transceiver, a transceiving circuit, or a transceiver, and the like, which is used to realize the transceiving function of the communication apparatus through the antenna 906.

[0388] The processing unit 501 shown in FIG. 5 can be the processor 901. The transceiving unit 502 shown in FIG. 5 can be a communication interface, which can be the transceiver 905 in FIG. 9, and the transceiver 905 can include an input interface and an output interface. Alternatively, the transceiver 905 can also be a transceiving circuit, which can include an input interface circuit and an output interface circuit.

[0389] The embodiments of the present application further provide a computer readable storage medium for storing one or more computer-executable instructions, which, when executed by a processor, cause the processor to perform the method described in the possible implementation manners of the first communication device or the second communication device.

[0390] The embodiments of the present application further provide a computer program product (or computer program), which, when executed by a processor, causes the processor to perform the method described in the possible implementation manners of the first communication device or the second communication device.

[0391] The embodiments of the present application further provide a chip system, which includes at least one processor for supporting the communication device to implement the functions involved in the possible implementation manners of the communication device. Optionally, the chip system further includes an interface circuit for providing program instructions and / or data for the at least one processor. In a possible design, the chip system can further include a memory for storing necessary program instructions and data of the communication device. The chip system can be composed of a chip, or can include a chip and other discrete devices. The communication device can be the first communication device or the second communication device in the method embodiments.

[0392] The embodiments of the present application further provide a communication system, which includes the first communication device and / or the second communication device in any of the above embodiments.

[0393] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the described device embodiments are merely schematic. For example, the division of the units is only a logical function division. There can be another division manner for the actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different units, can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0394] The units described as separated components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments of the present application.

[0395] In addition, each functional unit in the embodiments of the present application can be integrated in one processing unit, or each unit can exist physically as a separate unit, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware or in the form of a software functional unit. When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or substantially, or all or part of the technical solutions, can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application. The foregoing storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and various other media that can store program codes.

Claims

A communication method characterized by comprising: The method comprises: obtaining first information used to determine whether performance of a first model deployed on a first communication device meets a first condition; receiving second information used to determine whether performance of a second model deployed on a second communication device meets a second condition; wherein the second model and the first model are used to jointly complete an AI task; determining third information based on the first information and the second information, the third information being used to indicate adjustment of the first model and / or adjustment of the second model. The method of claim 1, wherein The first information is used to indicate performance of the first model using a first parameter for data processing; The first information is obtained in a case where performance of the first model using a second parameter for data processing is lower than a threshold value; wherein the first parameter is a parameter corresponding to the first model after a model adjustment operation is performed on the first model, and the second parameter is a parameter corresponding to the first model before the model adjustment operation is performed on the first model. The second information is obtained based on processing of a first processing result by the second model, the first processing result being obtained by processing of the first model using the second parameter. The method according to claim 2, characterized in that The second information is used to indicate performance of the second model using a third parameter for data processing, and the second information is obtained in a case where performance of the second model using a fourth parameter for data processing is lower than a threshold value; The method according to claim 2 or 3, characterized in that wherein the third parameter is a parameter corresponding to the second model after a model adjustment operation is performed on the second model, and the fourth parameter is a parameter corresponding to the second model before the model adjustment operation is performed on the second model. The method further comprises: The method according to any one of claims 2 to 4, characterized in that in a case where it is determined based on the first information that the performance of the first model meets the first condition, sending request information used to request the second information. The first information is obtained in a case where it is determined based on the second information that the performance of the second model meets the second condition. The method of claim 1, wherein The third information meets any one of the following conditions: in a case where it is determined based on the first information that the performance of the first model does not meet the first condition and it is determined based on the second information that the performance of the second model meets the second condition, the third information is used to indicate a model adjustment operation on the first model; The method according to any one of claims 1 to 6, characterized in that or, in a case where it is determined based on the first information that the performance of the first model meets the first condition and it is determined based on the second information that the performance of the second model does not meet the second condition, the third information is used to indicate a model adjustment operation on the second model; or, in a case where it is determined based on the first information that the performance of the first model does not meet the first condition and it is determined based on the second information that the performance of the second model does not meet the second condition, the third information is used to indicate a model adjustment operation on the first model and the second model. ​ ​ In a case where it is determined based on the first information that the performance of the first model meets the first condition and based on the second information that the performance of the second model meets the second condition, and the performance of the AI task is lower than a threshold, the third information is used to instruct a model adjustment operation on the first model and the second model. The method of claim 7, wherein In a case where the third information is used to instruct a model adjustment operation on the second model, the method further includes: sending the third information. The method of claim 7, wherein In a case where the third information is used to instruct a model adjustment operation on the first model and the second model, the method further includes: sending fourth information, the fourth information being used to instruct a model adjustment operation on the second model. The method according to any one of claims 1 to 9, characterized in that The first information includes at least one of: first indication information used to indicate output data of the first model; second indication information used to indicate a difference between the output data of the first model and an expected output; third indication information used to indicate a difference between a second processing result and a third processing result; the second processing result indicates a processing result obtained by processing, by the first model, output data of the second model, and the third processing result is used to indicate a processing result obtained by processing, by the first model, the second processing result; wherein the processing of the first model and the processing of the second model are partially or entirely the same; fourth indication information used to indicate whether the performance of the first model meets the first condition. The method according to any one of claims 1 to 10, characterized in that The second information includes at least one of: fifth indication information used to indicate output data of the second model; sixth indication information used to indicate a difference between the output data of the second model and an expected output; seventh indication information used to indicate a difference between a third processing result and a fourth processing result; the third processing result indicates a processing result obtained by processing, by the second model, output data of the first model, and the fourth processing result is used to indicate a processing result obtained by processing, by the second model, the third processing result; eighth indication information used to indicate whether the performance of the second model meets the first condition. A communication method characterized by comprising: includes: determining second information, the second information being used to determine whether the performance of a second model deployed on a second communication device meets a second condition; wherein the second model and a first model deployed on a first communication device are used to jointly complete an AI task, and the second information is used to determine third information, the third information being used to instruct adjustment of the first model and / or adjustment of the second model; sending the second information. The method of claim 12, wherein The second information and first information are used to determine the third information, wherein the first information is used to determine whether the performance of a first model deployed on a first communication device meets a first condition. The method of claim 13, wherein The first information is used to indicate the performance of the first model obtained by using a first parameter for data processing; The first information is obtained in a case where performance of the first model obtained by using a second parameter for data processing is lower than a threshold value; the first parameter is a parameter corresponding to the first model after a model adjustment operation is completed, and the second parameter is a parameter corresponding to the first model before the model adjustment operation is performed. The method of claim 14, wherein The second information is obtained based on processing of a first processing result by the second model, and the first processing result is obtained by processing of the first model by using the second parameter. The method according to claim 14 or 15, characterized in that The second information is used to indicate performance of the second model obtained by using a third parameter for data processing, and the second information is obtained in a case where performance of the second model obtained by using a fourth parameter for data processing is lower than a threshold value. The third parameter is a parameter corresponding to the second model after a model adjustment operation is performed, and the fourth parameter is a parameter corresponding to the second model before the model adjustment operation is performed. The method of claim 13, wherein The first information is obtained in a case where it is determined based on the second information that the performance of the second model satisfies the second condition. The method according to any one of claims 12 to 17, characterized in that The method further includes: receiving request information used for requesting the second information. The method according to any one of claims 12 to 18, characterized in that The third information satisfies any one of the following conditions: In a case where it is determined based on the first information that the performance of the first model does not satisfy the first condition and based on the second information that the performance of the second model satisfies the second condition, the third information is used to indicate that a model adjustment operation is performed on the first model. Or, In a case where it is determined based on the first information that the performance of the first model satisfies the first condition and based on the second information that the performance of the second model does not satisfy the second condition, the third information is used to indicate that a model adjustment operation is performed on the second model. Or, In a case where it is determined based on the first information that the performance of the first model does not satisfy the first condition and based on the second information that the performance of the second model does not satisfy the second condition, the third information is used to indicate that a model adjustment operation is performed on the first model and the second model. In a case where it is determined based on the first information that the performance of the first model satisfies the first condition and based on the second information that the performance of the second model satisfies the second condition, and the performance of the AI task is lower than a threshold value, the third information is used to indicate that a model adjustment operation is performed on the first model and the second model. The method of claim 19, wherein In a case where the third information is used to indicate that a model adjustment operation is performed on the second model, the method further includes: receiving the third information. The method of claim 19, wherein In a case where the third information is used to indicate that a model adjustment operation is performed on the first model and the second model, the method further includes: receiving fourth information, and the fourth information is used to indicate that a model adjustment operation is performed on the second model. The method according to any one of claims 13 to 21, characterized in that The first information includes at least one of the following: first indication information used to indicate output data of the first model; second indication information used to indicate a difference between the output data of the first model and expected output. third indication information, used for indicating a difference between a second processing result and a third processing result; the second processing result is used for indicating a processing result obtained by processing, by the first model, output data of the second model, and the third processing result is used for indicating a processing result obtained by processing, by the first model, the second processing result; wherein the processing of the first model and the processing of the second model are partially same or totally same; fourth indication information, used for indicating whether a performance of the first model satisfies the first condition. The method according to any one of claims 12 to 22, characterized in that The second information comprises at least one of: fifth indication information, used for indicating the output data of the second model; sixth indication information, used for indicating a difference between the output data of the second model and an expected output; seventh indication information, used for indicating a difference between a third processing result and a fourth processing result; the third processing result is used for indicating a processing result obtained by processing, by the second model, output data of the first model, and the fourth processing result is used for indicating a processing result obtained by processing, by the second model, the third processing result; eighth indication information, used for indicating whether a performance of the second model satisfies the first condition. A communication device, characterized by A module for performing the method according to any one of claims 1 to 23. A communication device, characterized by At least one processor for performing the method according to any one of claims 1 to 23. The communication apparatus according to claim 25, characterized in that, The communication device is a chip or a chip system. A computer-readable storage medium, characterized by The computer readable storage medium stores a computer program or instructions, which, when executed by the communication device, implement the method according to any one of claims 1 to 23. A computer program product, characterized in that A computer program or instructions, which, when executed by a computer, implement the method according to any one of claims 1 to 23. A computer program or instructions, which, when executed by a computer, implement the method according to any one of claims 1 to 23.

Images