Information processing method, device and equipment

Abstract

The invention provides an information processing method, device and equipment, and the information processing method comprises the steps: obtaining the model monitoring precision according to at least two test sample points; monitoring event related information aiming at the model monitoring precision is determined; the monitoring event related information comprises a definition of a monitoring event and an entry condition and / or an exit condition of the monitoring event; according to the monitoring precision of the model and the related information of the monitoring event, controlling the model to monitor the reported triggering operation; wherein the first device is a network device or a terminal. According to the scheme, the control of model monitoring report triggering can be realized according to the model monitoring precision (namely the monitoring result), so that decision operations such as switching, rollback, updating, activating and / or deactivating of the model can be accurately realized; the problem that in the prior art, an information processing scheme for model monitoring cannot support accurate implementation of model decision operation is well solved.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to an information processing method, device and equipment. Background Art

[0002] In the prior art, in order to ensure the applicability of the model, a solution for monitoring the performance of the model is proposed. However, it does not involve the use of the monitoring results (such as reporting), resulting in the inability to accurately support decision-making operations such as switching, rolling back, updating, activating and / or deactivating the model.

[0003] As described above, the information processing solutions for model monitoring in the prior art have problems such as being unable to support accurate implementation of model decision operations. Summary of the invention

[0004] The purpose of this application is to provide an information processing method, device and equipment to solve the problem that the information processing scheme for model monitoring in the prior art cannot support the accurate implementation of model decision operations.

[0005] In order to solve the above technical problems, an embodiment of the present application provides an information processing method, which is applied to a first device, including:

[0006] Acquire a model monitoring accuracy according to at least two test sample points; the test sample points are determined according to target channel state information CSI and CSI corresponding to the target CSI and recovered by the model; the model monitoring accuracy is based on statistical results of the at least two test sample points;

[0007] Determine monitoring event related information for the monitoring accuracy of the model; the monitoring event related information includes: a definition of the monitoring event, and an entry condition and / or an exit condition of the monitoring event;

[0008] Controlling the triggering operation of model monitoring reporting according to the model monitoring accuracy and monitoring event related information;

[0009] The first device is a network device or a terminal.

[0010] Optionally, the triggering operation of controlling the model monitoring report according to the model monitoring accuracy and the monitoring event related information includes:

[0011] When the monitoring event is that the model monitoring accuracy does not meet the use condition, and the model monitoring accuracy meets the entry condition of the monitoring event, triggering the operation of model monitoring reporting;

[0012] Alternatively, when the monitoring event is that the model monitoring accuracy meets the use condition, and the model monitoring accuracy meets the exit condition of the monitoring event, the model monitoring reporting operation is triggered.

[0013] Optionally, also include:

[0014] Determine monitoring and reporting conditions; the monitoring and reporting conditions include: reporting threshold and / or reporting timing duration;

[0015] The operation of triggering model monitoring reporting when the monitoring event is that the model monitoring accuracy does not meet the use condition and the model monitoring accuracy meets the entry condition of the monitoring event includes:

[0016] When the monitoring event is that the model monitoring accuracy does not meet the usage conditions, and the first number of times that the model monitoring accuracy meets the entry conditions of the monitoring event within the reporting timing duration is greater than the reporting threshold, the model monitoring reporting operation is triggered; wherein the content of the report is used to indicate that the currently used model is invalid; the first number of times is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event, or the first number of times is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event and the real-time number of times that the model monitoring accuracy meets the exit conditions of the monitoring event, or the first number of times refers to the continuous number of times that the model monitoring accuracy meets the entry conditions of the monitoring event;

[0017] And / or, when the monitoring event is that the model monitoring accuracy meets the use condition, and the model monitoring accuracy meets the exit condition of the monitoring event, the operation of triggering the model monitoring reporting includes:

[0018] When the monitoring event is that the model monitoring accuracy meets the usage conditions, and the second number of times that the model monitoring accuracy meets the exit conditions of the monitoring event within the reporting timing period is greater than the reporting threshold, the model monitoring reporting operation is triggered; wherein the reported content is used to indicate that the currently used model is invalid; the second number of times is determined based on the real-time number of times the model monitoring accuracy meets the exit conditions of the monitoring event, or the second number of times is determined based on the real-time number of times the model monitoring accuracy meets the exit conditions of the monitoring event and the real-time number of times the model monitoring accuracy meets the entry conditions of the monitoring event, or the second number of times refers to the consecutive number of times the model monitoring accuracy meets the exit conditions of the monitoring event.

[0019] Optionally, the model includes a currently used model and at least one candidate model;

[0020] The obtaining of the model monitoring accuracy according to at least two test sample points includes:

[0021] Acquire a first model monitoring accuracy for the currently used model according to at least two test sample points corresponding to the currently used model;

[0022] According to at least two test sample points corresponding to the candidate model, a second model monitoring accuracy for the candidate model is obtained.

[0023] Optionally, the monitoring event includes: a first monitoring event for the currently used model, and a second monitoring event for the candidate model;

[0024] The reporting threshold includes: a first reporting threshold for the currently used model, and a second reporting threshold for the candidate model;

[0025] The reporting timing duration includes: a first reporting timing duration for the currently used model, and a second reporting timing duration for the candidate model.

[0026] Optionally, when the monitoring event is that the model monitoring accuracy does not meet the use condition, and the model monitoring accuracy meets the entry condition of the monitoring event, the operation of triggering the model monitoring reporting includes:

[0027] When the first monitoring event is that the monitoring accuracy of the first model does not meet the use condition and meets the first condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model meets the use condition and meets the second condition, trigger the operation of model monitoring reporting; the content of the report is used to indicate model switching and carries the information of the candidate model;

[0028] Alternatively, when the first monitoring event is that the monitoring accuracy of the first model does not meet the use condition and meets the first condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model does not meet the use condition and meets the third condition, the operation of model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model;

[0029] Among them, the first condition includes: the third number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event within the first reporting timing duration is greater than the first reporting threshold; the third number is determined according to the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event, or the third number is determined according to the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event and the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event, or the third number refers to the consecutive number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event;

[0030] The second condition includes: the fourth number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the fourth number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event, or the fourth number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event, or the fourth number refers to the continuous number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event;

[0031] The third condition includes: the fifth number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event within the second reporting timing period is greater than the second reporting threshold; the fifth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event, or the fifth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event and the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event, or the fifth number refers to the consecutive number of times the second model monitoring accuracy meets the exit condition of the second monitoring event.

[0032] Optionally, when the monitoring event is that the model monitoring accuracy meets the use condition, and the model monitoring accuracy meets the exit condition of the monitoring event, triggering the model monitoring reporting operation includes:

[0033] When the first monitoring event is that the monitoring accuracy of the first model meets the use condition and the fourth condition is met; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model meets the use condition and the fifth condition is met, the operation of model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model;

[0034] Alternatively, when the first monitoring event is that the monitoring accuracy of the first model meets the use conditions and meets the fourth condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model does not meet the use conditions and meets the sixth condition, the operation of model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model;

[0035] Among them, the fourth condition includes: the sixth number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event within the first reporting timing duration is greater than the first reporting threshold; the sixth number is determined according to the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event, or the sixth number is determined according to the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event and the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event, or the sixth number refers to the consecutive number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event;

[0036] The fifth condition includes: the seventh number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the seventh number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event, or the seventh number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event, or the seventh number refers to the consecutive number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event;

[0037] The sixth condition includes: the eighth number of times the second model monitoring accuracy meets the exit condition of the second monitoring event within the second reporting timing period is greater than the second reporting threshold; the eighth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event, or the eighth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event and the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event, or the eighth number refers to the consecutive number of times the second model monitoring accuracy meets the exit condition of the second monitoring event.

[0038] Optionally, the triggering operation of controlling the model monitoring report according to the model monitoring accuracy and the monitoring event related information further includes:

[0039] When the monitoring event is that the model monitoring accuracy of the currently used model is less than the model monitoring accuracy of any candidate model, triggering the operation of model monitoring reporting;

[0040] The reported content is used to instruct model switching and carries information about the candidate model.

[0041] Optionally, also include:

[0042] A time domain sliding window of a first length is used to acquire the at least two test sample points.

[0043] Optionally, before obtaining the model monitoring accuracy according to at least two test sample points, the method further includes:

[0044] Preprocessing the at least two test sample points;

[0045] The preprocessing includes at least one of weighted processing and abnormal elimination processing.

[0046] Optionally, the entry condition and / or exit condition is determined based on a blocking parameter and a corresponding threshold value.

[0047] Optionally, the monitoring event related information is predefined according to a protocol and / or determined according to a network side configuration;

[0048] and / or, the monitoring reporting condition is predefined according to the protocol and / or determined by the network side configuration;

[0049] and / or, the first length is predefined according to a protocol and / or determined according to a network side configuration;

[0050] And / or, the sliding step size of the time domain sliding window is determined according to protocol pre-definition and / or network side configuration;

[0051] and / or, the blocking parameter is determined according to protocol pre-definition and / or network side configuration;

[0052] And / or, the threshold value is predefined according to the protocol and / or determined according to the network side configuration.

[0053] The embodiment of the present application further provides an information processing method, which is applied to a first device, including:

[0054] Using a time domain sliding window of a first length to obtain at least two test sample points;

[0055] The test sample point is determined according to a target CSI and a CSI corresponding to the target CSI and restored by a model;

[0056] The first device is a network device or a terminal.

[0057] Optionally, also include:

[0058] Preprocessing the at least two test sample points;

[0059] The preprocessing includes at least one of weighted processing and abnormal elimination processing.

[0060] Optionally, the first length is predefined according to a protocol and / or determined according to a network side configuration;

[0061] And / or, the sliding step size of the time domain sliding window is predefined according to the protocol and / or determined according to the network side configuration.

[0062] The embodiment of the present application further provides an information processing device, which is a first device and includes a memory, a transceiver, and a processor:

[0063] A memory for storing a computer program; a transceiver for transmitting and receiving data under the control of the processor; and a processor for reading the computer program in the memory and performing the following operations:

[0064] Acquire a model monitoring accuracy according to at least two test sample points; the test sample points are determined according to target channel state information CSI and CSI corresponding to the target CSI and recovered by the model; the model monitoring accuracy is based on statistical results of the at least two test sample points;

[0065] Determine monitoring event related information for the monitoring accuracy of the model; the monitoring event related information includes: a definition of the monitoring event, and an entry condition and / or an exit condition of the monitoring event;

[0066] Controlling the triggering operation of model monitoring reporting according to the model monitoring accuracy and monitoring event related information;

[0067] The first device is a network device or a terminal.

[0068] Optionally, the triggering operation of controlling the model monitoring report according to the model monitoring accuracy and the monitoring event related information includes:

[0069] When the monitoring event is that the model monitoring accuracy does not meet the use condition, and the model monitoring accuracy meets the entry condition of the monitoring event, triggering the operation of model monitoring reporting;

[0070] Alternatively, when the monitoring event is that the model monitoring accuracy meets the use condition, and the model monitoring accuracy meets the exit condition of the monitoring event, the model monitoring reporting operation is triggered.

[0071] Optionally, the operation further includes:

[0072] Determine monitoring and reporting conditions; the monitoring and reporting conditions include: reporting threshold and / or reporting timing duration;

[0073] The operation of triggering model monitoring reporting when the monitoring event is that the model monitoring accuracy does not meet the use condition and the model monitoring accuracy meets the entry condition of the monitoring event includes:

[0074] When the monitoring event is that the model monitoring accuracy does not meet the usage conditions, and the first number of times that the model monitoring accuracy meets the entry conditions of the monitoring event within the reporting timing duration is greater than the reporting threshold, the model monitoring reporting operation is triggered; wherein the content of the report is used to indicate that the currently used model is invalid; the first number of times is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event, or the first number of times is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event and the real-time number of times that the model monitoring accuracy meets the exit conditions of the monitoring event, or the first number of times refers to the continuous number of times that the model monitoring accuracy meets the entry conditions of the monitoring event;

[0075] And / or, when the monitoring event is that the model monitoring accuracy meets the use condition, and the model monitoring accuracy meets the exit condition of the monitoring event, the operation of triggering the model monitoring reporting includes:

[0076] When the monitoring event is that the model monitoring accuracy meets the usage conditions, and the second number of times that the model monitoring accuracy meets the exit conditions of the monitoring event within the reporting timing period is greater than the reporting threshold, the model monitoring reporting operation is triggered; wherein the reported content is used to indicate that the currently used model is invalid; the second number of times is determined based on the real-time number of times the model monitoring accuracy meets the exit conditions of the monitoring event, or the second number of times is determined based on the real-time number of times the model monitoring accuracy meets the exit conditions of the monitoring event and the real-time number of times the model monitoring accuracy meets the entry conditions of the monitoring event, or the second number of times refers to the consecutive number of times the model monitoring accuracy meets the exit conditions of the monitoring event.

[0077] Optionally, the model includes a currently used model and at least one candidate model;

[0078] The obtaining of the model monitoring accuracy according to at least two test sample points includes:

[0079] Acquire a first model monitoring accuracy for the currently used model according to at least two test sample points corresponding to the currently used model;

[0080] According to at least two test sample points corresponding to the candidate model, a second model monitoring accuracy for the candidate model is obtained.

[0081] Optionally, the monitoring event includes: a first monitoring event for the currently used model, and a second monitoring event for the candidate model;

[0082] The reporting threshold includes: a first reporting threshold for the currently used model, and a second reporting threshold for the candidate model;

[0083] The reporting timing duration includes: a first reporting timing duration for the currently used model, and a second reporting timing duration for the candidate model.

[0084] Optionally, when the monitoring event is that the model monitoring accuracy does not meet the use condition, and the model monitoring accuracy meets the entry condition of the monitoring event, the operation of triggering the model monitoring reporting includes:

[0085] When the first monitoring event is that the monitoring accuracy of the first model does not meet the use condition and meets the first condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model meets the use condition and meets the second condition, trigger the operation of model monitoring reporting; the content of the report is used to indicate model switching and carries the information of the candidate model;

[0086] Alternatively, when the first monitoring event is that the monitoring accuracy of the first model does not meet the use condition and meets the first condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model does not meet the use condition and meets the third condition, the operation of model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model;

[0087] Among them, the first condition includes: the third number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event within the first reporting timing duration is greater than the first reporting threshold; the third number is determined according to the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event, or the third number is determined according to the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event and the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event, or the third number refers to the consecutive number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event;

[0088] The second condition includes: the fourth number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the fourth number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event, or the fourth number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event, or the fourth number refers to the continuous number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event;

[0089] The third condition includes: the fifth number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event within the second reporting timing period is greater than the second reporting threshold; the fifth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event, or the fifth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event and the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event, or the fifth number refers to the consecutive number of times the second model monitoring accuracy meets the exit condition of the second monitoring event.

[0090] Optionally, when the monitoring event is that the model monitoring accuracy meets the use condition, and the model monitoring accuracy meets the exit condition of the monitoring event, triggering the model monitoring reporting operation includes:

[0091] When the first monitoring event is that the monitoring accuracy of the first model meets the use condition and the fourth condition is met; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model meets the use condition and the fifth condition is met, the operation of model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model;

[0092] Alternatively, when the first monitoring event is that the monitoring accuracy of the first model meets the use conditions and meets the fourth condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model does not meet the use conditions and meets the sixth condition, the operation of model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model;

[0093] Among them, the fourth condition includes: the sixth number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event within the first reporting timing duration is greater than the first reporting threshold; the sixth number is determined according to the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event, or the sixth number is determined according to the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event and the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event, or the sixth number refers to the consecutive number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event;

[0094] The fifth condition includes: the seventh number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the seventh number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event, or the seventh number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event, or the seventh number refers to the consecutive number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event;

[0095] The sixth condition includes: the eighth number of times the second model monitoring accuracy meets the exit condition of the second monitoring event within the second reporting timing period is greater than the second reporting threshold; the eighth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event, or the eighth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event and the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event, or the eighth number refers to the consecutive number of times the second model monitoring accuracy meets the exit condition of the second monitoring event.

[0096] Optionally, the triggering operation of controlling the model monitoring report according to the model monitoring accuracy and the monitoring event related information further includes:

[0097] When the monitoring event is that the model monitoring accuracy of the currently used model is less than the model monitoring accuracy of any candidate model, triggering the operation of model monitoring reporting;

[0098] The reported content is used to instruct model switching and carries information about the candidate model.

[0099] Optionally, the operation further includes:

[0100] A time domain sliding window of a first length is used to acquire the at least two test sample points.

[0101] Optionally, the operation further includes:

[0102] Before obtaining the model monitoring accuracy according to the at least two test sample points, preprocessing the at least two test sample points;

[0103] The preprocessing includes at least one of weighted processing and abnormal elimination processing.

[0104] Optionally, the entry condition and / or exit condition is determined based on a blocking parameter and a corresponding threshold value.

[0105] Optionally, the monitoring event related information is predefined according to a protocol and / or determined according to a network side configuration;

[0106] and / or, the monitoring reporting condition is predefined according to the protocol and / or determined by the network side configuration;

[0107] and / or, the first length is predefined according to a protocol and / or determined according to a network side configuration;

[0108] And / or, the sliding step size of the time domain sliding window is determined according to protocol pre-definition and / or network side configuration;

[0109] and / or, the blocking parameter is determined according to protocol pre-definition and / or network side configuration;

[0110] And / or, the threshold value is predefined according to the protocol and / or determined according to the network side configuration.

[0111] The embodiment of the present application further provides an information processing device, which is a first device and includes a memory, a transceiver, and a processor:

[0112] A memory for storing a computer program; a transceiver for transmitting and receiving data under the control of the processor; and a processor for reading the computer program in the memory and performing the following operations:

[0113] Using a time domain sliding window of a first length to obtain at least two test sample points;

[0114] The test sample point is determined according to a target CSI and a CSI corresponding to the target CSI and restored by a model;

[0115] The first device is a network device or a terminal.

[0116] Optionally, the operation further includes:

[0117] Preprocessing the at least two test sample points;

[0118] The preprocessing includes at least one of weighted processing and abnormal elimination processing.

[0119] Optionally, the first length is predefined according to a protocol and / or determined according to a network side configuration;

[0120] And / or, the sliding step size of the time domain sliding window is predefined according to the protocol and / or determined according to the network side configuration.

[0121] The embodiment of the present application further provides an information processing apparatus, applied to a first device, including:

[0122] A first acquisition unit is used to acquire a model monitoring accuracy according to at least two test sample points; the test sample points are determined according to target channel state information CSI and CSI corresponding to the target CSI and obtained by model recovery; the model monitoring accuracy is based on statistical results of the at least two test sample points;

[0123] A first determining unit is used to determine monitoring event related information for the model monitoring accuracy; the monitoring event related information includes: a definition of the monitoring event, and an entry condition and / or an exit condition of the monitoring event;

[0124] A first control unit, configured to control a triggering operation of a model monitoring report according to the model monitoring accuracy and monitoring event related information;

[0125] The first device is a network device or a terminal.

[0126] Optionally, the triggering operation of controlling the model monitoring report according to the model monitoring accuracy and the monitoring event related information includes:

[0127] When the monitoring event is that the model monitoring accuracy does not meet the use condition, and the model monitoring accuracy meets the entry condition of the monitoring event, triggering the operation of model monitoring reporting;

[0128] Alternatively, when the monitoring event is that the model monitoring accuracy meets the use condition, and the model monitoring accuracy meets the exit condition of the monitoring event, the model monitoring reporting operation is triggered.

[0129] Optionally, also include:

[0130] A second determining unit is used to determine monitoring reporting conditions; the monitoring reporting conditions include: reporting threshold and / or reporting timing duration;

[0131] The operation of triggering model monitoring reporting when the monitoring event is that the model monitoring accuracy does not meet the use condition and the model monitoring accuracy meets the entry condition of the monitoring event includes:

[0132] When the monitoring event is that the model monitoring accuracy does not meet the usage conditions, and the first number of times that the model monitoring accuracy meets the entry conditions of the monitoring event within the reporting timing duration is greater than the reporting threshold, the model monitoring reporting operation is triggered; wherein the content of the report is used to indicate that the currently used model is invalid; the first number of times is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event, or the first number of times is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event and the real-time number of times that the model monitoring accuracy meets the exit conditions of the monitoring event, or the first number of times refers to the continuous number of times that the model monitoring accuracy meets the entry conditions of the monitoring event;

[0133] And / or, when the monitoring event is that the model monitoring accuracy meets the use condition, and the model monitoring accuracy meets the exit condition of the monitoring event, the operation of triggering the model monitoring reporting includes:

[0134] When the monitoring event is that the model monitoring accuracy meets the usage conditions, and the second number of times that the model monitoring accuracy meets the exit conditions of the monitoring event within the reporting timing period is greater than the reporting threshold, the model monitoring reporting operation is triggered; wherein the reported content is used to indicate that the currently used model is invalid; the second number of times is determined based on the real-time number of times the model monitoring accuracy meets the exit conditions of the monitoring event, or the second number of times is determined based on the real-time number of times the model monitoring accuracy meets the exit conditions of the monitoring event and the real-time number of times the model monitoring accuracy meets the entry conditions of the monitoring event, or the second number of times refers to the consecutive number of times the model monitoring accuracy meets the exit conditions of the monitoring event.

[0135] Optionally, the model includes a currently used model and at least one candidate model;

[0136] The obtaining of the model monitoring accuracy according to at least two test sample points includes:

[0137] Acquire a first model monitoring accuracy for the currently used model according to at least two test sample points corresponding to the currently used model;

[0138] According to at least two test sample points corresponding to the candidate model, a second model monitoring accuracy for the candidate model is obtained.

[0139] Optionally, the monitoring event includes: a first monitoring event for the currently used model, and a second monitoring event for the candidate model;

[0140] The reporting threshold includes: a first reporting threshold for the currently used model, and a second reporting threshold for the candidate model;

[0141] The reporting timing duration includes: a first reporting timing duration for the currently used model, and a second reporting timing duration for the candidate model.

[0142] Optionally, when the monitoring event is that the model monitoring accuracy does not meet the use condition, and the model monitoring accuracy meets the entry condition of the monitoring event, the operation of triggering the model monitoring reporting includes:

[0143] When the first monitoring event is that the monitoring accuracy of the first model does not meet the use condition and meets the first condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model meets the use condition and meets the second condition, trigger the operation of model monitoring reporting; the content of the report is used to indicate model switching and carries the information of the candidate model;

[0144] Alternatively, when the first monitoring event is that the monitoring accuracy of the first model does not meet the use condition and meets the first condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model does not meet the use condition and meets the third condition, the operation of model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model;

[0145] Among them, the first condition includes: the third number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event within the first reporting timing duration is greater than the first reporting threshold; the third number is determined according to the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event, or the third number is determined according to the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event and the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event, or the third number refers to the consecutive number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event;

[0146] The second condition includes: the fourth number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the fourth number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event, or the fourth number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event, or the fourth number refers to the continuous number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event;

[0147] The third condition includes: the fifth number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event within the second reporting timing period is greater than the second reporting threshold; the fifth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event, or the fifth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event and the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event, or the fifth number refers to the consecutive number of times the second model monitoring accuracy meets the exit condition of the second monitoring event.

[0148] Optionally, when the monitoring event is that the model monitoring accuracy meets the use condition, and the model monitoring accuracy meets the exit condition of the monitoring event, triggering the model monitoring reporting operation includes:

[0149] When the first monitoring event is that the monitoring accuracy of the first model meets the use condition and the fourth condition is met; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model meets the use condition and the fifth condition is met, the operation of model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model;

[0150] Alternatively, when the first monitoring event is that the monitoring accuracy of the first model meets the use conditions and meets the fourth condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model does not meet the use conditions and meets the sixth condition, the operation of model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model;

[0151] Among them, the fourth condition includes: the sixth number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event within the first reporting timing duration is greater than the first reporting threshold; the sixth number is determined according to the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event, or the sixth number is determined according to the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event and the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event, or the sixth number refers to the consecutive number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event;

[0152] The fifth condition includes: the seventh number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the seventh number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event, or the seventh number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event, or the seventh number refers to the consecutive number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event;

[0153] The sixth condition includes: the eighth number of times the second model monitoring accuracy meets the exit condition of the second monitoring event within the second reporting timing period is greater than the second reporting threshold; the eighth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event, or the eighth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event and the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event, or the eighth number refers to the consecutive number of times the second model monitoring accuracy meets the exit condition of the second monitoring event.

[0154] Optionally, the triggering operation of controlling the model monitoring report according to the model monitoring accuracy and the monitoring event related information further includes:

[0155] When the monitoring event is that the model monitoring accuracy of the currently used model is less than the model monitoring accuracy of any candidate model, triggering the operation of model monitoring reporting;

[0156] The reported content is used to instruct model switching and carries information about the candidate model.

[0157] Optionally, also include:

[0158] The second acquisition unit is used to acquire the at least two test sample points by adopting a time domain sliding window of a first length.

[0159] Optionally, also include:

[0160] A first processing unit, configured to pre-process the at least two test sample points before obtaining the model monitoring accuracy according to the at least two test sample points;

[0161] The preprocessing includes at least one of weighted processing and abnormal elimination processing.

[0162] Optionally, the entry condition and / or exit condition is determined based on a blocking parameter and a corresponding threshold value.

[0163] Optionally, the monitoring event related information is predefined according to a protocol and / or determined according to a network side configuration;

[0164] and / or, the monitoring reporting condition is predefined according to the protocol and / or determined by the network side configuration;

[0165] and / or, the first length is predefined according to a protocol and / or determined according to a network side configuration;

[0166] And / or, the sliding step size of the time domain sliding window is determined according to protocol pre-definition and / or network side configuration;

[0167] and / or, the blocking parameter is determined according to protocol pre-definition and / or network side configuration;

[0168] And / or, the threshold value is predefined according to the protocol and / or determined according to the network side configuration.

[0169] The embodiment of the present application further provides an information processing apparatus, applied to a first device, including:

[0170] A third acquisition unit, configured to acquire at least two test sample points by using a time domain sliding window of a first length;

[0171] The test sample point is determined according to a target CSI and a CSI corresponding to the target CSI and restored by a model;

[0172] The first device is a network device or a terminal.

[0173] Optionally, also include:

[0174] A second processing unit, configured to pre-process the at least two test sample points;

[0175] The preprocessing includes at least one of weighted processing and abnormal elimination processing.

[0176] Optionally, the first length is predefined according to a protocol and / or determined according to a network side configuration;

[0177] And / or, the sliding step size of the time domain sliding window is predefined according to the protocol and / or determined according to the network side configuration.

[0178] An embodiment of the present application further provides a non-transitory readable storage medium, wherein the non-transitory readable storage medium stores a computer program, and the computer program is used to enable a processor to execute the above method.

[0179] The beneficial effects of the above technical solution of the present application are as follows:

[0180] In the above scheme, the information processing method obtains the model monitoring accuracy based on at least two test sample points; the test sample points are determined based on the target channel state information CSI and the CSI corresponding to the target CSI and obtained by model recovery; the model monitoring accuracy is based on the statistical results of the at least two test sample points; the monitoring event related information for the model monitoring accuracy is determined; the monitoring event related information includes: the definition of the monitoring event, and the entry condition and / or exit condition of the monitoring event; according to the model monitoring accuracy and the monitoring event related information, the triggering operation of the model monitoring report is controlled; wherein the first device is a network device or a terminal; it can support the control of the triggering of the model monitoring report according to the model monitoring accuracy (i.e., the monitoring result), so as to support the accurate implementation of the decision-making operations such as switching, fallback, updating, activation and / or deactivation of the model, which well solves the problem that the information processing scheme for model monitoring in the prior art cannot support the accurate implementation of the model decision-making operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0181] Figure 1 A schematic diagram of the wireless communication system architecture of an embodiment of the present application;

[0182] Figure 2 A schematic diagram of the network side model monitoring process of an embodiment of the present application;

[0183] Figure 3 The UE side model monitoring process of the embodiment of the present application is shown as follows: Figure 1 ;

[0184] Figure 4 The UE side model monitoring process of the embodiment of the present application is shown as follows: Figure 2 ;

[0185] Figure 5 The information processing method of the present application is shown in the following figure: Figure 1 ;

[0186] Figure 6 The information processing method of the present application is shown in the following figure: Figure 2 ;

[0187] Figure 7 A schematic diagram of the timing of obtaining test samples in an embodiment of the present application;

[0188] Figure 8 The structure of the information processing device of the embodiment of the present application is shown as follows: Figure 1 ;

[0189] Fig. 9 The structure of the information processing device of the embodiment of the present application is shown as follows: Figure 2 ;

[0190] Fig.10 The structure of the information processing device of the embodiment of the present application is shown as follows: Figure 1 ;

[0191] Fig.11 The structure of the information processing device of the embodiment of the present application is shown as follows: Figure 2 . DETAILED DESCRIPTION

[0192] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0193] In the embodiments of the present application, the term "and / or" describes the association relationship of the associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0194] In the embodiments of the present application, the term "plurality" refers to two or more than two, and other quantifiers are similar.

[0195] It is explained here that the technical solution provided in the embodiment of the present application can be applicable to a variety of systems, especially 5G systems. For example, the applicable system can be a global system of mobile communication (GSM) system, a code division multiple access (CDMA) system, a wideband code division multiple access (WCDMA) general packet radio service (GPRS) system, a long term evolution (LTE) system, a LTE frequency division duplex (FDD) system, a LTE time division duplex (TDD) system, an advanced long term evolution (LTE-A) system, a universal mobile telecommunication system (UMTS), a world-wide interoperability for microwave access (WiMAX) system, a 5G new air interface (NR) system, etc. These various systems include terminal equipment and network equipment. The system may also include a core network part, such as an evolved packet system (EPS), a 5G system (5GS), etc.

[0196] Figure 1 A block diagram of a wireless communication system applicable to the embodiments of the present application is shown. The wireless communication system includes a terminal device (also referred to as a terminal) and a network device.

[0197] The terminal device involved in the embodiment of the present application may be a device that provides voice and / or data connectivity to a user, a handheld device with a wireless connection function, or other processing devices connected to a wireless modem. In different systems, the name of the terminal device may also be different. For example, in a 5G system, the terminal device may be called a user equipment (UE). The wireless terminal device can communicate with one or more core networks (CN) via a radio access network (RAN). The wireless terminal device can be a mobile terminal device, such as a mobile phone (or a "cellular" phone) and a computer with a mobile terminal device. For example, it can be a portable, pocket-sized, handheld, computer-built-in or vehicle-mounted mobile device that exchanges language and / or data with a wireless access network. For example, personal communication service (PCS) phones, cordless phones, session initiation protocol (SIP) phones, wireless local loop (WLL) stations, personal digital assistants (PDA) and other devices. The wireless terminal device may also be referred to as a system, a subscriber unit, a subscriber station, a mobile station, a mobile station, a remote station, an access point, a remote terminal device, an access terminal device, a user terminal device, a user agent, and a user device, but is not limited in the embodiments of the present application.

[0198] The network device involved in the embodiments of the present application may be a base station, which may include multiple cells providing services for the terminal. Depending on the specific application scenario, the base station may also be called an access point, or may be a device in the access network that communicates with the wireless terminal device through one or more sectors on the air interface, or other names. The network device may be used to interchange received air frames with Internet Protocol (IP) packets, and serve as a router between the wireless terminal device and the rest of the access network, wherein the rest of the access network may include an Internet Protocol (IP) communication network. The network device may also coordinate the attribute management of the air interface. For example, the network device involved in the embodiments of the present application may be a network device (Base Transceiver Station, BTS) in the Global System for Mobile communications (Global System for Mobile communications, GSM) or Code Division Multiple Access (Code Division Multiple Access, CDMA), or a network device (NodeB) in Wide-band Code Division Multiple Access (WCDMA), or an evolutionary network device (evolutional Node B, eNB or e-NodeB) in the long-term evolution (long term evolution, LTE) system, a 5G base station (the next Generation Node B, gNB) in the 5G network architecture (next generation system), or a home evolved Node B (Home evolved Node B, HeNB), a relay node, a home base station (femto), a pico base station (pico), etc., which is not limited in the embodiments of the present application. In some network structures, the network device may include a centralized unit (CU) node and a distributed unit (DU) node, and the centralized unit and the distributed unit may also be arranged geographically separately.

[0199] Network devices and terminal devices can each use one or more antennas for multiple-input multiple-output (MIMO) transmission. MIMO transmission can be single-user MIMO (SU-MIMO) or multi-user MIMO (MU-MIMO). Depending on the form and number of antenna combinations, MIMO transmission can be 2D-MIMO, 3D-MIMO, FD-MIMO or massive-MIMO, or it can be diversity transmission, precoded transmission or beamforming transmission, etc.

[0200] The following first introduces the contents involved in the solution provided in the embodiment of the present application.

[0201] The monitoring of artificial intelligence (AI) / machine learning (ML) models (i.e. AI and / or ML models) is a new problem that arises after the introduction of AI / ML algorithms into the new radio (5G NR). Each AI / ML model has its scope of application, and AI / ML model monitoring means that when the AI / ML model usage scenario is different from its training scenario and the intrinsic relationship between input and output changes, the performance of the model needs to be monitored to determine whether the current model is applicable.

[0202] In actual systems, the performance of the AI / ML model monitoring the current model should not be judged based on a single result, because operations based on a single evaluation result may lead to frequent switching, rollback, updating, activation and / or deactivation of the AI / ML model, the so-called ping-pong effect.

[0203] In the current scheme, the conclusion proposed to improve the stability of Channel State Information (CSI) compression AI / ML model monitoring is: introduce a data set with K test sample points, and the AI / ML model monitoring accuracy is the statistical result of the performance evaluation on these K sample points. Based on the above conclusions, the following issues need to be addressed: (1) How to obtain these K test sample point data in the time series data, (2) How to trigger the AI / ML model monitoring report based on the AI / ML model monitoring accuracy indication obtained from the statistical results of the performance evaluation of these K sample points, so as to request the network side to make decisions on the activation, deactivation, update, rollback and / or switching of the AI / ML model. Furthermore, it is also necessary to solve the potential ping-pong effect problem of frequent changes in the AI / ML model.

[0204] In response to the above, the relevant content is introduced in detail below.

[0205] 1. For AI / ML-based CSI compression, the following AI / ML model monitoring metrics are summarized in the current solution:

[0206] 1. Intermediate KPI (Key performance indicator) as monitoring metric (e.g., Squared Generalized Cosine Similarity (SGCS)), i.e., monitoring based on inference accuracy;

[0207] 2. Eventual KPI as monitoring metric (e.g., throughput, hypothetical throughput, BLER (Block Error Rate), hypothetical BLER, NACK / ACK), i.e., monitoring based on system performance;

[0208] 3. Model monitoring based on traditional CSI: A solution based on additional traditional CSI reporting;

[0209] 4. Monitoring based on input and output data: for example, monitoring based on data drift between training data sets and observation data sets, monitoring based on output distribution, etc.

[0210] Second, the AI / ML model monitoring subject of CSI compression can be the terminal (i.e., model monitoring on the terminal (User Equipment, UE) side) or the network (i.e., model monitoring on the network side). In the current solution, for AI / ML model monitoring with Intermediate KPI as the monitoring metric, the following conclusions are reached:

[0211] 1. For network-side AI / ML model monitoring, it can be based on the target CSI reported by the UE side (referring to the input of the CSI generation part of the AI / ML model obtained on the UE side, also called ground truth CSI); among them, the CSI generation part of the AI / ML model can realize CSI compression, which may specifically include: CSI generation reasoning and quantization.

[0212] 2. For UE-side AI / ML model monitoring:

[0213] (1) The recovered CSI can be based on the network side feedback (referring to the output of the CSI recovery part of the AI / ML model obtained by the network side); wherein the CSI recovery part of the AI / ML model can realize CSI recovery, which may specifically include: dequantization and CSI recovery reasoning.

[0214] (2) It can be an alternative model based on the AI / ML model of the network-side CSI recovery part existing on the UE side;

[0215] 3. Other possibilities are not excluded, such as a model based on the UE side that directly outputs an intermediate KPI; or a target CSI (time division duplex (TDD) mode) measured by the network side based on the UE side channel sounding reference signal (SRS).

[0216] 3. Based on the above description, the network-side AI / ML model monitoring process based on the target CSI reported by the UE side can be as follows Figure 2 As shown, including:

[0217] The UE obtains the target CSI based on the received CSI-related reference signal. Then, the AI / ML model (CSI generation part) is used to obtain the CSI compressed by the AI / ML model. After that, the UE side feedbacks and reports the compressed CSI, and the UE side reports the target CSI.

[0218] The network inputs the received compressed CSI into the AI / ML model (CSI recovery part) to obtain the CSI recovered by the AI / ML model, and then performs network-side monitoring based on this and the target CSI reported by the UE side obtained by the network side: Calculate the Intermediate KPI based on the CSI recovered by the AI / ML model and the target CSI reported by the UE side obtained by the network side.

[0219] 4. Based on the above description, the UE-side AI / ML model monitoring process based on network-side backhaul recovery CSI can be as follows Figure 3 As shown, including:

[0220] The UE obtains the target CSI based on the received CSI-related reference signal. Then, the AI / ML model (CSI generation part) is used to obtain the CSI compressed by the AI / ML model. After that, the UE side feedbacks and reports the compressed CSI.

[0221] The network inputs the received compressed CSI into the AI / ML model (CSI recovery part) to obtain the CSI recovered by the AI / ML model, and then performs network-side backhaul to recover the CSI (i.e., the CSI recovered by the AI / ML model);

[0222] The UE calculates the IntermediateKPI according to the network-side backhaul recovery CSI obtained by the UE side and the target CSI, so as to implement UE-side monitoring.

[0223] 5. Based on the above description, the UE-side AI / ML model monitoring process based on the existence of an alternative model on the UE side (i.e., an alternative model based on the existence of the UE side and the network side CSI recovery part of the AI / ML model) can be as follows: Figure 4 As shown, including:

[0224] The UE obtains the target CSI based on the received CSI-related reference signal. Then, the AI / ML model (CSI generation part) is used to obtain the CSI compressed by the AI / ML model. After that, the UE side feedbacks and reports the compressed CSI.

[0225] The network inputs the received compressed CSI into the AI / ML model (CSI recovery part) to obtain the CSI recovered by the AI / ML model;

[0226] The UE inputs the CSI compressed by the AI / ML model into the AI / ML model (the alternative model of the CSI recovery part) to obtain the CSI recovered by the UE side using the alternative model; then the UE calculates the Intermediate KPI based on the target CSI and the CSI recovered by the UE side using the alternative model to achieve UE-side monitoring.

[0227] 6. In the current solution, the following conclusions are reached for the AI / ML model monitoring method under the CSI compression use case based on the intermediate KPI of model accuracy:

[0228] 1. For CSI compression, the following model monitoring method is used when conducting performance evaluation based on intermdediate KPI monitoring:

[0229] Step 1: Generate a test data set consisting of K test samples;

[0230] Step 2: For each of the K test points, the intermediate KPI deviation KPI Diff The following formula is used for calculation: KPI Diff =f(KPI Actual ,KPI Genie ), where KPI Actual For the actual intermediate KPI, KPI Genie It is an ideal intermediate KPI (genie-aided intermediate KPI);

[0231] Step 3: Calculate KPIs Diff The statistical results on K test points represent the monitoring accuracy performance.

[0232] Among them, (1) Introducing KPIGenie It is intended for evaluation and comparison purposes; it may not be available in an actual network.

[0233] (2) The complexity, cost, and latency of monitoring solutions are reported by each company.

[0234] The above conclusions introduce an AI / ML model monitoring test data set containing K test points. The statistical results of the K test points calculated on this data set can represent the performance of monitoring accuracy. Therefore, the impact of a single test data point on monitoring accuracy is limited, which can ensure the stability of the model monitoring performance evaluation results to a certain extent.

[0235] In the current solution, the method proposed to improve the stability of CSI compression AI / ML model monitoring is: introduce a data set with K test sample points, and the AI / ML model monitoring accuracy is the statistical result of the performance evaluation on these K sample points. In this method, the impact of a single test data point on the monitoring accuracy is limited, and the stability of the model monitoring performance evaluation results can be guaranteed to a certain extent. However, the current solution still has the following disadvantages:

[0236] (1) The current scheme does not have a specific implementation plan to describe how the dataset of K test sample points is obtained, and when the subject of model monitoring should calculate the monitoring accuracy based on the dataset of K test sample points.

[0237] (2) After calculating the monitoring accuracy based on the statistical results of the data set of K test samples, the current solution does not have a specific implementation plan that describes how to use the obtained monitoring accuracy to trigger AI / ML model monitoring reporting. The design of this trigger mechanism needs to be able to suppress the ping-pong effect of frequent changes in potential AI / ML models.

[0238] Based on the above, the embodiments of the present application provide an information processing method, device and equipment to solve the problem that the information processing scheme for model monitoring in the prior art cannot support the accurate implementation of model decision operations. Among them, the method, device and equipment are based on the same application concept. Since the principles of solving the problems by the method, device and equipment are similar, the implementation of the method, device and equipment can refer to each other, and the repeated parts will not be repeated.

[0239] The information processing method provided in the embodiment of the present application is applied to a first device, such as Figure 5 As shown, including:

[0240] Step 51: acquiring a model monitoring accuracy according to at least two test sample points; the test sample points are determined according to target channel state information CSI and CSI corresponding to the target CSI and recovered by the model; the model monitoring accuracy is based on statistical results of the at least two test sample points;

[0241] Step 52: Determine monitoring event related information for the monitoring accuracy of the model; the monitoring event related information includes: a definition of the monitoring event (indicating what the event is), and an entry condition and / or an exit condition of the monitoring event (indicating how to determine entering and / or exiting the monitoring event);

[0242] Step 53: Control the triggering operation of model monitoring reporting according to the model monitoring accuracy and monitoring event related information; wherein the first device is a network device or a terminal.

[0243] Among them, depending on whether the monitoring is done on the network side or the terminal side (i.e., the first device is a network device or a terminal), the target channel state information (i.e., the target CSI) can be various, for example: (1) obtained from a CSI-related reference signal (e.g., the above Figure 3 and 4 (2) obtained from the CSI-related reference signal and reported by the terminal (i.e., the terminal report will be quantized and / or compressed again to generate new errors, as mentioned above Figure 2 Of course, other possibilities may also be included, which are not limited here.

[0244] There are also multiple possibilities for obtaining the “CSI corresponding to the target CSI and recovered by the model”, for example: (1) the output of the CSI recovery part of the AI / ML model, as mentioned above Figure 2 ; (2) The output of the CSI recovery part of the AI / ML model is then transmitted back to the terminal via the network side, as mentioned above Figure 3 ; (3) The UE side uses the alternative model to recover, as mentioned above Figure 4 ; Of course, there may be other possibilities, which are not limited here.

[0245] The information processing method provided in the embodiment of the present application obtains the model monitoring accuracy based on at least two test sample points; the test sample points are determined based on the target channel state information CSI and the CSI corresponding to the target CSI and obtained by model recovery; the model monitoring accuracy is based on the statistical results of the at least two test sample points; the monitoring event related information for the model monitoring accuracy is determined; the monitoring event related information includes: the definition of the monitoring event, and the entry condition and / or exit condition of the monitoring event; according to the model monitoring accuracy and the monitoring event related information, the triggering operation of the model monitoring report is controlled; wherein the first device is a network device or a terminal; it can support the control of the triggering of the model monitoring report according to the model monitoring accuracy (i.e., the monitoring result), so as to support the accurate implementation of the decision-making operations such as switching, fallback, updating, activation and / or deactivation of the model, and well solves the problem that the information processing scheme for model monitoring in the prior art cannot support the accurate implementation of the model decision-making operation.

[0246] Among them, the triggering operation of controlling the model monitoring report based on the model monitoring accuracy and the information related to the monitoring event includes: triggering the model monitoring reporting operation when the monitoring event is that the model monitoring accuracy does not meet the usage conditions, and the model monitoring accuracy meets the entry conditions of the monitoring event; or triggering the model monitoring reporting operation when the monitoring event is that the model monitoring accuracy meets the usage conditions, and the model monitoring accuracy meets the exit conditions of the monitoring event.

[0247] In this way, the triggering operation of controlling the model monitoring report can be specifically implemented. In addition, a special case can be specifically implemented: as long as a certain monitoring event is entered or left, the model monitoring report can be triggered, but it is not limited to this.

[0248] Furthermore, the information processing method further includes: determining monitoring reporting conditions; the monitoring reporting conditions include: reporting threshold and / or reporting timing duration; (1) triggering the model monitoring reporting operation when the monitoring event is that the model monitoring accuracy does not meet the usage conditions, and the model monitoring accuracy meets the entry conditions of the monitoring event, including: triggering the model monitoring reporting operation when the monitoring event is that the model monitoring accuracy does not meet the usage conditions (for example, the model monitoring accuracy is less than a certain threshold), and the first number of times that the model monitoring accuracy meets the entry conditions of the monitoring event within the reporting timing duration is greater than the reporting threshold. Monitoring reporting operation; wherein the reported content is used to indicate that the currently used model is invalid; the first number is determined according to the real-time number of times that the model monitoring accuracy meets the entry condition of the monitoring event (for example, if the entry condition is met once, the count of the first number is +1), or, the first number is determined according to the real-time number of times that the model monitoring accuracy meets the entry condition of the monitoring event and the real-time number of times that the model monitoring accuracy meets the exit condition of the monitoring event (for example, if the entry condition is met once, the count of the first number is +1, and if the exit condition is met once, the count of the first number is -1), or, the first number refers to the the number of consecutive times that the model monitoring accuracy satisfies the entry condition of the monitoring event; and / or, (2) the operation of triggering model monitoring reporting when the monitoring event is that the model monitoring accuracy satisfies the use condition and the model monitoring accuracy satisfies the exit condition of the monitoring event, comprising: triggering the model monitoring reporting operation when the monitoring event is that the model monitoring accuracy satisfies the use condition (for example, the model monitoring accuracy is greater than a certain threshold) and the second number of times that the model monitoring accuracy satisfies the exit condition of the monitoring event within the reporting timing duration is greater than the reporting threshold; wherein the reported content is used to indicate that the currently used model is invalid; the second number is determined according to the real-time number of times that the model monitoring accuracy satisfies the exit condition of the monitoring event (for example, if the exit condition is satisfied once, the count of the second number is +1), or the second number is determined according to the real-time number of times that the model monitoring accuracy satisfies the exit condition of the monitoring event and the real-time number of times that the model monitoring accuracy satisfies the entry condition of the monitoring event (for example, if the entry condition is satisfied once, the count of the second number is -1, and if the exit condition is satisfied once, the count of the second number is +1), or the second number refers to the consecutive number of times that the model monitoring accuracy satisfies the exit condition of the monitoring event.

[0249] This can more accurately control the triggering operations of model monitoring reports. Specifically, there are two levels in this solution:

[0250] (1) The first level is: based on the monitoring accuracy calculated by the K test sample points (corresponding to the at least two test sample points mentioned above) in the current sliding window (i.e., the time domain sliding window), and compared with the monitoring event related information (the definition of the monitoring event, the entry condition and / or exit condition of the monitoring event), the monitoring event situation in the current sliding window can be obtained. Therefore, the first level is to compare the monitoring accuracy calculated by the K test sample points of a certain window with the monitoring event state threshold (including the threshold of the entry condition and / or exit condition) to determine the current single monitoring event state (i.e., whether a monitoring event is currently entered or exited). The first level can also be understood as: the monitoring accuracy determines the state of a single monitoring event.

[0251] (2) The second level is: after sliding the window, the monitoring accuracy calculated by the K test sample points of multiple windows can obtain the corresponding monitoring event status at multiple moments (entering and / or leaving a monitoring event); compare the cumulative effect of the monitoring event status within a certain period of time (defined by the timer) (for example, the number of times a certain monitoring event enters and / or leaves at present) with the monitoring event trigger reporting threshold (for example, the maximum threshold of the number of times entering a certain monitoring event, the maximum threshold of the number of times leaving a certain monitoring event, the maximum threshold of the number of consecutive monitoring events, or the maximum threshold of the number of consecutive times leaving a certain monitoring event, etc.) to determine whether to trigger the model monitoring report. The first level can also be understood as: judging whether the monitoring event triggers the model monitoring report based on the cumulative effect of multiple monitoring event states within a certain period of time.

[0252] Based on the above two levels, the monitoring accuracy calculated by the K test sample points of a single window only determines the monitoring event status of the window at the current moment, and the cumulative effect of the monitoring event status of multiple sliding windows can determine whether to trigger the model monitoring report; further, since the above two levels are based on the cumulative effect of the monitoring event status observed by multiple sliding windows within a certain period of time to determine whether to trigger the model monitoring report, the possibility of false alarm caused by a single result disturbance of one or more sliding windows can be minimized, thereby reducing the possibility of the ping-pong effect of frequent model switching.

[0253] In addition, this solution may also be a special case. For example, it is not based on the second level mentioned above, but only determines whether to trigger the report based on the monitoring accuracy calculated based on K test sample points in a single window (it can be understood that as long as a certain monitoring event is entered or left, the model monitoring report is triggered).

[0254] Among them, the above-mentioned reporting timing duration can correspond to the "expiration timer" mentioned below, which can be specifically set to the length of one or more windows (that is, the time domain sliding window), so it covers the above-mentioned levels one and two.

[0255] In an embodiment of the present application, the model includes a currently used model and at least one candidate model; obtaining the model monitoring accuracy based on at least two test sample points includes: obtaining a first model monitoring accuracy for the currently used model based on at least two test sample points corresponding to the currently used model; and obtaining a second model monitoring accuracy for the candidate model based on at least two test sample points corresponding to the candidate model.

[0256] This can support simultaneous monitoring of the currently used model and at least one candidate model.

[0257] Among them, the monitoring events include: a first monitoring event for the currently used model, and a second monitoring event for the candidate model; and / or, the reporting threshold includes: a first reporting threshold for the currently used model, and a second reporting threshold for the candidate model; and / or, the reporting timing duration includes: a first reporting timing duration for the currently used model, and a second reporting timing duration for the candidate model.

[0258] This can support accurately controlling the triggering operation of model monitoring reporting based on the monitoring of the currently used model and at least one candidate model.

[0259] In an embodiment of the present application, the operation of triggering model monitoring reporting when the monitoring event is that the model monitoring accuracy does not meet the usage conditions and the model monitoring accuracy meets the entry conditions of the monitoring event includes: (1) when the first monitoring event is that the first model monitoring accuracy does not meet the usage conditions (for example, the first model monitoring accuracy is less than a certain threshold) and the first condition is met; and the second monitoring event is that the second model monitoring accuracy of any candidate model meets the usage conditions (for example, the second model monitoring accuracy is greater than a certain threshold) and the second condition is met, the operation of triggering model monitoring reporting; the content of the report is used to indicate model switching and carries information about the candidate model; or, (2) when the first monitoring event is that the first model monitoring accuracy does not meet the usage conditions (for example, the first model monitoring accuracy is less than a certain threshold) and the first condition is met. The event is that the monitoring accuracy of the first model does not meet the use condition (for example, the monitoring accuracy of the first model is less than a certain threshold) and meets the first condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model does not meet the use condition (for example, the monitoring accuracy of the second model is less than a certain threshold) and meets the third condition, triggering the operation of model monitoring reporting; the content of the report is used to indicate model switching and carries the information of the candidate model; wherein (1) the first condition includes: within the first reporting timing period, the third number of times that the monitoring accuracy of the first model meets the entry condition of the first monitoring event is greater than the first reporting threshold; the third number is based on the number of times that the monitoring accuracy of the first model meets the entry condition of the first monitoring event The third number is determined by the real-time number of times the first model monitoring accuracy meets the entry condition of the first monitoring event (for example, if the entry condition is met once, the count of the third number is +1), or the third number is determined based on the real-time number of times the first model monitoring accuracy meets the entry condition of the first monitoring event and the real-time number of times the first model monitoring accuracy meets the exit condition of the first monitoring event (for example, if the entry condition is met once, the count of the third number is +1, and if the exit condition is met once, the count of the third number is -1), or the third number refers to the consecutive number of times the first model monitoring accuracy meets the entry condition of the first monitoring event; (2) the second condition includes: within the second reporting timing period, the second model monitoring accuracy meets the entry condition of the second monitoring event the fourth number of times the second model monitoring accuracy meets the entry condition of the second monitoring event is greater than the second reporting threshold; the fourth number is determined according to the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event (for example, if the entry condition is met once, the fourth number is counted +1), or, the fourth number is determined according to the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event (for example, if the entry condition is met once, the fourth number is counted +1, and if the exit condition is met once, the fourth number is counted -1), or, the fourth number refers to the consecutive number of times the second model monitoring accuracy meets the entry condition of the second monitoring event;(3) The third condition includes: the fifth number of times that the second model monitoring accuracy satisfies the exit condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the fifth number is determined based on the real-time number of times that the second model monitoring accuracy satisfies the exit condition of the second monitoring event (for example, if the exit condition is satisfied once, the fifth number is counted +1), or the fifth number is determined based on the real-time number of times that the second model monitoring accuracy satisfies the exit condition of the second monitoring event and the real-time number of times that the second model monitoring accuracy satisfies the entry condition of the second monitoring event (for example, if the entry condition is satisfied once, the fifth number is counted -1, and if the exit condition is satisfied once, the fifth number is counted +1), or the fifth number refers to the consecutive number of times that the second model monitoring accuracy satisfies the exit condition of the second monitoring event. ;

[0260] In this way, the monitoring and reporting of model switching can be accurately realized. Among them, the above-mentioned "triggering the operation of model monitoring reporting; the reported content is used to indicate model switching" can be understood as: the performance of the currently used model is too poor (triggering the model monitoring and reporting conditions of the currently used model) and at least one candidate model has good performance (triggering the model monitoring and reporting conditions of the candidate model), and switching is performed; but it is not limited to this. Among them, regarding "the second model monitoring accuracy of any candidate model meets the use conditions", it can also be implemented as: the second model monitoring accuracy of at least two candidate models meets the use conditions. In this case, the reported content can carry the information of at least one candidate model in the at least two candidate models, such as the target candidate model after screening, but it is not limited to this. Similarly, regarding "the second model monitoring accuracy of any candidate model does not meet the use conditions", it can also be implemented as: the second model monitoring accuracy of at least two candidate models does not meet the use conditions. In this case, the reported content can carry the information of at least one candidate model in the at least two candidate models, such as the target candidate model after screening, but it is not limited to this.

[0261] Among them, the operation of triggering model monitoring reporting when the monitoring event is that the model monitoring accuracy meets the use condition and the model monitoring accuracy meets the leaving condition of the monitoring event includes: (1) when the first monitoring event is that the first model monitoring accuracy meets the use condition (for example, the first model monitoring accuracy is greater than a certain threshold) and satisfies the fourth condition; and the second monitoring event is that the second model monitoring accuracy of any candidate model meets the use condition (for example, the second model monitoring accuracy is greater than a certain threshold) and satisfies the fifth condition, the operation of triggering model monitoring reporting; the content of the report is used to indicate model switching and carries the information of the candidate model; or, (2) when the first monitoring event is the first The model monitoring accuracy meets the use conditions (for example, the first model monitoring accuracy is greater than a certain threshold) and meets the fourth condition; and the second monitoring event is that the second model monitoring accuracy of any candidate model does not meet the use conditions (for example, the second model monitoring accuracy is less than a certain threshold) and meets the sixth condition, triggering the model monitoring reporting operation; the reported content is used to indicate model switching and carries the information of the candidate model; wherein (1) the fourth condition includes: the sixth number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event within the first reporting timing period is greater than the first reporting threshold; the sixth number is based on the number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event The sixth number is determined by the real-time number of times the first model monitoring accuracy satisfies the leaving condition of the first monitoring event (for example, if the leaving condition is satisfied once, the sixth number is counted as +1), or the sixth number is determined according to the real-time number of times the first model monitoring accuracy satisfies the leaving condition of the first monitoring event and the real-time number of times the first model monitoring accuracy satisfies the entering condition of the first monitoring event (for example, if the entering condition is satisfied once, the sixth number is counted as -1, and if the leaving condition is satisfied once, the sixth number is counted as +1), or the sixth number refers to the consecutive number of times the first model monitoring accuracy satisfies the entering condition of the first monitoring event; (2) the fifth condition includes: within the second reporting timing duration, the second model monitoring accuracy satisfies the entering condition of the second monitoring event The seventh number of the event is greater than the second reporting threshold; the seventh number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event (for example, if the entry condition is met once, the count of the seventh number is +1), or, the seventh number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event (for example, if the entry condition is met once, the count of the seventh number is +1, and if the exit condition is met once, the count of the seventh number is -1), or, the seventh number refers to the consecutive number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event;(3) The sixth condition includes: the eighth number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the eighth number is determined according to the real-time number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event (for example, if the exit condition is met once, the eighth number is counted +1), or the eighth number is determined according to the real-time number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event and the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event (for example, if the entry condition is met once, the eighth number is counted -1, and if the exit condition is met once, the eighth number is counted +1), or the eighth number refers to the consecutive number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event. ;

[0262] In this way, the monitoring and reporting of model switching can also be accurately realized. Among them, the above-mentioned "triggering the operation of model monitoring reporting; the reported content is used to indicate model switching" can be understood as: the performance of the currently used model is too poor (triggering the model monitoring and reporting conditions of the currently used model) and at least one candidate model has good performance (triggering the model monitoring and reporting conditions of the candidate model), and switching is performed; but it is not limited to this. Among them, regarding "the second model monitoring accuracy of any candidate model meets the use conditions", it can also be implemented as: the second model monitoring accuracy of at least two candidate models meets the use conditions. In this case, the reported content can carry the information of at least one candidate model in the at least two candidate models, such as the target candidate model after screening, but it is not limited to this. Similarly, regarding "the second model monitoring accuracy of any candidate model does not meet the use conditions", it can also be implemented as: the second model monitoring accuracy of at least two candidate models does not meet the use conditions. In this case, the reported content can carry the information of at least one candidate model in the at least two candidate models, such as the target candidate model after screening, but it is not limited to this.

[0263] In an embodiment of the present application, the triggering operation of controlling the model monitoring report based on the model monitoring accuracy and the information related to the monitoring event also includes: triggering the model monitoring reporting operation when the monitoring event is that the model monitoring accuracy of the currently used model is less than the model monitoring accuracy of any candidate model; wherein the reported content is used to indicate model switching and carries the information of the candidate model.

[0264] In this way, it is possible to directly trigger the operation of model monitoring reporting to switch models when the model monitoring accuracy of the currently used model is less than the model monitoring accuracy of any candidate model (for example, the monitoring accuracy of the new model is better than the old model, so the model is switched). Among them, "the model monitoring accuracy of the currently used model is less than the model monitoring accuracy of any candidate model" can also be implemented as: the model monitoring accuracy of the currently used model is less than the model monitoring accuracy of at least two candidate models. In this case, the reported content can carry information of at least one of the at least two candidate models, such as a screened target candidate model, but is not limited to this.

[0265] Furthermore, the information processing method further includes: using a time domain sliding window of a first length to obtain the at least two test sample points.

[0266] In this way, the at least two test sample points can be accurately obtained.

[0267] In an embodiment of the present application, before obtaining the model monitoring accuracy based on at least two test sample points, it also includes: preprocessing the at least two test sample points; wherein the preprocessing includes: at least one of weighted processing and abnormal elimination processing (such as eliminating certain abnormal sample points).

[0268] This can improve the accuracy of the test samples used, so as to further improve the accuracy of the obtained model monitoring accuracy.

[0269] The entry condition and / or exit condition is determined according to a blocking parameter and a corresponding threshold value.

[0270] This can avoid the ping-pong effect of frequent switching of model monitoring to a certain extent. Specifically, the introduction of a blocking parameter can reduce the impact of observation disturbance on the decision; wherein, the values ​​of the blocking parameter and / or the threshold corresponding to different conditions may be different, but not limited to this.

[0271] In an embodiment of the present application, the monitoring event related information is determined according to the protocol predefinition and / or the network side configuration; and / or, the monitoring reporting condition is determined according to the protocol predefinition and / or the network side configuration; and / or, the first length is determined according to the protocol predefinition and / or the network side configuration; and / or, the sliding step of the time domain sliding window is determined according to the protocol predefinition and / or the network side configuration; and / or, the blocking parameter is determined according to the protocol predefinition and / or the network side configuration; and / or, the threshold value is determined according to the protocol predefinition and / or the network side configuration.

[0272] This can improve the flexibility of solution implementation.

[0273] The present application also provides an information processing method, which is applied to a first device, such as Figure 6 As shown, including:

[0274] Step 61: Use a time domain sliding window of a first length to obtain at least two test sample points; wherein the test sample points are determined according to a target CSI and a CSI corresponding to the target CSI and recovered by a model; and the first device is a network device or a terminal.

[0275] Among them, depending on whether the monitoring is done on the network side or the terminal side (i.e., the first device is a network device or a terminal), the target channel state information (i.e., the target CSI) can be various, for example: (1) obtained from a CSI-related reference signal (e.g., the above Figure 3 and 4 (2) obtained from the CSI-related reference signal and reported by the terminal (i.e., the terminal report will be quantized and / or compressed again to generate new errors, as mentioned above Figure 2 Of course, other possibilities may also be included, which are not limited here.

[0276] There are also multiple possibilities for obtaining the “CSI corresponding to the target CSI and recovered by the model”, for example: (1) the output of the CSI recovery part of the AI / ML model, as mentioned above Figure 2 ; (2) The output of the CSI recovery part of the AI / ML model is then transmitted back to the terminal via the network side, as mentioned above Figure 3 ; (3) The UE side uses the alternative model to recover, as mentioned above Figure 4 ; Of course, there may be other possibilities, which are not limited here.

[0277] The information processing method provided in the embodiment of the present application obtains at least two test sample points by adopting a time domain sliding window of a first length; wherein the test sample points are determined according to the target CSI and the CSI corresponding to the target CSI and obtained by model recovery; the first device is a network device or a terminal; it can support accurate acquisition of test sample points, thereby improving the accuracy of the acquired model monitoring accuracy, and further support accurate control of model monitoring reporting triggering according to the model monitoring accuracy (i.e., monitoring results), thereby supporting accurate implementation of decision operations such as model switching, fallback, update, activation and / or deactivation, and well solving the problem that the information processing scheme for model monitoring in the prior art cannot support accurate implementation of model decision operations.

[0278] Furthermore, the information processing method further includes: preprocessing the at least two test sample points; wherein the preprocessing includes: at least one of weighted processing and abnormal elimination processing (such as eliminating certain abnormal sample points).

[0279] This can improve the accuracy of the test samples used, so as to further improve the accuracy of the obtained model monitoring accuracy.

[0280] Among them, the first length is determined according to protocol pre-definition and / or network side configuration; and / or the sliding step size of the time domain sliding window is determined according to protocol pre-definition and / or network side configuration.

[0281] This can improve the flexibility of solution implementation.

[0282] The information processing method provided in the embodiments of the present application is described below with examples.

[0283] In view of the above technical problems, the embodiment of the present application provides an information processing method, which can be specifically implemented as an event-triggered information reporting model monitoring method. This method mainly targets Figure 2-4 The part in the dashed box solves: (1) how to obtain the K test sample data in the time series data (which can be understood as the acquisition of K intermediate KPI sample points); (2) how to trigger the AI / ML model monitoring report based on the AI / ML model monitoring accuracy represented by the performance evaluation statistics of these K sample points, so as to request the network side to make decisions on the activation, deactivation, update, rollback and / or switching of the AI / ML model. Furthermore, it can also solve the potential ping-pong effect problem of frequent changes in AI / ML models.

[0284] In addition, this solution is applicable to both network-side AI / ML model monitoring and UE-side AI / ML model monitoring, but there will be some differences in the specific details; for example: when monitoring the AI / ML model on the UE side, the network side may configure some relevant parameters on the UE side, but the network-side AI / ML model monitoring does not have the above parameter configuration problem, but what may need to be considered is the design of the trigger signaling for the network side to trigger the UE to "report the target CSI on the UE side". However, it may also be considered that the network side determines when to trigger the UE to report, which belongs to the specific implementation of the base station and can be unrestricted. Therefore, this solution will try to describe it in a universal (both network and UE side) AI / ML model monitoring manner, and then refine it according to the network side and UE side AI / ML model monitoring respectively.

[0285] This plan mainly includes the following contents:

[0286] 1. The AI / ML model monitoring entity (which can belong to the network side or the UE side; corresponding to the above-mentioned first device being a network device or a terminal) establishes a time-domain sliding window with a length of K (corresponding to the above-mentioned first length) to obtain a test data set of K test sample points, and the monitoring accuracy (corresponding to the above-mentioned model monitoring accuracy) is based on the statistical results of these K test sample points (corresponding to the above-mentioned model monitoring accuracy being based on the statistical results of the at least two test sample points).

[0287] Optionally, the configuration of the window length K is pre-defined by the protocol, configured by the network side, or a combination of protocol pre-definition and network side configuration (corresponding to the above-mentioned first length being determined according to protocol pre-definition and / or network side configuration).

[0288] Optionally, the sliding step Nstep of the sliding window (i.e., the time-domain sliding window) in the time domain is configured to be pre-defined by the protocol, configured by the network side, or a combination of protocol pre-definition and network side configuration (corresponding to the sliding step of the above-mentioned time-domain sliding window being determined according to protocol pre-definition and / or network side configuration).

[0289] Optionally, the AI / ML model monitoring entity can preprocess the K test sample points (including related operations to improve the accuracy of the calculation results). For example, the preprocessing includes, but is not limited to: taking a subset containing K' (<K) test sample points for calculating the monitoring accuracy, weighting the test sample points when calculating the monitoring accuracy based on the statistical results of the K test sample points, etc. (such as cleaning the data, etc.); corresponding to the above-mentioned preprocessing of the at least two test sample points; wherein, the preprocessing includes at least one of weighting processing and abnormal elimination processing.

[0290] 2. Define the information related to the monitoring events of the AI / ML model monitoring entity (network side or UE side). The information related to the monitoring events can include: the definition of the monitoring event, as well as the entry condition and the exit condition of the event. When the monitoring accuracy calculated based on the time-domain sliding window with a length of K meets the entry condition, it is considered that the entry of the monitoring event is satisfied. When the monitoring accuracy calculated based on the time-domain sliding window with a length of K meets the exit condition, it is considered that the exit of the monitoring event is satisfied.

[0291] Optionally, the entry condition and / or the exit condition includes a threshold parameter (corresponding to the above-mentioned threshold), and the configuration of this threshold parameter is pre-defined by the protocol, configured by the network side, or a combination of protocol pre-definition and network side configuration (corresponding to the above-mentioned threshold being determined according to protocol pre-definition and / or network side configuration).

[0292] 3. Define the conditions for the AI / ML model monitoring entity (network side or UE side) to trigger the AI / ML model monitoring report for the monitoring events (corresponding to the above-mentioned monitoring report conditions).

[0293] Optionally, the configuration of the threshold parameter (corresponding to the above reporting threshold) for the monitoring event of the monitoring entity to trigger the monitoring report of the AI / ML model is pre-defined by the protocol, configured by the network side, or a combination of protocol pre-definition and network side configuration (corresponding to the above monitoring report condition being determined according to protocol pre-definition and / or network side configuration).

[0294] IV. The monitoring entity (network side or UE side) of the AI / ML model uses a time-domain sliding window of length K to calculate the monitoring accuracy and defined monitoring events to determine the entry and exit of the monitoring event, and then determines whether to trigger the monitoring report of the AI / ML model according to the determination result of the monitoring event and the condition for the monitoring event to trigger the monitoring report of the AI / ML model (corresponding to controlling the trigger operation of the model monitoring report according to the model monitoring accuracy, monitoring event-related information, and monitoring report condition).

[0295] Combined with the above Figures 2 to 3 process, the present solution will be specifically introduced below.

[0296] Method 1: Network-side AI / ML model monitoring, based on the UE-side reported target CSI (corresponding to the above Figure 2 , the first device is a network device);

[0297] Part 1: The network side establishes a sliding window of length K (length K means including K test sample points) to obtain a test data set of K test sample points, and the monitoring accuracy is based on the statistical results of these K test sample points (corresponding to the above using a time-domain sliding window of the first length to obtain the at least two test sample points; the model monitoring accuracy is based on the statistical results of the at least two test sample points). The sliding window length K is pre-defined by the protocol, configured by the network side itself, or a combination of protocol pre-definition and network side self-configuration (corresponding to the above first length being determined according to protocol pre-definition and / or network side configuration). The configuration of the sliding step Nstep of the sliding window is pre-defined by the protocol, configured by the network side itself, or a combination of protocol pre-definition and network side self-configuration (corresponding to the above sliding step of the time-domain sliding window being determined according to protocol pre-definition and / or network side configuration). Optionally, the network side can preprocess the test data samples in the sliding window. For example, the network side can discard some samples that it considers inaccurate or unrepresentative (corresponding to the above anomaly rejection processing), that is, select a subset with a cardinality of K' (K'<K) for the next calculation; or the network side can weight each data sample (corresponding to the above weighting processing) to change its influence on the final statistical result.

[0298] Taking Figure 2For example, one test sample point refers to: an intermediate KPI calculated based on a CSI recovered by an AI / ML model (the CSI recovered by the AI / ML model is the output of the compressed CSI received by the network side and reported back by the UE side after the CSI recovery part of the AI / ML model on the network side) and a corresponding target CSI reported by the UE side obtained by the network side (for example, the target CSI reported by the UE side can be based on the physical uplink shared channel (Physical Uplink Shared Channel, PUSCH), can be based on the traditional enhanced type II codebook CSI reporting (for example: parameter configuration (Parameter Configuration, PC) 6 and / or PC8 configuration), can be based on similar enhanced type II codebook CSI reporting (based on a more refined codebook configuration), can be based on the SRS sent by the UE side, etc.). Specifically, from the perspective of the time axis, the acquisition time of each intermediate KPI is a discrete point on the time axis. Since the reporting methods of compressed CSI and / or target CSI are various (periodic, semi-periodic, non-periodic), and the reporting strategies are flexible and changeable (the target CSI is reported once each time the compressed CSI is fed back and reported, and the target CSI corresponding to the previously reported compressed CSI is reported together after multiple compressed CSIs are fed back and reported, etc.), the interval between any two intermediate KPI acquisition times can be any number greater than or equal to zero. Assume that Figure 7 (Example of intermediate KPI acquisition time and sliding window on the time axis): window length K = 3, sliding step Nstep = 3 (corresponding to the solid line box in the figure), or window length K = 4, sliding step Nstep = 2 (corresponding to the dotted line box in the figure). It should be noted that the above discrete points (i.e., intermediate KPI acquisition time) are not necessarily equally spaced, and one discrete point can correspond to multiple intermediate KPI acquisition times, such as Figure 7 The acquisition times 4 to 7 in the figure all correspond to the fourth discrete point from left to right in the figure (i.e., the 4th to 7th intermediate KPI acquisition times are the same, corresponding to the above-mentioned situation of "reporting the target CSI corresponding to the previously reported compressed CSI together", that is, the 4th to 7th times are reported together after accumulating multiple intermediate KPIs, and the corresponding intermediate KPI acquisition times are the same).

[0299] Among them, it is pre-assumed in this solution that both the UE-side feedback reporting compressed CSI method and the UE-side reporting target CSI method have been determined and the corresponding relationship has also been clarified. Considering the latency of the AI / ML model monitoring entity when calculating the intermediate KPI, it can be considered that the time when each intermediate KPI is obtained on the time axis = max [the time when the CSI recovered by the AI / ML model (the output obtained through the CSI recovery part of the AI / ML model based on the UE-side feedback reporting compressed CSI) is obtained, the time when the UE-side reporting target CSI corresponding to it (i.e., the time when the CSI recovered by the AI / ML model is obtained)] + the latency of calculating the intermediate KPI, but not limited to this.

[0300] Part 2: The network side defines the AI / ML model monitoring events Event1 (for example: the monitoring accuracy of the network side AI / ML model is greater than a certain threshold, and it can be understood that this event is established when the entry condition is met) and Event2 (for example: the monitoring accuracy of the network side AI / ML model is less than a certain threshold).

[0301] (1) Define the network side event Event1 (the monitoring accuracy of the network side AI / ML model is greater than a certain threshold):

[0302] The network shall:

[0303] 1) When the following specified condition Condition1-1 is met, consider that the entry condition of this event Event1 is satisfied. The condition Condition1-1 is as follows;

[0304] 2) When the following specified condition Condition1-2 is met, consider that the departure condition of this event Event1 is satisfied. The condition Condition1-2 is as follows;

[0305] The condition Condition1-1 (entry condition) can be the following inequality:

[0306] M NW -H ys > Thresh1; the above entry condition is determined according to the blocking parameter and the corresponding threshold value;

[0307] The condition Condition1-2 (departure condition) can be the following inequality:

[0308] M NW +H ys < Thresh1; the above departure condition is determined according to the blocking parameter and the corresponding threshold value;

[0309] Among them, the variable definitions in the above formula are as follows:

[0310] M NW Refers to the monitoring accuracy of the AI / ML model calculated by the network side (corresponding to the above-mentioned model monitoring accuracy), and this calculated value does not include any offset value;

[0311] H ys Refers to the hysteresis parameter of Event 1. This parameter is configured as pre-defined by the protocol, configured by the network side itself, or a combination of pre-defined by the protocol and configured by the network side; among them, the hysteresis parameter can be understood as a kind of margin or the confidence interval in statistics (such as 3σ). When the difference between the observed value (corresponding to the above-mentioned model monitoring accuracy) and the threshold is significantly greater than the hysteresis parameter, a decision can be made. Introducing the hysteresis parameter can reduce the impact of observation perturbation on the decision.

[0312] Thresh1 refers to the threshold parameter of Event 1. This parameter is configured as pre-defined by the protocol, configured by the network side itself, or a combination of pre-defined by the protocol and configured by the network side itself.

[0313] Where M NW 、H ys and Thresh1 have the same dimension.

[0314] Meanwhile, (2) defines the network side event Event2 (the monitoring accuracy of the network side AI / ML model is less than a certain threshold):

[0315] The network shall:

[0316] 1) When the following specified condition Condition2-1 is met, consider meeting the entry condition of this event Event2. The condition Condition2-1 is as follows;

[0317] 2) When the following specified condition Condition2-2 is met, consider meeting the departure condition of this event Event2. The condition Condition2-2 is as follows;

[0318] The condition Condition2-1 (entry condition) can be the following inequality:

[0319] M NW +H ys ′ < Thresh2; corresponding to the above entry condition is determined according to the hysteresis parameter and the corresponding threshold value;

[0320] The condition Condition2-2 (departure condition) can be the following inequality:

[0321] M NW -H ys′>Thresh2; corresponding to the above-mentioned leaving condition is determined according to the blocking parameter and the corresponding threshold value;

[0322] The variables in the above formula are defined as follows:

[0323] M NW Refers to the AI / ML model monitoring accuracy calculated by the network side. This calculated value does not include any offset value.

[0324] H ys ' refers to the blocking parameter of event Event2, which is configured as protocol predefined, network self-configured, or a mixture of protocol predefined and network self-configured;

[0325] Thresh2 refers to the threshold parameter of Event2, which can be predefined by the protocol, configured by the network side, or a mixture of predefined by the protocol and configured by the network side.

[0326] Among them, M NW , H ys ′ and Thresh2 have the same dimension.

[0327] Part 3: Define the conditions for triggering AI / ML model monitoring reporting by AI / ML model monitoring events Event1 and / or Event2 (corresponding to the above-mentioned monitoring reporting conditions). The conditions for triggering AI / ML model monitoring reporting are related to the entry and exit of Event1 (and / or Event2) monitored by the network side, the threshold parameters for the occurrence of monitoring events Event1 (and / or Event2) configured on the network side (for example, the maximum or minimum number of occurrences of the event, the maximum or minimum number of consecutive occurrences of the event, etc.), and the timer parameters corresponding to the monitoring events Event1 (and / or Event2) (corresponding to the above-mentioned reporting timing duration); the timer parameters are predefined by the protocol, configured on the network side, or a mixture of protocol predefined and network side configured.

[0328] The following is an example of a solution under this method.

[0329] Implementation example 1a: Network side AI / ML model monitoring, based on the number of entry and / or exit times of monitoring events of a single or multiple sliding windows being greater than a threshold, it is determined that the currently used AI / ML model M1 is invalid (here, the AI / ML model M1 may refer to a pair of models for UE side CSI generation and network side CSI recovery);

[0330] The network side defines the maximum number of occurrences N of monitoring event Event2 for AI / ML model M1 max_count_E2(corresponding to the above reporting threshold), this parameter configuration is predefined by the protocol, self-configured by the network side, or a mixture of predefined by the protocol and self-configured by the network side. max_count_E2 The associated invalidation timer (corresponding to the above-mentioned reporting time duration) is configured as a parameter predefined by the protocol, configured by the network side, or a mixture of a predefined protocol and a self-configured network side. max_count_E2 =1. Under this configuration, whether a monitoring event triggers an AI / ML model monitoring report can be determined by the network-side monitoring accuracy calculated by the K data sample points in a single sliding window. The expiration timer is an expiration counter that counts the number of occurrences of the monitoring event Event2. When the timer returns to zero (countdown) or reaches the upper limit (countup), the previously counted number of occurrences of the monitoring event Event2 returns to zero and starts counting again.

[0331] The network side monitors the number of occurrences of AI / ML model M1 Event2, where M NW According to the sliding window calculation described above. When the network side calculates and monitors (satisfies) the entry condition (Condition2-1) of Event2 based on the current sliding window (that is, the time domain sliding window), the network side records the current number of occurrences of Event2 plus 1; when the network side calculates and monitors (satisfies) the exit condition (Condition2-2) of Event2 based on the current sliding window, the network side records the current number of occurrences of Event2 minus 1. When the expiration timer has not returned to zero (countdown) or reached the upper limit (countup), and the current number of occurrences of Event2 recorded by the network side is greater than the maximum number of occurrences of monitored event Event2 N max_count_E2 When the network side triggers the AI / ML model monitoring report (which can be a report between internal modules on the network side, such as a report from the monitoring module to the decision module), the network side reports that the currently used AI / ML model M1 is invalid (corresponding to the above-mentioned operation of triggering the model monitoring report when the monitoring event is that the model monitoring accuracy does not meet the usage conditions, and the first number of times that the model monitoring accuracy meets the entry conditions of the monitoring event within the reporting timing is greater than the reporting threshold; wherein the reported content is used to indicate that the currently used model is invalid; the first number is determined according to the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event and the real-time number of times that the model monitoring accuracy meets the exit conditions of the monitoring event).

[0332] Implementation example 1b: Network-side AI / ML model monitoring, based on the number of monitoring events entering a single or multiple sliding windows being greater than a threshold, determines that the currently used AI / ML model M1 is invalid.

[0333] Similar to implementation example 1a, the difference is that the occurrence of the exit condition (Condition2-2) of Event2 is not considered: only the occurrence of the entry condition (Condition2-1) of Event2 monitored by the network side calculated according to the current sliding window or the number of consecutive occurrences is greater than the maximum number of occurrences N of the monitored event Event2 max_count_E2 Or monitor the maximum number of consecutive occurrences of event Event2 N max_consec_count_E2 When the network side triggers the AI / ML model monitoring report, the network side reports that the currently used AI / ML model M1 is invalid; corresponding to the above-mentioned operation of triggering the model monitoring report when the monitoring event is that the model monitoring accuracy does not meet the usage conditions, and the first number of times that the model monitoring accuracy meets the entry conditions of the monitoring event within the reporting timing is greater than the reporting threshold; wherein the reported content is used to indicate that the currently used model is invalid; the first number is determined according to the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event, or the first number refers to the continuous number of times that the model monitoring accuracy meets the entry conditions of the monitoring event.

[0334] Implementation Example 1c (can be used in combination with Implementation Example 1a or Implementation Example 1b): Network-side AI / ML model monitoring, based on the number of monitoring event entries of a single or multiple sliding windows being greater than a threshold, determines that the currently used AI / ML model M1 is invalid, and the count of consecutive monitoring events Event2 is reset to zero.

[0335] Similar to implementation example 1a or implementation example 1b, the difference is that when the network side calculates the monitoring accuracy based on the sliding window and finds that N consecutive max_consec_count_E2out If the exit condition of Event2 (Condition2-2) is met for the first time, the network side will reset the number of times Event2 occurs to zero and reset the invalidation timer to the upper limit (countdown) / zero (countup). max_consec_count_E2out It can be predefined by the protocol, configured by the network side, or a mixture of predefined by the protocol and configured by the network side.

[0336] Among them: The above implementation examples 1a-1c are all network-side AI / ML model monitoring, and are several possible solutions based on event-triggered monitoring and reporting.

[0337] Implementation example 2: Network-side AI / ML model monitoring, based on the number of entry and / or exit times of monitoring events of a single or multiple sliding windows being greater than a threshold, determines that the currently used network-side CSI restores the AI / ML model switching.

[0338] The network side monitors the currently used network side CSI recovery AI / ML model C1, and simultaneously monitors the unused, potential network side CSI recovery AI / ML model C2 (because the network side also has the CSI recovery model C2, the received UE side feedback reported compressed CSI can be obtained through model C2 to obtain the recovered CSI based on C2, and then the intermediate KPI is calculated with the target CSI reported by the UE side). Here, it is explained that here one potential CSI recovery model C2 (corresponding to the above candidate model) is taken as an example, of course, it can also be applied to multiple potential CSI recovery models C3, C4, ..., CN, which is not limited here.

[0339] The network side configures the parameters of Event1 and / or Event2 for C1 and C2 respectively. The parameters are predefined by the protocol, configured by the network side, or a mixture of predefined by the protocol and configured by the network side. The network side defines the maximum number of occurrences of monitoring event Event2 for C1: N max_count_E2 The network side defines the minimum occurrence number N of monitoring event Event1 for C2 min_count_E1 The above parameters are predefined by the protocol, configured by the network side, or a mixture of predefined by the protocol and configured by the network side. max_count_E2 and N min_count_E1 (corresponding to the above-mentioned reporting threshold) and the associated invalidation timer (corresponding to the above-mentioned reporting timing duration), the above-mentioned parameter configuration is predefined by the protocol, self-configured by the network side, or a mixture of predefined by the protocol and self-configured by the network side. Among them, N can also be configured max_count_E2 =1, at this time, the K data samples in a single sliding window determine whether to trigger the AI / ML model monitoring report; you can also configure N min_count_E1 =1, at this time, the K data samples in a single sliding window determine whether to trigger the AI / ML model monitoring report.

[0340] The network side monitors the number of occurrences of Event2 related to C1, where M NW According to the sliding window calculation described above, when the network side monitors the entry condition of Event2 (Condition2-1) based on the current sliding window calculation, the network side records the number of occurrences of Event2 plus 1; when the network side monitors the exit condition of Event2 (Condition2-2) based on the current sliding window calculation, the network side records the number of occurrences of Event2 minus 1.

[0341] The network side monitors the number of occurrences of C2-related Event1, where M NWAccording to the sliding window calculation described above, when the network side monitors the entry condition (Condition1-1) of Event1 based on the current sliding window calculation, the network side records the number of occurrences of Event1 plus 1; when the network side monitors the exit condition (Condition1-2) of Event1 based on the current sliding window calculation, the network side records the number of occurrences of Event1 minus 1.

[0342] When the invalidation timer configured for C1 has not yet reached zero (countdown) or reached the upper limit (countup), and the number of occurrences of C1's current Event2 recorded on the network side is greater than its configured N max_count_E2 , and the invalidation timer configured on C2 has not yet reached zero (countdown) or reached the upper limit (countup), and the number of occurrences of the current Event1 of C2 recorded on the network side is greater than its configured N min_count_E1 , the network side triggers the AI / ML model monitoring report, and reports that the performance of the currently used CSI recovery AI / ML model C1 is worse than that of the CSI recovery AI / ML model C2, and C2 is reported as the candidate switching model; corresponding to the above-mentioned operation of triggering the model monitoring report when the first monitoring event is that the monitoring accuracy of the first model does not meet the use condition and meets the first condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model meets the use condition and meets the second condition; the reported content is used to indicate model switching and carries the information of the candidate model; wherein the first condition includes: within the first reporting timing duration, the monitoring accuracy of the first model meets the second condition. The third number of entry conditions for a monitoring event is greater than the first reporting threshold; the third number is determined based on the real-time number of times the first model monitoring accuracy meets the entry conditions for the first monitoring event and the real-time number of times the first model monitoring accuracy meets the exit conditions for the first monitoring event; the second condition includes: the fourth number of times the second model monitoring accuracy meets the entry conditions for the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the fourth number is determined based on the real-time number of times the second model monitoring accuracy meets the entry conditions for the second monitoring event and the real-time number of times the second model monitoring accuracy meets the exit conditions for the second monitoring event.

[0343] Among them: the above-mentioned implementation example 2 can also have variants similar to the above-mentioned implementation examples 1a-1c, that is, using a combination of different event entry and / or exit conditions, event (continuous) maximum or minimum number of occurrences, counter initialization or zeroing and other conditions to define the trigger monitoring report, which will not be repeated here.

[0344] Method 2: UE-side AI / ML model monitoring, based on network-side backhaul recovery CSI (corresponding to the above Figure 3 , the first device is a terminal);

[0345] Part 1: The UE side establishes a time domain sliding window of length K (the length K means that it contains K test sample points) to obtain a test data set of K test sample point data, and the monitoring accuracy is based on the statistical results of these K test sample points (corresponding to the above-mentioned use of the time domain sliding window of the first length to obtain the at least two test sample points; the model monitoring accuracy is based on the statistical results of the at least two test sample points). The sliding window length K is predefined by the protocol, configured on the network side (for example, based on RRC configuration), or a mixture of protocol predefined and network side configuration (corresponding to the above-mentioned first length is determined according to protocol predefined and / or network side configuration). The sliding step size Nstep of the sliding window in the time domain is configured as predefined by the protocol, configured on the network side (for example, based on RRC configuration), or a mixture of protocol predefined and network side configuration (corresponding to the above-mentioned sliding step size of the time domain sliding window is determined according to protocol predefined and / or network side configuration).

[0346] by Figure 3 For example, one test sample point refers to an intermediate KPI calculated based on one target CSI and one corresponding network-side backhaul recovery CSI obtained by the UE side (for example, the network-side backhaul recovery CSI can be based on the physical downlink shared channel (PDSCH), can be based on the precoded channel state information reference signal (CSI-RS), etc.). Among them, the examples of the intermediate KPI acquisition time, the sliding window length K, and the sliding window sliding step length Nstep are similar to the method 1, and will not be repeated here.

[0347] Part 2: The UE side defines AI / ML model monitoring events Event3 (for example, the AI / ML model monitoring accuracy on the UE side is greater than a certain threshold) and Event4 (for example, the AI / ML model monitoring accuracy on the UE side is less than a certain threshold).

[0348] (1) Define UE side event Event3 (UE side AI / ML model monitoring accuracy is greater than a certain threshold):

[0349] The UE shall:

[0350] 1) When the following condition Condition3-1 is met, the entry condition of the event Event3 is considered to be met. Condition Condition3-1 is as follows;

[0351] 2) When the condition Condition3-2 as specified below is satisfied, consider the departure condition for satisfying the event Event3, and the condition Condition3-2 is as follows;

[0352] The condition Condition3-1 (entry condition) can be the following inequality:

[0353] M UE -H ys ″ > Thresh3; The above-mentioned entry condition is determined according to the blocking parameter and the corresponding threshold value;

[0354] The condition Condition3-2 (departure condition) can be the following inequality:

[0355] M UE +H ys ″ < Thresh3; The above-mentioned departure condition is determined according to the blocking parameter and the corresponding threshold value;

[0356] Among them, the variable definitions in the above formula are as follows:

[0357] M UE Refers to the AI / ML model monitoring accuracy calculated on the UE side (corresponding to the above-mentioned model monitoring accuracy), and this calculated value does not include any offset value;

[0358] H ys ″ Refers to the blocking parameter of the event Event3, and this parameter is configured as pre-defined by the protocol, configured by the network side (for example, based on RRC configuration), or a mixture of pre-defined by the protocol and configured by the network side;

[0359] Thresh3 refers to the threshold parameter of Event3, and this parameter is configured as pre-defined by the protocol, configured by the network side (for example, based on Radio Resource Control (RRC) configuration), or a mixture of pre-defined by the protocol and configured by the network side.

[0360] Among them M UE 、H″ ys and Thresh3 have the same dimension.

[0361] Meanwhile, (2) define the event Event4 on the UE side (the AI / ML model monitoring accuracy on the UE side is less than a certain threshold):

[0362] The UE shall:

[0363] 1) When the condition Condition4-1 as specified below is satisfied, consider the entry condition for satisfying the event Event4, and the condition Condition4-1 is as follows;

[0364] 2) When the condition Condition4-2, which meets the following specified conditions, is satisfied, consider the departure condition for Event4. The condition Condition4-2 is as follows:

[0365] The condition Condition4-1 (entry condition) can be the following inequality:

[0366] M UE +H ys ″′ < Thresh4; The above corresponding entry condition is determined according to the hysteresis parameter and the corresponding threshold value.

[0367] The condition Condition4-2 (departure condition) can be the following inequality:

[0368] M UE -H ys ″′ > Thresh4; The above corresponding departure condition is determined according to the hysteresis parameter and the corresponding threshold value.

[0369] Among them, the variable definitions in the above formula are as follows:

[0370] M UE Refers to the AI / ML model monitoring accuracy calculated on the UE side, and this calculated value does not include any offset value.

[0371] H ys ″′ Refers to the hysteresis parameter of Event4, and this parameter is configured as pre-defined by the protocol, network-side configured (such as based on RRC configuration), or a mixture of pre-defined by the protocol and network-side configured.

[0372] Thresh4 refers to the threshold parameter of Event4, and this parameter is configured as pre-defined by the protocol, network-side configured (such as based on RRC configuration), or a mixture of pre-defined by the protocol and network-side configured.

[0373] Among them, M UE 、H″′ ys and Thresh4 have the same dimension.

[0374] Part 3: Define the conditions for triggering AI / ML model monitoring reporting by AI / ML model monitoring events Event3 and / or Event4 (corresponding to the above-mentioned monitoring reporting conditions). The conditions for triggering AI / ML model monitoring reporting are related to the entry and exit of Event3 (and / or Event4) monitored on the UE side, the threshold parameters for the occurrence of monitoring events Event3 (and / or Event4) configured on the network side (for example, the maximum or minimum number of occurrences of the event, the maximum or minimum number of consecutive occurrences of the event, etc.), and the timer parameters corresponding to the monitoring event Event3 (and / or Event4) (corresponding to the above-mentioned reporting timing duration).

[0375] The following is an example of a solution under this method.

[0376] Implementation example 3: UE-side AI / ML model monitoring (based on network-side backhaul recovery of CSI), determining that the currently used AI / ML model M1 is invalid (here, the AI / ML model M1 may refer to a pair of models for UE-side CSI generation and network-side CSI recovery);

[0377] The UE side defines the maximum number of occurrences N of monitoring event Event4 for AI / ML model M1 max_count_E4 (corresponding to the above reporting threshold), this parameter configuration is protocol predefined, network side configuration (for example, based on RRC configuration), or a mixture of protocol predefined and network side configuration. max_count_E4 The associated expiration timer (corresponding to the above-mentioned reporting timing duration), the parameter configuration is protocol predefined, network side configured (for example, based on RRC configuration), or a mixture of protocol predefined and network side configured.

[0378] The UE side monitors the number of occurrences of AI / ML model M1 Event4, where M UE According to the sliding window calculation described above. When the UE side calculates and monitors (satisfies) the entry condition (Condition4-1) of Event4 based on the current sliding window (that is, the time domain sliding window), the UE side records the number of occurrences of the current Event4 plus 1; when the UE side calculates and monitors (satisfies) the exit condition (Condition4-2) of Event4 based on the current sliding window, the UE side records the number of occurrences of the current Event4 minus 1. When the expiration timer has not returned to zero (countdown) or reached the upper limit (countup), and the number of occurrences of the current Event4 recorded by the UE side is greater than the maximum number of occurrences N of the monitored event Event4 max_count_E4When the UE triggers the AI / ML model monitoring report, reports that the currently used AI / ML model M1 is invalid, and requests the network to make a decision (corresponding to the above-mentioned operation of triggering the model monitoring reporting when the monitoring event is that the model monitoring accuracy does not meet the usage conditions, and the first number of times that the model monitoring accuracy meets the entry conditions of the monitoring event within the reporting timing duration is greater than the reporting threshold; wherein the reported content is used to indicate that the currently used model is invalid; the first number is determined according to the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event and the real-time number of times that the model monitoring accuracy meets the exit conditions of the monitoring event).

[0379] Among them: the above-mentioned implementation example 3 can also have variants similar to the above-mentioned implementation examples 1a-1c, that is, using a combination of different event entry and / or exit conditions, event (continuous) maximum or minimum number of occurrences, counter initialization or zeroing and other conditions to define the trigger monitoring report, which will not be repeated here.

[0380] Method 3: UE-side AI / ML model monitoring, based on the UE-side existing, network-side CSI recovery partial AI / ML model alternative model (corresponding to the above Figure 4 , the first device is a terminal);

[0381] Part 1: The UE side establishes a time domain sliding window of length K (the length K means that it contains K test sample points) to obtain a test data set of K test sample point data, and the monitoring accuracy is based on the statistical results of these K test sample points (corresponding to the above-mentioned use of the time domain sliding window of the first length to obtain the at least two test sample points; the model monitoring accuracy is based on the statistical results of the at least two test sample points). The sliding window length K is predefined by the protocol, configured on the network side (for example, based on RRC configuration), or a mixture of protocol predefined and network side configuration (corresponding to the above-mentioned first length is determined according to protocol predefined and / or network side configuration). The sliding step size Nstep of the sliding window in the time domain is configured as predefined by the protocol, configured on the network side (for example, based on RRC configuration), or a mixture of protocol predefined and network side configuration (corresponding to the above-mentioned sliding step size of the time domain sliding window is determined according to protocol predefined and / or network side configuration).

[0382] by Figure 4 For example, 1 test sample point refers to: 1 intermediate KPI calculated based on 1 target CSI and 1 corresponding CSI restored by the UE side using the alternative model. Among them, the example of intermediate KPI acquisition time, sliding window length K, sliding window sliding step length Nstep is similar to the method 1, and will not be repeated here.

[0383] Part 2: The UE side defines AI / ML model monitoring events Event3 (for example, the UE side AI / ML model monitoring accuracy is greater than a certain threshold) and Event4 (for example, the UE side AI / ML model monitoring accuracy is less than a certain threshold). The definitions of Event3 and Event4 are the same as those in Part 2 of Method 2, and will not be repeated here, but the values ​​of the configuration parameters may be different.

[0384] Part 3: Define the conditions for triggering AI / ML model monitoring reporting by AI / ML model monitoring events Event3 and / or Event4 (corresponding to the above-mentioned monitoring reporting conditions). The conditions for triggering AI / ML model monitoring reporting are related to the entry and exit of Event3 (and / or Event4) monitored on the UE side, the threshold parameters for the occurrence of monitoring events Event3 (and / or Event4) configured on the network side (for example, the maximum or minimum number of occurrences of the event, the maximum or minimum number of consecutive occurrences of the event, etc.), and the timer parameters corresponding to the monitoring event Event3 (and / or Event4) (corresponding to the above-mentioned reporting timing duration).

[0385] Implementation example 4: UE-side AI / ML model monitoring (based on the alternative model of the AI / ML model of the network-side CSI recovery part existing on the UE side), determining that the currently used AI / ML model M1 is invalid (the AI / ML model M1 here may refer to a pair of models of UE-side CSI generation and UE-side CSI recovery, that is, the AI / ML model on the UE side and the alternative model of the AI / ML model of the network-side CSI recovery part existing on the UE side);

[0386] Among them, (1) the relevant parts of this implementation example are the same as those of implementation example 3 and will not be repeated here.

[0387] (2) The above-mentioned implementation example 4 may also have variations similar to the above-mentioned implementation examples 1a-1c, that is, a combination of conditions such as different event entry and / or exit conditions, maximum or minimum number of event (continuous) occurrences, counter initialization or reset to zero, etc. is used to define the triggering of monitoring reports, which will not be repeated here.

[0388] Implementation example 5a: UE-side AI / ML model monitoring (based on the alternative model of the network-side CSI recovery part AI / ML model existing on the UE side), determining the switch of the currently used UE-side CSI generation AI / ML model;

[0389] The UE side monitors the currently used UE-side CSI generation AI / ML model D1, and simultaneously monitors the unused, potential UE-side CSI generation AI / ML model D2 (because the UE side also has the CSI generation model D2, the target CSI can be obtained through model D2 to obtain the compressed CSI based on D2, and then the compressed CSI of D2 is obtained through the replacement model of the CSI recovery part AI / ML model to obtain the new CSI, and then the intermediate KPI is calculated with the target CSI). Here, it is explained that here one potential CSI generation model D2 (corresponding to the above-mentioned candidate model) is taken as an example, of course, it can also be applied to the case of multiple potential CSI generation models D3, D4, ..., DN.

[0390] The UE side configures the relevant parameters of Event3 and / or Event4 for D1 and D2 respectively. The parameter configuration is predefined by the protocol, configured by the network side (for example, based on RRC configuration), or a mixture of predefined by the protocol and configured by the network side. The UE side defines the maximum number of occurrences of monitoring event Event4 for D1, N max_count_E4 UE side defines the minimum occurrence number N of monitoring event Event3 for D2 min_count_E3 The above parameter configuration is predefined by the protocol, configured by the network side (for example, based on RRC configuration), or a mixture of predefined by the protocol and configured by the network side. max_count_E4 and N min_count_E3 (corresponding to the above-mentioned reporting threshold) associated with the failure timer (corresponding to the above-mentioned reporting timing duration), the above-mentioned parameter configuration is protocol predefined, network side configured (for example, based on RRC configuration), or a mixture of protocol predefined and network side configuration.

[0391] The UE side monitors the number of occurrences of D1-related Event4, where M NW According to the sliding window calculation described above, when the UE side monitors the entry condition (Condition4-1) of Event4 based on the current sliding window calculation, the UE side records the number of occurrences of the current Event4 plus 1; when the UE side monitors the exit condition (Condition4-2) of Event4 based on the current sliding window calculation, the UE side records the number of occurrences of the current Event4 minus 1.

[0392] The UE side monitors the number of occurrences of D2-related Event3, where M NWAccording to the sliding window calculation described above, when the UE side monitors the entry condition (Condition3-1) of Event3 based on the current sliding window calculation, the UE side records the number of occurrences of the current Event3 plus 1; when the UE side monitors the exit condition (Condition3-2) of Event3 based on the current sliding window calculation, the UE side records the number of occurrences of the current Event3 minus 1.

[0393] When the invalidation timer configured for D1 has not yet reached zero (countdown) or reached the upper limit (countup), and the number of occurrences of the current Event 4 of D1 recorded on the UE side is greater than the configured N max_count_E4 , and the invalidation timer configured for D2 has not yet returned to zero (countdown) or reached the upper limit (countup), and the number of occurrences of D2's current Event3 recorded on the network side is greater than its configured N min_count_E3 When the UE side triggers the AI / ML model monitoring report, reports that the performance of the currently used CSI generated AI / ML model D1 is worse than that of the CSI generated AI / ML model D2, and D2 is reported as a candidate switching model to request the network to make a decision; corresponding to the above-mentioned operation of triggering the model monitoring report when the first monitoring event is that the monitoring accuracy of the first model does not meet the use condition and meets the first condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model meets the use condition and meets the second condition; the reported content is used to indicate model switching and carries the information of the candidate model; wherein the first condition includes: the monitoring accuracy of the first model meets the use condition within the first reporting timing duration The third number of times the monitoring accuracy of the first model meets the entry condition of the first monitoring event is greater than the first reporting threshold; the third number is determined according to the real-time number of times the monitoring accuracy of the first model meets the entry condition of the first monitoring event and the real-time number of times the monitoring accuracy of the first model meets the exit condition of the first monitoring event; the second condition includes: within the second reporting timing period, the fourth number of times the monitoring accuracy of the second model meets the entry condition of the second monitoring event is greater than the second reporting threshold; the fourth number is determined according to the real-time number of times the monitoring accuracy of the second model meets the entry condition of the second monitoring event and the real-time number of times the monitoring accuracy of the second model meets the exit condition of the second monitoring event.

[0394] Among them: the above-mentioned implementation example 5a can also have variants similar to the above-mentioned implementation examples 1a-1c, that is, using a combination of different events and / or entry and exit conditions, maximum or minimum number of occurrences of events (continuous), counter initialization or zeroing, etc. to define the triggering of monitoring reports, which will not be repeated here.

[0395] Implementation example 5b: UE-side AI / ML model monitoring (based on the alternative model of the network-side CSI recovery part AI / ML model existing on the UE side), based on Event3 and Event4 entry events, determine the current UE-side CSI generation AI / ML model switch.

[0396] This implementation example can be used as a special case of implementation example 5a. max_count_E4 and D2's N min_count_E3 Set them to 1, and set the threshold thresh in the entry condition of D1 Event4 E4in and the threshold thresh in the entry condition of D2 Event3 E3in For joint configuration, the combined event "Event4 meets the entry condition and Event3 meets the entry condition" is equivalent to the event "The monitoring accuracy of D2 is greater than the monitoring accuracy of D1 by X = thresh E3in -thresh E4in ". Using the above configuration, it can be set to trigger monitoring reporting when the potential AI / ML model has a monitoring accuracy better than the current model by X, report the monitoring accuracy of the candidate model D2 and / or each model, and request the network to make a decision (corresponding to the above operation of triggering model monitoring reporting when the monitoring event is that the model monitoring accuracy of the currently used model is less than the model monitoring accuracy of any candidate model; wherein the reported content is used to indicate model switching and carries the information of the candidate model).

[0397] It is to be noted that the relevant contents of the above-mentioned exemplary solutions can refer to each other, and the repeated parts will not be repeated. In addition, the numerical settings of the above-mentioned thresholds and / or thresholds can be the same or different, which is not limited here.

[0398] From the above, the solution provided by the embodiment of the present application has the following advantages:

[0399] 1. AI / ML model monitoring based on sliding windows allows the monitoring subject to continuously monitor the AI / ML model on the timeline, without the need for the network side to use signaling (Media Access Control-Control Element (MAC CE) or Downlink Control Information (DCI)) to trigger the initialization of AI / ML model monitoring, saving signaling overhead. In particular, when the monitoring subject is the UE side, a preferred example of this solution is that the network side initiates RRC configuration signaling to the UE to configure sliding window related parameters (monitoring configuration); and after receiving the RRC configuration, the UE will initiate perpetual, continuous, sliding window-based AI / ML model monitoring.

[0400] 2. The AI / ML model monitoring calculation based on sliding windows can be more flexible. Different sliding window sizes and sliding window steps can be configured for different UE capabilities (such as computing power, monitoring accuracy requirements, monitoring immediacy requirements, etc.), so that the sliding window calculation is not too frequent, or the change in monitoring accuracy calculated after the sliding window is more significant. That is, the window length and sliding window step can be flexibly configured and jointly optimized according to the needs and the computing power of the terminal.

[0401] 3. Defining information related to monitoring events allows the AI / ML model monitoring subject to determine the triggering of AI / ML model monitoring reports based on the entry and exit of monitoring events, as well as threshold parameters related to monitoring events (for example, the maximum or minimum number of occurrences of monitoring events, the maximum or minimum number of consecutive occurrences of monitoring events, etc.). This can effectively reduce disturbances and misjudgments, and solve the ping-pong effect of frequent switching of AI / ML model monitoring.

[0402] The embodiment of the present application also provides an information processing device, wherein the information processing device is a first device, such as Figure 8 As shown, it includes a memory 81, a transceiver 82, and a processor 83:

[0403] The memory 81 is used to store computer programs; the transceiver 82 is used to send and receive data under the control of the processor 83; the processor 83 is used to read the computer program in the memory 81 and perform the following operations:

[0404] Acquire a model monitoring accuracy according to at least two test sample points; the test sample points are determined according to target channel state information CSI and CSI corresponding to the target CSI and recovered by the model; the model monitoring accuracy is based on statistical results of the at least two test sample points;

[0405] Determine monitoring event related information for the monitoring accuracy of the model; the monitoring event related information includes: a definition of the monitoring event, and an entry condition and / or an exit condition of the monitoring event;

[0406] Controlling the triggering operation of model monitoring reporting according to the model monitoring accuracy and monitoring event related information;

[0407] The first device is a network device or a terminal.

[0408] The information processing device provided in the embodiment of the present application obtains the model monitoring accuracy based on at least two test sample points; the test sample points are determined based on the target channel state information CSI and the CSI corresponding to the target CSI and obtained by model recovery; the model monitoring accuracy is based on the statistical results of the at least two test sample points; the monitoring event related information for the model monitoring accuracy is determined; the monitoring event related information includes: the definition of the monitoring event, and the entry condition and / or exit condition of the monitoring event; according to the model monitoring accuracy and the monitoring event related information, the triggering operation of the model monitoring report is controlled; wherein the first device is a network device or a terminal; it can support the control of the triggering of the model monitoring report according to the model monitoring accuracy (i.e., the monitoring result), so as to support the accurate implementation of the decision-making operations such as switching, fallback, updating, activation and / or deactivation of the model, and well solves the problem that the information processing scheme for model monitoring in the prior art cannot support the accurate implementation of the model decision-making operation.

[0409] Specifically, the transceiver 82 is used to receive and send data under the control of the processor 83.

[0410] Among them, Figure 8 In the embodiment, the bus architecture may include any number of interconnected buses and bridges, specifically one or more processors represented by processor 83 and various circuits of memory represented by memory 81 are linked together. The bus architecture may also link together various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface. The transceiver 82 may be a plurality of components, namely, a transmitter and a receiver, providing a unit for communicating with various other devices on a transmission medium, which transmission medium may include a wireless channel, a wired channel, an optical cable, and other transmission media. The processor 83 is responsible for managing the bus architecture and general processing, and the memory 81 may store data used by the processor 83 when performing operations.

[0411] The processor 83 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or a complex programmable logic device (CPLD). The processor may also adopt a multi-core architecture.

[0412] The processor calls the computer program stored in the memory to execute any of the methods provided in the embodiments of the present application according to the obtained executable instructions. The processor and the memory can also be arranged physically separately.

[0413] It is explained here that when the first device is implemented as a terminal, it can also include a user interface 84 that can communicate with the bus interface; for different user devices, the user interface 84 can also be an interface that can connect to required devices externally or internally, and the connected devices include but are not limited to keypads, displays, speakers, microphones, joysticks, etc.

[0414] Among them, the triggering operation of controlling the model monitoring report based on the model monitoring accuracy and the information related to the monitoring event includes: triggering the model monitoring reporting operation when the monitoring event is that the model monitoring accuracy does not meet the usage conditions, and the model monitoring accuracy meets the entry conditions of the monitoring event; or triggering the model monitoring reporting operation when the monitoring event is that the model monitoring accuracy meets the usage conditions, and the model monitoring accuracy meets the exit conditions of the monitoring event.

[0415] Further, the operation also includes: determining monitoring reporting conditions; the monitoring reporting conditions include: reporting threshold and / or reporting timing duration; the operation of triggering model monitoring reporting when the monitoring event is that the model monitoring accuracy does not meet the usage conditions, and the model monitoring accuracy meets the entry conditions of the monitoring event, includes: (1) when the monitoring event is that the model monitoring accuracy does not meet the usage conditions, and the first number of times that the model monitoring accuracy meets the entry conditions of the monitoring event within the reporting timing duration is greater than the reporting threshold, triggering the operation of model monitoring reporting; wherein the reported content is used to indicate that the currently used model is invalid; the first number is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event, or the first number is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event and the real-time number of times that the model monitoring accuracy meets the exit conditions of the monitoring event, or the first number refers to the model The number of consecutive times that the monitoring accuracy satisfies the entry condition of the monitoring event; and / or, (2) when the monitoring event is that the model monitoring accuracy satisfies the use condition and the model monitoring accuracy satisfies the exit condition of the monitoring event, the operation of triggering the model monitoring reporting comprises: when the monitoring event is that the model monitoring accuracy satisfies the use condition and the second number of times that the model monitoring accuracy satisfies the exit condition of the monitoring event within the reporting timing is greater than the reporting threshold, the operation of triggering the model monitoring reporting; wherein the content of the report is used to indicate that the currently used model is invalid; the second number is determined according to the real-time number of times that the model monitoring accuracy satisfies the exit condition of the monitoring event, or the second number is determined according to the real-time number of times that the model monitoring accuracy satisfies the exit condition of the monitoring event and the real-time number of times that the model monitoring accuracy satisfies the entry condition of the monitoring event, or the second number refers to the consecutive number of times that the model monitoring accuracy satisfies the exit condition of the monitoring event.

[0416] In an embodiment of the present application, the model includes a currently used model and at least one candidate model; obtaining the model monitoring accuracy based on at least two test sample points includes: obtaining a first model monitoring accuracy for the currently used model based on at least two test sample points corresponding to the currently used model; and obtaining a second model monitoring accuracy for the candidate model based on at least two test sample points corresponding to the candidate model.

[0417] Among them, the monitoring events include: a first monitoring event for the currently used model, and a second monitoring event for the candidate model; the reporting threshold includes: a first reporting threshold for the currently used model, and a second reporting threshold for the candidate model; the reporting timing duration includes: a first reporting timing duration for the currently used model, and a second reporting timing duration for the candidate model.

[0418] In an embodiment of the present application, when the monitoring event is that the model monitoring accuracy does not meet the usage condition, and the model monitoring accuracy meets the entry condition of the monitoring event, the operation of triggering model monitoring reporting includes: (1) when the first monitoring event is that the first model monitoring accuracy does not meet the usage condition and meets the first condition; and the second monitoring event is that the second model monitoring accuracy of any candidate model meets the usage condition and meets the second condition, the operation of triggering model monitoring reporting is used to indicate model switching and carries information about the candidate model; or, (2) when the first monitoring event is that the first model monitoring accuracy does not meet the usage condition and meets the first condition; and the second monitoring event is that the second model monitoring accuracy of any candidate model does not meet the usage condition and meets the third condition, the operation of triggering model monitoring reporting is used to indicate model switching and carries information about the candidate model; wherein (1) the first condition includes: the third number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event within the first reporting timing duration is greater than the first reporting threshold; the third number is based on the first model monitoring accuracy meeting the entry condition of the first monitoring event. The third number is determined by the real-time number of times the first model monitoring accuracy meets the entry condition of the first monitoring event, or the third number is determined by the real-time number of times the first model monitoring accuracy meets the entry condition of the first monitoring event and the real-time number of times the first model monitoring accuracy meets the exit condition of the first monitoring event, or the third number refers to the consecutive number of times the first model monitoring accuracy meets the entry condition of the first monitoring event; (2) the second condition includes: the fourth number of times the second model monitoring accuracy meets the entry condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the fourth number is determined by the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event, or the fourth number is determined by the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event, or the fourth number refers to the consecutive number of times the second model monitoring accuracy meets the entry condition of the second monitoring event; (3) the third condition includes: the fifth number of times the second model monitoring accuracy meets the exit condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold;The fifth number is determined according to the real-time number of times the second model monitoring accuracy satisfies the exit condition of the second monitoring event, or the fifth number is determined according to the real-time number of times the second model monitoring accuracy satisfies the exit condition of the second monitoring event and the real-time number of times the second model monitoring accuracy satisfies the entry condition of the second monitoring event, or the fifth number refers to the consecutive number of times the second model monitoring accuracy satisfies the exit condition of the second monitoring event. ;

[0419] Wherein, when the monitoring event is that the model monitoring accuracy meets the use condition and the model monitoring accuracy meets the exit condition of the monitoring event, the operation of triggering model monitoring reporting includes: (1) when the first monitoring event is that the first model monitoring accuracy meets the use condition and meets the fourth condition; and the second monitoring event is that the second model monitoring accuracy of any candidate model meets the use condition and meets the fifth condition, the operation of triggering model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model; or, (2) when the first monitoring event is that the first model monitoring accuracy meets the use condition and meets the fourth condition; and the second monitoring event is that the second model monitoring accuracy of any candidate model does not meet the use condition and meets the sixth condition, the operation of triggering model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model; wherein, (1) the fourth condition includes: the sixth number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event within the first reporting timing duration is greater than the first reporting threshold; the sixth number is the number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event according to the actual number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event The sixth number is determined according to the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event and the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event, or the sixth number refers to the consecutive number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event; (2) the fifth condition includes: the seventh number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the seventh number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event, or the seventh number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event, or the seventh number refers to the consecutive number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event; (3) the sixth condition includes: the eighth number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold;The eighth number is determined according to the real-time number of times the second model monitoring accuracy satisfies the exit condition of the second monitoring event, or the eighth number is determined according to the real-time number of times the second model monitoring accuracy satisfies the exit condition of the second monitoring event and the real-time number of times the second model monitoring accuracy satisfies the entry condition of the second monitoring event, or the eighth number refers to the consecutive number of times the second model monitoring accuracy satisfies the exit condition of the second monitoring event. ;

[0420] Furthermore, the triggering operation of controlling the model monitoring report based on the model monitoring accuracy and the information related to the monitoring event also includes: triggering the model monitoring reporting operation when the monitoring event is that the model monitoring accuracy of the currently used model is less than the model monitoring accuracy of any candidate model; wherein the reported content is used to indicate model switching and carries the information of the candidate model.

[0421] In the embodiment of the present application, the operation further includes: using a time domain sliding window of a first length to obtain the at least two test sample points.

[0422] Furthermore, the operation also includes: before obtaining the model monitoring accuracy based on the at least two test sample points, preprocessing the at least two test sample points; wherein the preprocessing includes: at least one of weighted processing and abnormal elimination processing.

[0423] The entry condition and / or exit condition is determined according to a blocking parameter and a corresponding threshold value.

[0424] In an embodiment of the present application, the monitoring event related information is determined according to the protocol predefinition and / or the network side configuration; and / or, the monitoring reporting condition is determined according to the protocol predefinition and / or the network side configuration; and / or, the first length is determined according to the protocol predefinition and / or the network side configuration; and / or, the sliding step of the time domain sliding window is determined according to the protocol predefinition and / or the network side configuration; and / or, the blocking parameter is determined according to the protocol predefinition and / or the network side configuration; and / or, the threshold value is determined according to the protocol predefinition and / or the network side configuration.

[0425] It should be noted here that the above-mentioned device provided in the embodiment of the present application can implement all the method steps implemented in the above-mentioned method embodiment, and can achieve the same technical effect. The parts and beneficial effects of this embodiment that are the same as the method embodiment will not be described in detail here.

[0426] The embodiment of the present application also provides an information processing device, wherein the information processing device is a first device, such as Fig. 9 As shown, it includes a memory 91, a transceiver 92, and a processor 93:

[0427] The memory 91 is used to store computer programs; the transceiver 92 is used to send and receive data under the control of the processor 93; the processor 93 is used to read the computer program in the memory 91 and perform the following operations:

[0428] Using a time domain sliding window of a first length to obtain at least two test sample points;

[0429] The test sample point is determined according to a target CSI and a CSI corresponding to the target CSI and restored by a model;

[0430] The first device is a network device or a terminal.

[0431] The information processing device provided in the embodiment of the present application obtains at least two test sample points by adopting a time domain sliding window of a first length; wherein the test sample points are determined according to the target CSI and the CSI corresponding to the target CSI and obtained by model recovery; the first device is a network device or a terminal; it can support accurate acquisition of test sample points, thereby improving the accuracy of the acquired model monitoring accuracy, and further support accurate control of model monitoring reporting triggering according to the model monitoring accuracy (i.e., monitoring results), thereby supporting accurate implementation of decision operations such as model switching, fallback, update, activation and / or deactivation, and well solving the problem that the information processing scheme for model monitoring in the prior art cannot support accurate implementation of model decision operations.

[0432] Specifically, the transceiver 92 is used to receive and send data under the control of the processor 93.

[0433] Among them, Fig. 9 In the embodiment, the bus architecture may include any number of interconnected buses and bridges, specifically one or more processors represented by processor 93 and various circuits of memory represented by memory 91 are linked together. The bus architecture may also link together various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface. The transceiver 92 may be a plurality of components, namely, a transmitter and a receiver, providing a unit for communicating with various other devices on a transmission medium, such as a wireless channel, a wired channel, an optical cable, and the like. The processor 93 is responsible for managing the bus architecture and general processing, and the memory 91 may store data used by the processor 93 when performing operations.

[0434] The processor 93 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or a complex programmable logic device (CPLD). The processor may also adopt a multi-core architecture.

[0435] The processor calls the computer program stored in the memory to execute any of the methods provided in the embodiments of the present application according to the obtained executable instructions. The processor and the memory can also be arranged physically separately.

[0436] It is explained here that when the first device is implemented as a terminal, it can also include a user interface 94 that can communicate with the bus interface; for different user devices, the user interface 94 can also be an interface that can be externally or internally connected to required devices, and the connected devices include but are not limited to keypads, displays, speakers, microphones, joysticks, etc.

[0437] Furthermore, the operation also includes: preprocessing the at least two test sample points; wherein the preprocessing includes: at least one of weighted processing and abnormal elimination processing.

[0438] In an embodiment of the present application, the first length is determined according to protocol pre-definition and / or network side configuration; and / or, the sliding step size of the time domain sliding window is determined according to protocol pre-definition and / or network side configuration.

[0439] It should be noted here that the above-mentioned device provided in the embodiment of the present application can implement all the method steps implemented in the above-mentioned method embodiment, and can achieve the same technical effect. The parts and beneficial effects of this embodiment that are the same as the method embodiment will not be described in detail here.

[0440] The present application also provides an information processing device, which is applied to a first device, such as Fig.10 As shown, including:

[0441] A first acquisition unit 101 is configured to acquire a model monitoring accuracy according to at least two test sample points; the test sample points are determined according to target channel state information CSI and CSI corresponding to the target CSI and recovered by the model; the model monitoring accuracy is based on statistical results of the at least two test sample points;

[0442] A first determining unit 102 is used to determine monitoring event related information for the model monitoring accuracy; the monitoring event related information includes: a definition of a monitoring event, and an entry condition and / or an exit condition of the monitoring event;

[0443] A first control unit 103, configured to control a triggering operation of a model monitoring report according to the model monitoring accuracy and monitoring event related information;

[0444] The first device is a network device or a terminal.

[0445] The information processing device provided in the embodiment of the present application obtains the model monitoring accuracy based on at least two test sample points; the test sample points are determined based on the target channel state information CSI and the CSI corresponding to the target CSI and obtained by model recovery; the model monitoring accuracy is based on the statistical results of the at least two test sample points; the monitoring event related information for the model monitoring accuracy is determined; the monitoring event related information includes: the definition of the monitoring event, and the entry condition and / or exit condition of the monitoring event; according to the model monitoring accuracy and the monitoring event related information, the triggering operation of the model monitoring report is controlled; wherein the first device is a network device or a terminal; it can support the control of the triggering of the model monitoring report according to the model monitoring accuracy (i.e., the monitoring result), so as to support the accurate implementation of the decision-making operations such as switching, fallback, updating, activation and / or deactivation of the model, and well solves the problem that the information processing scheme for model monitoring in the prior art cannot support the accurate implementation of the model decision-making operation.

[0446] Among them, the triggering operation of controlling the model monitoring report based on the model monitoring accuracy and the information related to the monitoring event includes: triggering the model monitoring reporting operation when the monitoring event is that the model monitoring accuracy does not meet the usage conditions, and the model monitoring accuracy meets the entry conditions of the monitoring event; or triggering the model monitoring reporting operation when the monitoring event is that the model monitoring accuracy meets the usage conditions, and the model monitoring accuracy meets the exit conditions of the monitoring event.

[0447] Furthermore, the information processing device further includes: a second determination unit, configured to determine monitoring reporting conditions; the monitoring reporting conditions include: a reporting threshold and / or a reporting timing duration; (1) when the monitoring event is that the model monitoring accuracy does not meet the use conditions, and the model monitoring accuracy meets the entry conditions of the monitoring event, the operation of triggering model monitoring reporting includes: when the monitoring event is that the model monitoring accuracy does not meet the use conditions, and the first number of times that the model monitoring accuracy meets the entry conditions of the monitoring event within the reporting timing duration is greater than the reporting threshold, the operation of triggering model monitoring reporting; wherein the content of the report is used to indicate that the currently used model is invalid; the first number is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event, or the first number is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event and the real-time number of times that the model monitoring accuracy meets the exit conditions of the monitoring event, or the first number is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event and the real-time number of times that the model monitoring accuracy meets the exit conditions of the monitoring event, or the first number is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event The number refers to the consecutive number of times that the model monitoring accuracy meets the entry condition of the monitoring event; and / or, (2) the operation of triggering the model monitoring reporting when the monitoring event is that the model monitoring accuracy meets the use condition and the model monitoring accuracy meets the exit condition of the monitoring event, including: triggering the model monitoring reporting operation when the monitoring event is that the model monitoring accuracy meets the use condition and the second number of times that the model monitoring accuracy meets the exit condition of the monitoring event within the reporting timing is greater than the reporting threshold; wherein the reported content is used to indicate that the currently used model is invalid; the second number is determined based on the real-time number of times that the model monitoring accuracy meets the exit condition of the monitoring event, or the second number is determined based on the real-time number of times that the model monitoring accuracy meets the exit condition of the monitoring event and the real-time number of times that the model monitoring accuracy meets the entry condition of the monitoring event, or the second number refers to the consecutive number of times that the model monitoring accuracy meets the exit condition of the monitoring event.

[0448] Wherein, the model includes a currently used model and at least one candidate model; the obtaining of the model monitoring accuracy based on at least two test sample points includes: obtaining a first model monitoring accuracy for the currently used model based on at least two test sample points corresponding to the currently used model; and obtaining a second model monitoring accuracy for the candidate model based on at least two test sample points corresponding to the candidate model.

[0449] In an embodiment of the present application, the monitoring event includes: a first monitoring event for the currently used model, and a second monitoring event for the candidate model; the reporting threshold includes: a first reporting threshold for the currently used model, and a second reporting threshold for the candidate model; the reporting timing duration includes: a first reporting timing duration for the currently used model, and a second reporting timing duration for the candidate model.

[0450] Wherein, when the monitoring event is that the model monitoring accuracy does not meet the use condition, and the model monitoring accuracy meets the entry condition of the monitoring event, the operation of triggering model monitoring reporting includes: (1) when the first monitoring event is that the first model monitoring accuracy does not meet the use condition and meets the first condition; and the second monitoring event is that the second model monitoring accuracy of any candidate model meets the use condition and meets the second condition, the operation of triggering model monitoring reporting is used to indicate model switching and carries the information of the candidate model; or, (2) when the first monitoring event is that the first model monitoring accuracy does not meet the use condition and meets the first condition; and the second monitoring event is that the second model monitoring accuracy of any candidate model does not meet the use condition and meets the third condition, the operation of triggering model monitoring reporting is used to indicate model switching and carries the information of the candidate model; wherein, (1) the first condition includes: the third number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event within the first reporting timing duration is greater than the first reporting threshold; the third number is based on the first model monitoring accuracy meeting the entry condition of the first monitoring event. The third number is determined by the real-time number of times the first model monitoring accuracy meets the entry condition of the first monitoring event and the real-time number of times the first model monitoring accuracy meets the exit condition of the first monitoring event, or the third number refers to the consecutive number of times the first model monitoring accuracy meets the entry condition of the first monitoring event; (2) the second condition includes: the fourth number of times the second model monitoring accuracy meets the entry condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the fourth number is determined by the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event, or the fourth number is determined by the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event, or the fourth number refers to the consecutive number of times the second model monitoring accuracy meets the entry condition of the second monitoring event; (3) the third condition includes: the fifth number of times the second model monitoring accuracy meets the exit condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold;The fifth number is determined according to the real-time number of times the second model monitoring accuracy satisfies the exit condition of the second monitoring event, or the fifth number is determined according to the real-time number of times the second model monitoring accuracy satisfies the exit condition of the second monitoring event and the real-time number of times the second model monitoring accuracy satisfies the entry condition of the second monitoring event, or the fifth number refers to the consecutive number of times the second model monitoring accuracy satisfies the exit condition of the second monitoring event. ;

[0451] In an embodiment of the present application, when the monitoring event is that the model monitoring accuracy meets the use condition, and the model monitoring accuracy meets the exit condition of the monitoring event, the operation of triggering model monitoring reporting includes: (1) when the first monitoring event is that the first model monitoring accuracy meets the use condition and meets the fourth condition; and the second monitoring event is that the second model monitoring accuracy of any candidate model meets the use condition and meets the fifth condition, the operation of triggering model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model; or, (2) when the first monitoring event is that the first model monitoring accuracy meets the use condition and meets the fourth condition; and the second monitoring event is that the second model monitoring accuracy of any candidate model does not meet the use condition and meets the sixth condition, the operation of triggering model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model; wherein (1) the fourth condition includes: the sixth number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event within the first reporting timing duration is greater than the first reporting threshold; the sixth number is based on the first model monitoring accuracy meeting the exit condition of the first monitoring event. The sixth number is determined according to the real-time number of times the first model monitoring accuracy meets the exit condition of the first monitoring event and the real-time number of times the first model monitoring accuracy meets the entry condition of the first monitoring event, or the sixth number refers to the consecutive number of times the first model monitoring accuracy meets the entry condition of the first monitoring event; (2) the fifth condition includes: the seventh number of times the second model monitoring accuracy meets the entry condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the seventh number is determined according to the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event, or the seventh number is determined according to the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event, or the seventh number refers to the consecutive number of times the second model monitoring accuracy meets the entry condition of the second monitoring event; (3) the sixth condition includes: the eighth number of times the second model monitoring accuracy meets the exit condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold;The eighth number is determined according to the real-time number of times the second model monitoring accuracy satisfies the exit condition of the second monitoring event, or the eighth number is determined according to the real-time number of times the second model monitoring accuracy satisfies the exit condition of the second monitoring event and the real-time number of times the second model monitoring accuracy satisfies the entry condition of the second monitoring event, or the eighth number refers to the consecutive number of times the second model monitoring accuracy satisfies the exit condition of the second monitoring event. ;

[0452] Furthermore, the triggering operation of controlling the model monitoring report based on the model monitoring accuracy and the information related to the monitoring event also includes: triggering the model monitoring reporting operation when the monitoring event is that the model monitoring accuracy of the currently used model is less than the model monitoring accuracy of any candidate model; wherein the reported content is used to indicate model switching and carries the information of the candidate model.

[0453] In the embodiment of the present application, the information processing device further includes: a second acquisition unit, configured to acquire the at least two test sample points by adopting a time domain sliding window of a first length.

[0454] Furthermore, the information processing device also includes: a first processing unit, used to preprocess the at least two test sample points before obtaining the model monitoring accuracy based on the at least two test sample points; wherein the preprocessing includes: at least one of weighted processing and abnormal elimination processing.

[0455] The entry condition and / or exit condition is determined according to a blocking parameter and a corresponding threshold value.

[0456] In an embodiment of the present application, the monitoring event related information is determined according to the protocol predefinition and / or the network side configuration; and / or, the monitoring reporting condition is determined according to the protocol predefinition and / or the network side configuration; and / or, the first length is determined according to the protocol predefinition and / or the network side configuration; and / or, the sliding step of the time domain sliding window is determined according to the protocol predefinition and / or the network side configuration; and / or, the blocking parameter is determined according to the protocol predefinition and / or the network side configuration; and / or, the threshold value is determined according to the protocol predefinition and / or the network side configuration.

[0457] It should be noted here that the above-mentioned device provided in the embodiment of the present application can implement all the method steps implemented in the above-mentioned method embodiment, and can achieve the same technical effect. The parts and beneficial effects of this embodiment that are the same as those in the method embodiment will not be described in detail here.

[0458] The present application also provides an information processing device, which is applied to a first device, such as Fig.11 As shown, including:

[0459] A third acquisition unit 111 is used to acquire at least two test sample points by using a time domain sliding window of a first length;

[0460] The test sample point is determined according to a target CSI and a CSI corresponding to the target CSI and restored by a model;

[0461] The first device is a network device or a terminal.

[0462] The information processing device provided in the embodiment of the present application obtains at least two test sample points by adopting a time domain sliding window of a first length; wherein the test sample points are determined according to the target CSI and the CSI corresponding to the target CSI and obtained by model recovery; the first device is a network device or a terminal; it can support accurate acquisition of test sample points, thereby improving the accuracy of the acquired model monitoring accuracy, and further support accurate control of model monitoring reporting triggering according to the model monitoring accuracy (i.e., monitoring results), thereby supporting accurate implementation of decision operations such as model switching, fallback, update, activation and / or deactivation, and well solving the problem that the information processing scheme for model monitoring in the prior art cannot support accurate implementation of model decision operations.

[0463] Furthermore, the information processing device further includes: a second processing unit, configured to pre-process the at least two test sample points; wherein the pre-processing includes: at least one of weighted processing and abnormal elimination processing.

[0464] In an embodiment of the present application, the first length is determined according to protocol pre-definition and / or network side configuration; and / or, the sliding step size of the time domain sliding window is determined according to protocol pre-definition and / or network side configuration.

[0465] It should be noted here that the above-mentioned device provided in the embodiment of the present application can implement all the method steps implemented in the above-mentioned method embodiment, and can achieve the same technical effect. The parts and beneficial effects of this embodiment that are the same as those in the method embodiment will not be described in detail here.

[0466] It should be noted that the division of units in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional unit in each embodiment of the present application may be integrated into a processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0467] If 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 processor-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store program codes.

[0468] An embodiment of the present application further provides a non-transitory readable storage medium, wherein the non-transitory readable storage medium stores a computer program, and the computer program is used to enable a processor to execute the above method.

[0469] The non-transitory readable storage medium can be any available medium or data storage device that can be accessed by the processor, including but not limited to magnetic storage (such as floppy disks, hard disks, magnetic tapes, magneto-optical (MO)), etc.), optical storage (such as compact discs (CD), digital video discs (DVD), Blu-ray discs (BD), high-definition versatile discs (HVD), etc.), and semiconductor memories (such as ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read only memory (EEPROM), non-volatile memory (NAND (Non-volatile Memory Device) FLASH), solid-state drives (SSD)), etc.

[0470] Among them, the implementation embodiments of the above-mentioned information processing method are all applicable to the embodiments of the non-transitory readable storage medium, and can also achieve the same technical effect.

[0471] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) that contain computer-usable program code.

[0472] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer executable instructions. These computer executable instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0473] These processor executable instructions may also be stored in a processor readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the processor readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0474] These processor-executable instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0475] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. An information processing method, applied to a first device, characterized in that: include: Obtaining the model monitoring accuracy based on at least two test sample points; The test sample points are determined according to target channel state information CSI and CSI corresponding to the target CSI and obtained by model recovery; the model monitoring accuracy is based on the statistical results of the at least two test sample points; Determining monitoring event related information for monitoring accuracy of the model; The monitoring event related information includes: a definition of the monitoring event, and an entry condition and / or an exit condition of the monitoring event; Controlling the triggering operation of model monitoring reporting according to the model monitoring accuracy and monitoring event related information; The first device is a network device or a terminal.

2. The information processing method according to claim 1, characterized in that: The triggering operation of controlling the model monitoring report according to the model monitoring accuracy and the monitoring event related information includes: When the monitoring event is that the model monitoring accuracy does not meet the use condition, and the model monitoring accuracy meets the entry condition of the monitoring event, triggering the operation of model monitoring reporting; Alternatively, when the monitoring event is that the model monitoring accuracy meets the use condition, and the model monitoring accuracy meets the exit condition of the monitoring event, the model monitoring reporting operation is triggered.

3. The information processing method according to claim 2, characterized in that: Also includes: Determine monitoring and reporting conditions; The monitoring reporting conditions include: reporting threshold and / or reporting timing duration; The operation of triggering model monitoring reporting when the monitoring event is that the model monitoring accuracy does not meet the use condition and the model monitoring accuracy meets the entry condition of the monitoring event includes: When the monitoring event is that the model monitoring accuracy does not meet the usage conditions, and the first number of times that the model monitoring accuracy meets the entry conditions of the monitoring event within the reporting timing duration is greater than the reporting threshold, the model monitoring reporting operation is triggered; wherein the reported content is used to indicate that the currently used model is invalid; the first number of times is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event, or the first number of times is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event and the real-time number of times that the model monitoring accuracy meets the exit conditions of the monitoring event, or the first number of times refers to the consecutive number of times that the model monitoring accuracy meets the entry conditions of the monitoring event; And / or, when the monitoring event is that the model monitoring accuracy meets the use condition, and the model monitoring accuracy meets the exit condition of the monitoring event, the operation of triggering the model monitoring reporting includes: When the monitoring event is that the model monitoring accuracy meets the usage conditions, and the second number of times that the model monitoring accuracy meets the exit conditions of the monitoring event within the reporting timing period is greater than the reporting threshold, the model monitoring reporting operation is triggered; wherein the reported content is used to indicate that the currently used model is invalid; the second number of times is determined based on the real-time number of times the model monitoring accuracy meets the exit conditions of the monitoring event, or the second number of times is determined based on the real-time number of times the model monitoring accuracy meets the exit conditions of the monitoring event and the real-time number of times the model monitoring accuracy meets the entry conditions of the monitoring event, or the second number of times refers to the consecutive number of times the model monitoring accuracy meets the exit conditions of the monitoring event.

4. The information processing method according to claim 3, characterized in that: The model includes a currently used model and at least one candidate model; The obtaining of the model monitoring accuracy according to at least two test sample points includes: Acquire a first model monitoring accuracy for the currently used model according to at least two test sample points corresponding to the currently used model; According to at least two test sample points corresponding to the candidate model, a second model monitoring accuracy for the candidate model is obtained.

5. The information processing method according to claim 4, characterized in that: The monitoring event includes: a first monitoring event for the currently used model, and a second monitoring event for the candidate model; The reporting threshold includes: a first reporting threshold for the currently used model, and a second reporting threshold for the candidate model; The reporting timing duration includes: a first reporting timing duration for the currently used model, and a second reporting timing duration for the candidate model.

6. The information processing method according to claim 5, characterized in that: The operation of triggering model monitoring reporting when the monitoring event is that the model monitoring accuracy does not meet the use condition and the model monitoring accuracy meets the entry condition of the monitoring event includes: When the first monitoring event is that the monitoring accuracy of the first model does not meet the use condition and meets the first condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model meets the use condition and meets the second condition, trigger the operation of model monitoring reporting; the content of the report is used to indicate model switching and carries the information of the candidate model; Alternatively, when the first monitoring event is that the monitoring accuracy of the first model does not meet the use condition and meets the first condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model does not meet the use condition and meets the third condition, the operation of model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model; Among them, the first condition includes: the third number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event within the first reporting timing duration is greater than the first reporting threshold; the third number is determined according to the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event, or the third number is determined according to the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event and the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event, or the third number refers to the consecutive number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event; The second condition includes: the fourth number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the fourth number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event, or the fourth number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event, or the fourth number refers to the continuous number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event; The third condition includes: the fifth number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event within the second reporting timing period is greater than the second reporting threshold; the fifth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event, or the fifth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event and the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event, or the fifth number refers to the consecutive number of times the second model monitoring accuracy meets the exit condition of the second monitoring event.

7. The information processing method according to claim 5, characterized in that: The operation of triggering model monitoring reporting when the monitoring event is that the model monitoring accuracy meets the use condition and the model monitoring accuracy meets the exit condition of the monitoring event includes: When the first monitoring event is that the monitoring accuracy of the first model meets the use condition and the fourth condition is met; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model meets the use condition and the fifth condition is met, the operation of model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model; Alternatively, when the first monitoring event is that the monitoring accuracy of the first model meets the use conditions and meets the fourth condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model does not meet the use conditions and meets the sixth condition, the operation of model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model; Among them, the fourth condition includes: the sixth number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event within the first reporting timing duration is greater than the first reporting threshold; the sixth number is determined according to the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event, or the sixth number is determined according to the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event and the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event, or the sixth number refers to the consecutive number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event; The fifth condition includes: the seventh number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the seventh number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event, or the seventh number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event, or the seventh number refers to the consecutive number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event; The sixth condition includes: the eighth number of times the second model monitoring accuracy meets the exit condition of the second monitoring event within the second reporting timing period is greater than the second reporting threshold; the eighth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event, or the eighth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event and the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event, or the eighth number refers to the consecutive number of times the second model monitoring accuracy meets the exit condition of the second monitoring event.

8. The information processing method according to claim 2, characterized in that: The triggering operation of controlling the model monitoring report according to the model monitoring accuracy and the monitoring event related information further includes: When the monitoring event is that the model monitoring accuracy of the currently used model is less than the model monitoring accuracy of any candidate model, triggering the operation of model monitoring reporting; The reported content is used to instruct model switching and carries information about the candidate model.

9. The information processing method according to claim 1, characterized in that: Also includes: A time domain sliding window of a first length is used to acquire the at least two test sample points.

10. The information processing method according to claim 1, characterized in that: Before obtaining the model monitoring accuracy based on at least two test sample points, it also includes: Preprocessing the at least two test sample points; The preprocessing includes at least one of weighted processing and abnormal elimination processing.

11. The information processing method according to claim 1, characterized in that: The entry condition and / or exit condition is determined according to the blocking parameter and the corresponding threshold value.

12. The information processing method according to any one of claims 1 to 11, characterized in that: The monitoring event related information is determined according to protocol pre-definition and / or network side configuration; and / or, the monitoring reporting condition is predefined according to the protocol and / or determined by the network side configuration; and / or, the first length is predefined according to a protocol and / or determined according to a network side configuration; And / or, the sliding step size of the time domain sliding window is determined according to protocol pre-definition and / or network side configuration; and / or, the blocking parameter is determined according to protocol pre-definition and / or network side configuration; And / or, the threshold value is predefined according to the protocol and / or determined according to the network side configuration.

13. An information processing method, applied to a first device, characterized in that: include: Using a time domain sliding window of a first length to obtain at least two test sample points; The test sample point is determined according to a target CSI and a CSI corresponding to the target CSI and restored by a model; The first device is a network device or a terminal.

14. The information processing method according to claim 13, characterized in that: Also includes: Preprocessing the at least two test sample points; The preprocessing includes at least one of weighted processing and abnormal elimination processing.

15. The information processing method according to any one of claims 13 to 14, characterized in that: The first length is determined according to protocol pre-definition and / or network side configuration; And / or, the sliding step size of the time domain sliding window is predefined according to the protocol and / or determined according to the network side configuration.

16. An information processing device, the information processing device being a first device, characterized in that: Including memory, transceiver, processor: A memory for storing a computer program; a transceiver for transmitting and receiving data under the control of the processor; and a processor for reading the computer program in the memory and performing the following operations: Acquire a model monitoring accuracy according to at least two test sample points; the test sample points are determined according to target channel state information CSI and CSI corresponding to the target CSI and recovered by the model; the model monitoring accuracy is based on statistical results of the at least two test sample points; Determining monitoring event related information for monitoring accuracy of the model; The monitoring event related information includes: a definition of the monitoring event, and an entry condition and / or an exit condition of the monitoring event; Controlling the triggering operation of model monitoring reporting according to the model monitoring accuracy and monitoring event related information; The first device is a network device or a terminal.

17. The information processing device according to claim 16, characterized in that The triggering operation of controlling the model monitoring report according to the model monitoring accuracy and the monitoring event related information includes: When the monitoring event is that the model monitoring accuracy does not meet the use condition, and the model monitoring accuracy meets the entry condition of the monitoring event, triggering the operation of model monitoring reporting; Alternatively, when the monitoring event is that the model monitoring accuracy meets the use condition, and the model monitoring accuracy meets the exit condition of the monitoring event, the model monitoring reporting operation is triggered.

18. The information processing device according to claim 17, characterized in that: The operations also include: Determine monitoring and reporting conditions; the monitoring and reporting conditions include: reporting threshold and / or reporting timing duration; The operation of triggering model monitoring reporting when the monitoring event is that the model monitoring accuracy does not meet the use condition and the model monitoring accuracy meets the entry condition of the monitoring event includes: When the monitoring event is that the model monitoring accuracy does not meet the usage conditions, and the first number of times that the model monitoring accuracy meets the entry conditions of the monitoring event within the reporting timing duration is greater than the reporting threshold, the model monitoring reporting operation is triggered; wherein the reported content is used to indicate that the currently used model is invalid; the first number of times is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event, or the first number of times is determined based on the real-time number of times that the model monitoring accuracy meets the entry conditions of the monitoring event and the real-time number of times that the model monitoring accuracy meets the exit conditions of the monitoring event, or the first number of times refers to the consecutive number of times that the model monitoring accuracy meets the entry conditions of the monitoring event; And / or, when the monitoring event is that the model monitoring accuracy meets the use condition, and the model monitoring accuracy meets the exit condition of the monitoring event, the operation of triggering the model monitoring reporting includes: When the monitoring event is that the model monitoring accuracy meets the usage conditions, and the second number of times that the model monitoring accuracy meets the exit conditions of the monitoring event within the reporting timing period is greater than the reporting threshold, the model monitoring reporting operation is triggered; wherein the reported content is used to indicate that the currently used model is invalid; the second number of times is determined based on the real-time number of times the model monitoring accuracy meets the exit conditions of the monitoring event, or the second number of times is determined based on the real-time number of times the model monitoring accuracy meets the exit conditions of the monitoring event and the real-time number of times the model monitoring accuracy meets the entry conditions of the monitoring event, or the second number of times refers to the consecutive number of times the model monitoring accuracy meets the exit conditions of the monitoring event.

19. The information processing device according to claim 18, characterized in that The model includes a currently used model and at least one candidate model; The obtaining of the model monitoring accuracy according to at least two test sample points includes: Acquire a first model monitoring accuracy for the currently used model according to at least two test sample points corresponding to the currently used model; According to at least two test sample points corresponding to the candidate model, a second model monitoring accuracy for the candidate model is obtained.

20. The information processing device according to claim 19, characterized in that The monitoring event includes: a first monitoring event for the currently used model, and a second monitoring event for the candidate model; The reporting threshold includes: a first reporting threshold for the currently used model, and a second reporting threshold for the candidate model; The reporting timing duration includes: a first reporting timing duration for the currently used model, and a second reporting timing duration for the candidate model.

21. The information processing device according to claim 20, characterized in that The operation of triggering model monitoring reporting when the monitoring event is that the model monitoring accuracy does not meet the use condition and the model monitoring accuracy meets the entry condition of the monitoring event includes: When the first monitoring event is that the monitoring accuracy of the first model does not meet the use condition and meets the first condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model meets the use condition and meets the second condition, trigger the operation of model monitoring reporting; the content of the report is used to indicate model switching and carries the information of the candidate model; Alternatively, when the first monitoring event is that the monitoring accuracy of the first model does not meet the use condition and meets the first condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model does not meet the use condition and meets the third condition, the operation of model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model; Among them, the first condition includes: the third number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event within the first reporting timing duration is greater than the first reporting threshold; the third number is determined according to the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event, or the third number is determined according to the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event and the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event, or the third number refers to the consecutive number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event; The second condition includes: the fourth number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the fourth number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event, or the fourth number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event, or the fourth number refers to the continuous number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event; The third condition includes: the fifth number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event within the second reporting timing period is greater than the second reporting threshold; the fifth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event, or the fifth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event and the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event, or the fifth number refers to the consecutive number of times the second model monitoring accuracy meets the exit condition of the second monitoring event.

22. The information processing device according to claim 20, characterized in that The operation of triggering model monitoring reporting when the monitoring event is that the model monitoring accuracy meets the use condition and the model monitoring accuracy meets the exit condition of the monitoring event includes: When the first monitoring event is that the monitoring accuracy of the first model meets the use condition and the fourth condition is met; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model meets the use condition and the fifth condition is met, the operation of model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model; Alternatively, when the first monitoring event is that the monitoring accuracy of the first model meets the use conditions and meets the fourth condition; and the second monitoring event is that the monitoring accuracy of the second model of any candidate model does not meet the use conditions and meets the sixth condition, the operation of model monitoring reporting is triggered; the content of the report is used to indicate model switching and carries the information of the candidate model; Among them, the fourth condition includes: the sixth number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event within the first reporting timing duration is greater than the first reporting threshold; the sixth number is determined according to the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event, or the sixth number is determined according to the real-time number of times that the first model monitoring accuracy meets the exit condition of the first monitoring event and the real-time number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event, or the sixth number refers to the consecutive number of times that the first model monitoring accuracy meets the entry condition of the first monitoring event; The fifth condition includes: the seventh number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event within the second reporting timing duration is greater than the second reporting threshold; the seventh number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event, or the seventh number is determined according to the real-time number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event and the real-time number of times that the second model monitoring accuracy meets the exit condition of the second monitoring event, or the seventh number refers to the consecutive number of times that the second model monitoring accuracy meets the entry condition of the second monitoring event; The sixth condition includes: the eighth number of times the second model monitoring accuracy meets the exit condition of the second monitoring event within the second reporting timing period is greater than the second reporting threshold; the eighth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event, or the eighth number is determined based on the real-time number of times the second model monitoring accuracy meets the exit condition of the second monitoring event and the real-time number of times the second model monitoring accuracy meets the entry condition of the second monitoring event, or the eighth number refers to the consecutive number of times the second model monitoring accuracy meets the exit condition of the second monitoring event.

23. The information processing device according to claim 17, characterized in that The triggering operation of controlling the model monitoring report according to the model monitoring accuracy and the monitoring event related information further includes: When the monitoring event is that the model monitoring accuracy of the currently used model is less than the model monitoring accuracy of any candidate model, triggering the operation of model monitoring reporting; The reported content is used to instruct model switching and carries information about the candidate model.

24. The information processing device according to claim 16, characterized in that The operations also include: A time domain sliding window of a first length is used to acquire the at least two test sample points.

25. The information processing device according to claim 16, characterized in that The operations also include: Before obtaining the model monitoring accuracy according to the at least two test sample points, preprocessing the at least two test sample points; The preprocessing includes at least one of weighted processing and abnormal elimination processing.

26. The information processing device according to claim 16, characterized in that The entry condition and / or exit condition is determined according to the blocking parameter and the corresponding threshold value.

27. The information processing device according to any one of claims 16 to 26, characterized in that: The monitoring event related information is determined according to protocol pre-definition and / or network side configuration; and / or, the monitoring reporting condition is predefined according to the protocol and / or determined by the network side configuration; and / or, the first length is predefined according to a protocol and / or determined according to a network side configuration; And / or, the sliding step size of the time domain sliding window is determined according to protocol pre-definition and / or network side configuration; and / or, the blocking parameter is determined according to protocol pre-definition and / or network side configuration; And / or, the threshold value is predefined according to the protocol and / or determined according to the network side configuration.

28. An information processing device, the information processing device being a first device, characterized in that: Including memory, transceiver, processor: A memory for storing a computer program; a transceiver for transmitting and receiving data under the control of the processor; and a processor for reading the computer program in the memory and performing the following operations: Using a time domain sliding window of a first length to obtain at least two test sample points; The test sample point is determined according to a target CSI and a CSI corresponding to the target CSI and restored by a model; The first device is a network device or a terminal.

29. The information processing device according to claim 28, characterized in that The operations also include: Preprocessing the at least two test sample points; The preprocessing includes at least one of weighted processing and abnormal elimination processing.

30. The information processing device according to any one of claims 28 to 29, characterized in that: The first length is determined according to protocol pre-definition and / or network side configuration; And / or, the sliding step size of the time domain sliding window is predefined according to the protocol and / or determined according to the network side configuration.

31. An information processing device, applied to a first device, characterized in that: include: A first acquisition unit, configured to acquire a model monitoring accuracy according to at least two test sample points; The test sample points are determined according to target channel state information CSI and CSI corresponding to the target CSI and obtained by model recovery; the model monitoring accuracy is based on the statistical results of the at least two test sample points; A first determining unit, configured to determine monitoring event related information for the monitoring accuracy of the model; The monitoring event related information includes: a definition of the monitoring event, and an entry condition and / or an exit condition of the monitoring event; A first control unit, configured to control a triggering operation of a model monitoring report according to the model monitoring accuracy and monitoring event related information; The first device is a network device or a terminal.

32. An information processing device, applied to a first device, characterized in that: include: A third acquisition unit, configured to acquire at least two test sample points by using a time domain sliding window of a first length; The test sample point is determined according to a target CSI and a CSI corresponding to the target CSI and restored by a model; The first device is a network device or a terminal.

33. A non-transitory readable storage medium, characterized in that: The non-transitory readable storage medium stores a computer program, and the computer program is used to enable a processor to execute the method according to any one of claims 1 to 15.

Images