Method and communication device for obtaining training data in AI model training

Abstract

The present application provides a method for acquiring training data in training an AI model, which can be applied to a scenario in which an AI model training network element (second network element) and a training data collection network element (first network element) are logically deployed separately. The first network element receives first information from the second network element, the first information being for determining the validity of candidate training data collected by the first network element. The first network element collects candidate training data for the AI ​​model and determines the validity of the candidate training data based on the first information. When it is determined that the candidate training data is valid, the first network element sends the valid candidate training data to the second network element but does not send invalid candidate training data. When the collected candidate training data does not include valid candidate training data, the first network element indicates to the second network element that the currently collected training data is invalid and does not send the currently collected candidate training data to the second network element. In this way, waste of air interface resources can be reduced.

Description

[Technical Field]

[0001] TECHNICAL FIELD Embodiments of the present application relate to the field of machine learning, and more particularly to a method and communication device for obtaining training data in AI model training. [Background technology]

[0002] This application claims priority to Chinese Patent Application No. 202211203052.X, entitled "METHOD FOR OBTAINING TRAINING DATA IN AI MODEL TRAINING AND COMMUNICATION APPARATUS," filed with the State Intellectual Property Office of China on September 29, 2022, the entire contents of which are incorporated herein by reference.

[0003] When an AI model is applied to some application scenarios of air interface technology, AI model training and AI model training data collection may be deployed on different network elements. In this situation, training or updating an AI model requires the exchange of training data (e.g., reference signal measurement results and / or labels) between the AI ​​model training network element and the training data collection network element.

[0004] In existing solutions, training data between the AI ​​model training network element and the training data collection network element is typically sent periodically or continuously exchanged, but this information exchange method can result in wasted air interface resources. Summary of the Invention

[0005] The present application provides a method and a communication device for obtaining training data in AI model training to reduce the waste of air interface resources.

[0006] According to a first aspect, a method for acquiring training data in AI model training is provided, which can be applied to a training data collection network element, such as a terminal device or an access network device. The method includes:

[0007] a first network element receives first information from a second network element, the first information being for determining validity of the candidate training data collected by the first network element, and the determination result of the validity includes valid or invalid; The first network element collects candidate training data for the AI ​​model; The first network element sends second information to the second network element based on the candidate training data and the first information, the second information indicating a validity determination result.

[0008] In the technical solution of the present application, the first network element is a network element that collects training data for an AI model, and the second network element is a network element that trains the AI ​​model. When the second network element needs the first network element to collect training data for the AI ​​model, the second network element sends first information to the first network element. The first information indicates to the first network element to collect training data for the AI ​​model, and is further for determining the validity of the candidate training data collected by the first network element (also referred to as validity determination for short). The first network element collects candidate training data for the AI ​​model and determines the validity of the collected candidate training data based on the first information. Then, the first network element sends second information to the second network element to indicate the validity determination result. Based on this technical solution, after completing one collection of candidate training data, the first network element determines the validity of the collected candidate training data. Instead of providing collected data to the second network element without screening it, only valid candidate training data is provided by the first network element as training data to be used by the second network element, which reduces transmission of collected invalid candidate training data and thereby reduces waste of air interface resources.

[0009] Regarding the first aspect, in some implementations of the first aspect, the first information is for determining a constraint for validity determination. Optionally, the constraint is: quality indicator thresholds and quality indicator criteria; A training data quantity threshold and training data quantity criterion that meets the quality metric criteria; or Maximum duration of candidate training data collection corresponding to one validation may include one or more of:

[0010] Regarding the first aspect, in some implementations of the first aspect, the first information is: quality indicator thresholds, quality indicator criteria, A threshold for the amount of training data that satisfies the quality metric criteria; A measure of the amount of training data that meets the quality metric criteria, or Maximum duration of candidate training data collection corresponding to one validation indicates one or more of:

[0011] It may be understood that the part of the above information that is not indicated by the first information may be predefined in the protocol.

[0012] Optionally, the first information indicating the portion of the information may include explicitly indicating one or more pieces of information from the portion of the information or implicitly indicating one or more pieces of information from the portion of the information. An explicit indication may include the first information including one or more pieces of information from the portion of the information explicitly indicated by the first information. An implicit indication may include the first information including other information corresponding to one or more pieces of information from the portion of the information implicitly indicated by the first information. Optionally, the other information may include an index having a correspondence with one or more pieces of information from the portion of the information implicitly indicated by the other information. The other information may include one or more pieces of information, and the plurality of pieces of information each indicate all pieces of information from the implicitly indicated portion of the information.

[0013] Optionally, the correspondence may be predefined in a protocol or may be pre-stored or pre-configured, in which case the correspondence between multiple indexes and multiple values ​​of combinations of one or more items of the information may be configured by radio resource control (RRC) signaling.

[0014] Optionally, the first information may be carried in control information, for example, downlink control information (DCI).

[0015] Optionally, the quality indicators may include one or more quality indicators, each having a corresponding threshold and a corresponding judgment criterion. For example, the quality indicators may include one or more of a quality indicator of a reference signal measurement result or a quality indicator of a label. The label is used as a truth value of comparison for AI model training. For example, the label may include one or more of location information, a beam pattern, a channel measurement result, etc.

[0016] Optionally, the quality index threshold may comprise the above-mentioned amount threshold of training data that satisfies the quality index criterion, and the quality index criterion may comprise a criterion for the amount of training data that satisfies the quality index criterion.

[0017] In the present application, the first information may be for determining the constraint, which may be specifically implemented in several ways. In the following, some examples are used for illustration.

[0018] In an example, the first information indicates thresholds of one or more quality indicators, and criteria for the one or more quality indicators are predefined in the protocol. For example, the quality indicators include a signal to interference plus noise ratio (SINR) of training data and an amount of training data. The criterion for the SINR is that the SINR is greater than or equal to a threshold Q. The criterion for the amount of training data is that the amount of training data is greater than or equal to a threshold N. For example, the first information indicates Q and N, and both the criterion for the SINR and the criterion for the amount of training data are predefined in the protocol.

[0019] In this example, the criteria for the quality indicators are predefined in the protocol, and therefore the indication overhead can be reduced.

[0020] In another example, the first information indicates thresholds of one or more quality indicators and criteria for determining the one or more quality indicators. For example, the quality indicators include an SINR of training data and an amount of training data. The criterion for the SINR is that the SINR is equal to or greater than a threshold Q. The criterion for the amount of training data is that the amount of training data is equal to or greater than a threshold N. For example, the first information indicates Q and N. In addition, the first information includes an information field, which indicates the criterion for the SINR and the criterion for the amount of training data. For example, if the value of the information field is 1, it indicates that "the SINR is equal to or greater than Q and the amount of training data is equal to or greater than N," and if the value of the information field is 0, it indicates that "the SINR is greater than Q and the amount of training data is greater than N."

[0021] In this example, the first information indicates the quality index threshold and the quality index criterion, so that the second network element can adaptively update the constraints based on changes in the requirements for the training data, which is applicable to scenarios where the constraints change frequently, can improve the adaptability of the AI ​​model to different application scenarios, and can improve the probability of collecting training data that meets the requirements in different application scenarios.

[0022] In yet another example, the first information indicates thresholds of some quality indicators, and thresholds of other quality indicators and criteria for the quality indicators are predefined in the protocol. For example, the quality indicators include an SINR of training data and an amount of training data, and the criterion for the SINR is that the SINR of the training data is equal to or greater than a threshold Q, and the criterion for the amount of training data is that the amount of training data is equal to or greater than a threshold N. For example, the first information indicates Q, and the threshold N for the amount of training data, the criterion for the SINR, and the criterion for the amount of training data may be predefined in the protocol.

[0023] In this example, the threshold value of the quality indicator and the criterion for the quality indicator, which have a long change period in the application scenario, are predefined in the protocol, so that the signaling overhead can be reduced. The threshold value of the quality indicator and the criterion for the quality indicator, which change frequently, are indicated by the first information, so that the requirement for the required training data can be flexibly adjusted. In this example, both the signaling overhead and the flexibility of the constraint update can be taken into consideration.

[0024] In yet another example, the first information indicates thresholds of some quality indicators and one index information, where the index information is for determining criteria for some quality indicators, thresholds of other quality indicators in the constraints, and criteria for other quality indicators. For example, the first information indicates an SINR threshold Q and an index 0, where index 0 indicates that a threshold for the amount of training data is N, the SINR criterion is that the SINR of the training data is equal to or greater than Q, and the criterion for the amount of training data is that the amount of training data is at least N. Optionally, index 0 is an index value in one of multiple application scenarios. For example, the multiple application scenarios include, but are not limited to, CSI prediction, uplink positioning, downlink positioning, or beam management, and index 0 is one of one or more indexes corresponding to a beam management scenario. Optionally, index 0 is an index value in a particular application scenario. For example, index 0 is one of multiple indexes corresponding to an uplink positioning scenario.

[0025] In yet another example, the first information indicates one index information, and the index information is for determining thresholds of one or more quality indicators and criteria for determining the one or more quality indicators. For example, the first information indicates index 0, and index 0 indicates that a threshold for SINR of training data is Q, a threshold for the amount of training data is N, the SINR criteria are that the SINR of the training data is equal to or greater than Q, and the amount of training data criteria are that the amount of training data is at least N. Optionally, index 0 is an index value in one of a plurality of application scenarios. For example, the plurality of application scenarios include, but are not limited to, CSI prediction, uplink positioning, downlink positioning, or beam management, and index 0 is one of one or more indexes corresponding to a beam management scenario. Optionally, index 0 is an index value in a particular application scenario. For example, index 0 is one of a plurality of indexes corresponding to an uplink positioning scenario.

[0026] In the last two examples, the correspondence between the index information and the quality indicator thresholds and / or quality indicator criteria is predefined in the protocol and is used only as an example. Alternatively, in another implementation, the correspondence may include, but is not limited to, being pre-stored or pre-configured.

[0027] The above describes an example in which the first information is for determining the constraint, but the present application is not limited to the above example.

[0028] Optionally, with regard to the first aspect, in some implementations of the first aspect, the quality indicator indicated by the first information includes a quality indicator of a label of the AI ​​model.

[0029] Optionally, the training data of the AI ​​model further includes a label. In an example, in an application scenario of uplink positioning or downlink positioning, the label is location information. Optionally, the quality indicator in the constraint may further include a quality indicator of the label. For example, the quality indicator of the label may include a threshold value of the distance between the locations of different samples. Optionally, the quality indicator indicated by the first information further includes a quality indicator of the label of the AI ​​model.

[0030] With regard to the first aspect, in some implementations of the first aspect, the second information includes first training data, the second information indicates that the candidate training data collected by the first network element is valid, and the first training data is valid data among the candidate training data.

[0031] In this implementation, after the first network element determines the validity of the collected candidate training data based on the first information, if the first network element determines that the current collection is valid, the first network element sends second information to the second network element. The second information may be valid candidate training data (i.e., the first training data), and invalid candidate training data is not sent, thereby reducing waste of air interface resources.

[0032] In this application, valid candidate training data is provided to the second network element for training or updating the AI ​​model, i.e., valid candidate training data is actually training data. Invalid candidate training data is candidate training data that does not satisfy the constraints.

[0033] In addition, since the first network element does not send invalid candidate training data to the second network element, the second network element does not receive invalid or unqualified training data, thereby avoiding contamination of the entire training data set. In addition, adverse effects on the AI ​​model training by the second network element are also avoided, such as inaccurate AI performance gain evaluation, overfitting of the AI ​​model, weak generalization ability, and poor scenario adaptability caused by AI model training performed using invalid candidate training data.

[0034] Regarding the first aspect, in some implementations of the first aspect, the second information indicates that the candidate training data collected by the first network element is invalid.

[0035] In this implementation, if the first network element determines the validity of the collected candidate training data based on the first information and then determines that the current collection is invalid, the first network element sends second information to the second network element. The second information only indicates that the currently collected candidate training data is invalid and does not provide the collected candidate training data to the second network element, thereby reducing waste of air interface resources. In addition, since the second network element does not receive invalid or unqualified candidate training data, contamination of the entire training data set is avoided. In addition, adverse effects on AI model training by the second network element are also avoided, such as inaccurate AI performance gain evaluation, AI model overfitting, weak generalization ability, and poor scenario adaptability caused by AI model training performed using invalid candidate training data.

[0036] With respect to the first aspect, in some implementations of the first aspect, the first information is for determining constraints for determining the validity of candidate training data collected by the first network element.

[0037] In this implementation, when the second network element indicates to the first network element to collect training data for the AI ​​model based on the first information, the first information is further used by the first network element to determine constraints that the training data to be collected must satisfy, thus providing the first network element with a basis for performing screening (i.e., validity determination) after collecting candidate training data to determine whether the collected candidate training data is valid.

[0038] Regarding the first aspect, in some implementations of the first aspect, the method further includes:

[0039] If the first network element determines that the candidate training data includes first training data that satisfies the constraint, then the first network element determines that the candidate training data is valid; or If the first network element determines that the candidate training data does not include first training data that satisfies the constraint, the first network element determines that the candidate training data is invalid.

[0040] In this implementation, the set of candidate training data that is in the collected candidate training data and satisfies the constraints is called the first training data. If there is no candidate training data that satisfies the constraints, it indicates that the current collection is invalid.

[0041] Regarding the first aspect, in some implementations of the first aspect, when the candidate training data collected by the first network element is invalid, the method further includes:

[0042] The first network element receives third information from the second network element, where the third information indicates to the first network element to recollect candidate training data for the AI ​​model.

[0043] In this implementation, recollection is performed after one collection becomes invalid, and a validity determination is performed on the previously collected invalid candidate training data and the recollected candidate training data together to improve the probability of obtaining candidate training data that meets the requirements (i.e., obtaining training data). In addition, the air interface transmission configuration of the reference signal may be updated during recollection, thereby increasing the possibility of obtaining high-quality candidate training data and increasing the probability of collecting qualified training data.

[0044] Regarding the first aspect, in some implementations of the first aspect, the method further includes:

[0045] The first network element determines air interface transmission configuration information, where the air interface transmission configuration information corresponds to an updated air interface transmission configuration, and the air interface transmission configuration information indicates to the first network element to collect candidate training data for the AI ​​model based on the updated air interface transmission configuration.

[0046] Information about the updated air interface transmission configuration can be found at the transmit power of the reference signal, the amount of antenna ports used for the reference signal, the bandwidth of the reference signal, the frequency domain density of the reference signal, or Reference signal period The update may include one or more of the following:

[0047] In this implementation, when training data for an AI model needs to be recollected, the air interface transmission configuration of the reference signal related to the training data collection may be updated, thereby improving or ensuring the quality of the reference signal. In this manner, valid candidate training data may be collected, thereby providing assurance of AI model training, for example, during the initial training and / or update process. In addition, because updating the air interface transmission configuration of the reference signal helps collect valid candidate training data, AI model training efficiency may be further improved.

[0048] In addition, by learning the status of AI model training data collection, it can be learned whether there is a large amount of valid candidate training data for AI model training / updating.To ensure reliable AI model-based air interface performance, it is necessary to maintain the AI ​​model in a timely manner, switch to non-AI mode, or adaptively update the configuration for collecting training data.

[0049] Regarding the first aspect, in some implementations of the first aspect, the third information further indicates a maximum number of times k for determining validity, where k is a positive integer.

[0050] In this implementation, the third information indicates the maximum number of times k to determine validity, i.e., the second network element indicates the maximum number of times k to determine validity to the first network element only when it determines that recollection is necessary. This can avoid wasteful signaling caused by limiting the recollection process when the collection result is unknown. For example, the first network element may obtain valid candidate training data through one collection. In this case, recollection does not need to be performed. In this case, the second network element does not need to indicate the recollected related information to the first network element, thereby reducing signaling overhead.

[0051] Regarding the first aspect, in some implementations of the first aspect, the first information further indicates a maximum number of times k for determining validity, where k is a positive integer.

[0052] In this implementation, the first information indicates the maximum number of times k to determine validity, i.e., when training data collection starts, the second network element indicates the maximum number of times to determine validity, so that the first network element can quickly enter re-collection processing after one collection failure, thereby reducing the interaction time between the first network element and the second network element and improving training data collection efficiency.

[0053] In the above two implementations, the second network element indicates the maximum number k of times to determine validity to the first network element, so that the first network element can quickly collect candidate training data the next time when the initially collected candidate training data is invalid, and can repeatedly collect candidate training data when the maximum number k of times to determine validity is not exceeded, thereby reducing the signaling overhead of recollection indications and improving the efficiency of AI model training / update.

[0054] Regarding the first aspect, in some implementations of the first aspect, the method further includes:

[0055] The first network element collects candidate training data for the AI ​​model based on the updated air interface transmission configuration; When the maximum number k of validity determinations is reached and the first network element determines, based on the first information, that the result of the kth validity determination is invalid, the first network element stops collecting candidate training data for the AI ​​model.

[0056] In this implementation, based on the constraint of the maximum number k of times for determining validity, the collection process of the first network element can be prevented from entering an infinite loop, and resource occupation and resource waste can be avoided.

[0057] Regarding the first aspect, in some implementations of the first aspect, the method further includes:

[0058] If the first network element determines that the result of the jth validity determination is valid based on the first information before the maximum number k of validity determinations is exceeded, the first network element sends fourth information to the second network element, the fourth information including second training data, the fourth information indicating that the result of the jth validity determination is valid, the second training data including valid data among the candidate training data on which the jth validity determination was performed, j being less than or equal to k and j being a positive integer.

[0059] Regarding the first aspect, in some implementations of the first aspect, the first network element collecting candidate training data for the AI ​​model includes:

[0060] the first network element measures a reference signal from the second network element to obtain one or more measurement results, and the candidate training data for the AI ​​model includes the one or more measurement results; or The first network element measures a reference signal from a third network element to obtain one or more measurement results, and the candidate training data for the AI ​​model includes the one or more measurement results.

[0061] In this implementation, depending on different application scenarios, the candidate training data collected by the first network element for the AI ​​model may be a measurement result obtained by measuring a reference signal sent by a second network element or a reference signal sent by a third network element. Optionally, there may be one or more measurement results. For example, the first network element may obtain one measurement result by measuring the reference signal once, or the first network element may obtain multiple measurement results by measuring the reference signal multiple times, or the first network element may obtain multiple measurement results by measuring the reference signal once. This is not limited to this. In these implementations, the candidate training data includes one measurement result or multiple measurement results.

[0062] Regarding the first aspect, in some implementations of the first aspect, the first network element is a terminal device, the second network element is an access network device, and the first network element measures a reference signal from the second network element to obtain one or more measurement results.

[0063] Optionally, the signal from the second network element comprises one or more of a channel state information-reference signal (CSI-RS), a positioning reference signal (PRS), a synchronizing signal and physical broadcast channel block (SSB), and / or a signal on a physical broadcast channel.

[0064] In this implementation, the application of the AI ​​model is applicable to application scenarios such as AI model-based CSI feedback or CSI prediction, and AI model-based beam management, to solve problems such as CSI feedback or prediction and beam management, and improve the performance of the air interface in these application scenarios.

[0065] Regarding the first aspect, in some implementations of the first aspect, the first training data further includes reference signal information corresponding to K best measurement results of the one or more measurement results, where K is an integer greater than or equal to 1. It should be understood that when there is 1 measurement result, K is equal to 1, and when there are V measurement results, K is less than or equal to V, and K is greater than or equal to 1, where V is an integer greater than or equal to 2.

[0066] In this implementation, the AI ​​model is applicable to a beam management scenario, where the first training data further includes reference signal information corresponding to the K best measurement results, which is used as a label for the AI ​​model.

[0067] Regarding the first aspect, in some implementations of the first aspect, the first network element is an access network device, and the second network element is a positioning device; the first network element measures the sounding reference signal from the third network element to obtain one or more measurements; The first training data further includes location information of a third network element.

[0068] In this implementation, the AI ​​model is applicable to an uplink positioning scenario. In this case, a first network element measures a sounding reference signal of a third network element to obtain candidate training data, where the candidate training data includes location information of the third network element. When the candidate training data is valid, the first network element provides the valid candidate training data (i.e., the first training data) and the corresponding location information of the third network element to a positioning device to train or update the AI ​​model. The location information of the third network element is used as a label for the AI ​​model.

[0069] Regarding the first aspect, in some implementations of the first aspect, the first network element is a terminal device, and the second network element is a positioning device; the first network element measures a positioning reference signal from a third network element to obtain one or more measurement results, the third network element being an access network device; The first training data further includes location information of the first network element.

[0070] In this implementation, the AI ​​model is applicable to a downlink positioning scenario. If the candidate training data collected by the first network element (e.g., a location reference device) is valid, the first training data provided by the first network element to the second network element (i.e., a positioning device) further includes location information of the first network element, and the location information of the first network element is used as a label for the AI ​​model.

[0071] According to a second aspect, a method for obtaining training data in AI model training is provided, which can be applied to an AI model training network element, such as an access network device or a positioning device. The method includes:

[0072] The second network element sends first information to the first network element, the first information being for the AI ​​model to determine validity of the candidate training data collected by the first network element, and the determination result of the validity includes valid or invalid; The second network element receives second information from the first network element, the second information indicating the validity determination.

[0073] For the description of the first information, please refer to the description of the first aspect, and the details will not be described again in this specification.

[0074] With regard to the second aspect, in some implementations of the second aspect, the second information includes first training data, the second information indicates that the candidate training data collected by the first network element is valid, and the first training data is valid data among the candidate training data.

[0075] With regard to the second aspect, in some implementations of the second aspect, the second information indicates that the candidate training data collected by the first network element is invalid.

[0076] With respect to the second aspect, in some implementations of the second aspect, the first information is for determining constraints for determining the validity of candidate training data collected by the first network element.

[0077] Regarding the second aspect, in some implementations of the second aspect, the candidate training data is valid if it includes first training data that satisfies the constraint, or If the candidate training data does not include the first training data that satisfies the constraint, the candidate training data is invalid.

[0078] Regarding the second aspect, in some implementations of the second aspect, when the second information indicates that the candidate training data collected by the first network element is invalid, the method further includes:

[0079] The second network element sends third information to the first network element, where the third information indicates to the first network element to recollect candidate training data for the AI ​​model.

[0080] Regarding the second aspect, in some implementations of the second aspect, the method further includes:

[0081] The second network element determines air interface transmission configuration information, where the air interface transmission configuration information corresponds to the updated air interface transmission configuration, and the air interface transmission configuration information indicates to the first network element to collect candidate training data for the AI ​​model based on the updated air interface transmission configuration.

[0082] Information about the updated air interface transmission configuration can be found at the transmit power of the reference signal, the amount of antenna ports used for the reference signal, the bandwidth of the reference signal, the frequency domain density of the reference signal, or Reference signal period The update may include one or more of the following:

[0083] Regarding the second aspect, in some implementations of the second aspect, the third information further indicates a maximum number of times k for determining validity, where k is a positive integer.

[0084] Regarding the second aspect, in some implementations of the second aspect, the first information further indicates a maximum number of times k for determining validity, where k is a positive integer.

[0085] Regarding the second aspect, in some implementations of the second aspect, the method further includes:

[0086] The second network element receives fourth information from the first network element, the fourth information including second training data, the fourth information indicating that the determination result of the jth validity determination performed by the first network element is valid, the second training data being valid data among the candidate training data on which the jth validity determination was performed, j being less than or equal to k, and j being a positive integer.

[0087] Regarding the second aspect, in some implementations of the second aspect, the second network element is an access network device, the first network element is a terminal device, and the method further includes:

[0088] The second network element sends a reference signal to the first network element, the reference signal is used by the first network element to obtain one or more measurement results corresponding to the reference signal, and the candidate training data for the AI ​​model includes the one or more measurement results.

[0089] Regarding the second aspect, in some implementations of the second aspect, the first training data further includes reference signals corresponding to K best measurement results among the one or more measurement results, where K is an integer greater than or equal to 1.

[0090] Regarding the second aspect, in some implementations of the second aspect, the second network element is a positioning device, and the first network element is an access network device. The candidate training data for the AI ​​model includes one or more measurement results and location information of a third network element, and the one or more measurement results are obtained by the first network element by measuring a sounding reference signal sent by the third network element.

[0091] Regarding the second aspect, in some implementations of the second aspect, the second network element is a positioning device and the first network element is a terminal device. The candidate training data for the AI ​​model includes one or more measurement results and location information of the first network element, where the one or more measurement results are based on measurements of positioning reference signals sent by a third network element, and the third network element is an access network device. Optionally, the measurements are performed by the first network element.

[0092] Regarding the second aspect, in some implementations of the second aspect, the constraint is: quality indicator thresholds and quality indicator criteria; A training data quantity threshold and training data quantity criterion that meets the quality metric criteria; or Maximum duration of candidate training data collection corresponding to one validation Contains one or more of:

[0093] Regarding the second aspect, in some implementations of the second aspect, the first information is: quality indicator thresholds, quality indicator criteria, A threshold for the amount of training data that satisfies the quality metric criteria; A measure of the amount of training data that meets the quality metric criteria, or Maximum duration of candidate training data collection corresponding to one validation indicates one or more of:

[0094] Optionally, the quality indicators in the above implementations include one or more quality indicators, for example, one or more of a quality indicator of a label of an AI model or a quality indicator of a measurement result of a reference signal.

[0095] In some implementations of the first or second aspect, the constraint is based on an application scenario of the AI ​​model, and the application scenario of the AI ​​model includes: AI model-based CSI feedback or CSI prediction, AI model-based positioning, or AI model-based beam management Contains one or more of:

[0096] According to a third aspect, the present application provides a communication apparatus. The communication apparatus may be a terminal device, a device, module, chip, etc. disposed in a terminal device, or an apparatus capable of being used in conjunction with a terminal device. In design, the communication apparatus may include modules configured to implement and in one-to-one correspondence with the methods / operations / steps / actions described in the first aspect. The modules may be hardware circuits, software, or may be implemented by hardware circuits in combination with software. In design, the communication apparatus may include a processing module and a communication module.

[0097] According to a fourth aspect, the present application provides a communications device. In design, the communications device may include modules configured to implement the methods / operations / steps / actions described in the second aspect, in one-to-one correspondence. The modules may be hardware circuits, software, or implemented by hardware circuits in combination with software. In design, the communications device may include a processing module and a communications module. In an example, the communications device is an access network device or a positioning device, and the positioning device may be, for example, an LMF network element.

[0098] According to a fifth aspect, the present application provides a communication device. The communication device includes a processor configured to implement a method according to the first aspect or any one of the implementations of the first aspect. The processor is coupled to a memory. The memory is configured to store instructions and data. When the processor executes the instructions stored in the memory, the method described in the first aspect or any one of the implementations of the first aspect can be implemented. Optionally, the communication device may further include a memory. Optionally, the communication device may further include a communication interface. The communication interface is used by the device to communicate with another device. For example, the communication interface may be a transceiver, a hardware circuit, a bus, a module, a pin, or any other type of communication interface. In an example, the communication device may be a terminal device, a device, a module, a chip, etc. disposed in the terminal device, or a device capable of being used with the terminal device.

[0099] According to a sixth aspect, the present application provides a communications device. The communications device includes a processor configured to implement a method according to the second aspect or any one of the implementations of the second aspect. The processor is coupled to a memory. The memory is configured to store instructions and data. When the processor executes the instructions stored in the memory, the method described in the second aspect or any one of the implementations of the second aspect can be implemented. Optionally, the communications device may further include a memory. Optionally, the communications device may further include a communications interface. The communications interface is used by the device to communicate with another device. For example, the communications interface may be a transceiver, a hardware circuit, a bus, a module, a pin, or another type of communications interface. In an example, the communications device may be an access network device, a device, module, chip, etc. disposed in an access network device, or a device capable of being used with an access network device. In another example, the communications device may be a positioning device, a device, module, chip, etc. disposed in a positioning device, or a device capable of being used with a positioning device.

[0100] According to a seventh aspect, the present application provides a communication system, including a first network element and a second network element. For example, an interaction between the first network element and the second network element is as follows:

[0101] The second network element sends first information to the first network element, the first information being for determining validity of the candidate training data collected by the first network element, and the determination result of the validity includes valid or invalid; the first network element receives the first information from the second network element; The first network element collects candidate training data for the AI ​​model; the first network element sends second information to the second network element based on the candidate training data and the first information, the second information indicating a validity determination result; The second network element receives the second information from the first network element.

[0102] In particular, the solution on the first network element side should be understood with reference to the implementation in the first aspect, and the solution on the second network element side should be understood with reference to the implementation in the second aspect. Details will not be described again in this specification. For example, the communication system includes a terminal device and an access network device. Optionally, the communication system includes a terminal device, an access network device, and a positioning device. Optionally, the terminal device is a location reference device, and the positioning device is an LMF network element.

[0103] According to an eighth aspect, the present application provides a communication system including a communication device as described in the third aspect or the fifth aspect and a communication device as described in the fourth aspect or the sixth aspect.

[0104] According to a ninth aspect, the present application further provides a computer program, which when run on a computer enables the computer to perform the method according to the first aspect, the second aspect, or any one of the implementations of the first or second aspect.

[0105] According to a tenth aspect, the present application further provides a computer program product comprising instructions which, when executed on a computer, enable the computer to perform a method according to the first aspect, the second aspect, or any one of the implementations of the first or second aspect.

[0106] According to an eleventh aspect, the present application further provides a computer-readable storage medium storing a computer program or instructions, which, when executed on a computer, enables the computer to perform a method according to the first aspect, the second aspect, or any one of the implementations of the first or second aspect.

[0107] According to a twelfth aspect, the present application further provides a chip, wherein the chip is configured to read a computer program stored in a memory to perform a method according to the first aspect, the second aspect, or any one of the implementations of the first or second aspect, or the chip includes circuitry configured to perform a method according to the first aspect, the second aspect, or any one of the implementations of the first or second aspect.

[0108] According to a thirteenth aspect, the present application further provides a chip system. The chip system includes a processor configured to support an apparatus in implementing a method according to the first aspect, the second aspect, or any one of the implementations of the first or second aspects. In a possible design, the chip system further includes a memory configured to store programs and data required by the apparatus. The chip system may include a chip, or may include a chip and another individual component.

[0109] For the technical effects of the solutions provided in any one of the second to thirteenth aspects or the implementations of the second to thirteenth aspects, please refer to the corresponding description in the first aspect, and the details will not be described again. [Brief explanation of the drawings]

[0110] [Figure 1] FIG. 1 is a diagram of a neural network iteration process. [Figure 2] 1 is a diagram of the architecture of a communication system to which embodiments of the present application are applicable; [Figure 3] 1 is a schematic flowchart of a method for obtaining training data in AI model training according to the present application. [Figure 4] 1 is a diagram of an AI model-based CSI feedback mechanism. [Figure 5] FIG. 1 is a diagram of an example of obtaining training data in AI model-based CSI feedback according to the present application. [Figure 6] FIG. 1 is a diagram of a technical solution in an AI model-based uplink positioning scenario according to the present application. [Figure 7] FIG. 10 is a diagram of an example of obtaining training data in AI model-based uplink positioning according to the present application. [Figure 8] FIG. 1 is a diagram of a technical solution in an AI model-based downlink positioning scenario according to the present application. [Figure 9] FIG. 10 is a diagram of an example of obtaining training data in AI model-based downlink positioning according to the present application. [Figure 10] A diagram of AI-assisted sparse beam scanning processing. [Figure 11] FIG. 10 is a diagram of an example of obtaining training data in AI model-based beam management according to the present application. [Figure 12] 1 is a diagram of a configuration of a communication device according to the present application; [Figure 13] 1 is a diagram of a configuration of a communication device according to the present application; DETAILED DESCRIPTION OF THE INVENTION

[0111] The technical solutions of the present application are described below with reference to the accompanying drawings.

[0112] First, the concepts and techniques involved in the embodiments of the present application are briefly described.

[0113] An AI model is a function model that maps inputs of a specific dimension to outputs of a specific dimension. The model parameters of an AI model are obtained through machine learning training. For example, f(x)=ax 2 + b is a quadratic function model, which may be considered as an AI model, and a and b are parameters of the AI ​​model, which may be obtained through machine learning training. For example, the AI ​​model described in the following embodiments of the present application is not limited to a neural network, a linear regression model, a decision tree model, a support vector machine (SVM), a Bayesian network, a Q-learning model, or another machine learning (ML) model.

[0114] A training dataset is data used for model training, validation, and testing in machine learning. The quantity and quality of data affect the effectiveness of machine learning. The training data may include the input of an AI model, or may include the input and target output of an AI model. The target output is the target value output by the AI ​​model, and may also be called a truth value, an output truth value, a label, or a label sample.

[0115] Model training is the process of selecting an appropriate loss function and using an optimization algorithm to train the model parameters so that the value of the loss function is less than a threshold or meets a target requirement.

[0116] AI model design mainly includes a data collection phase (e.g., collection of training data and / or inference data), a model training phase, and a model inference phase. It may also include an inference result application phase. In the data collection phase, a data source is used to provide a training dataset and inference data. In the model training phase, training data provided by the data source is analyzed or trained to obtain an AI model. The AI ​​model represents a mapping relationship between the model's input and output. Obtaining an AI model through learning by using a model training node is equivalent to obtaining a mapping relationship between the model's input and output through learning by using training data. In the model inference phase, the AI ​​model obtained through training in the model training phase is used to perform inference based on inference data provided by the data source to obtain an inference result. This phase may also be understood as follows: inference data is input to the AI ​​model to obtain an output through the AI ​​model. This output is the inference result. The inference result may indicate the configuration parameters used (acted upon) by the execution object and / or the operation performed by the execution object. The inference result is released in the inference result application phase. For example, the inference results may be planned in a unified manner by an execution (actor) entity. For example, the execution entity may send the inference results to one or more execution objects (e.g., a core network device, an access network device, or a terminal device) for execution. In another example, the execution entity may further feed back the model performance to the data source to facilitate subsequent model update training.

[0117] The loss function is used to measure the difference between the model's prediction and the truth value.

[0118] Model application is the use of a trained model to solve a real-world problem.

[0119] Machine learning (ML) is an important technological approach to realizing artificial intelligence (AI). Machine learning can be divided into supervised learning, unsupervised learning, and reinforcement learning.

[0120] For example, in supervised learning, based on collected sample values ​​and sample labels, a mapping relationship between the sample values ​​and sample labels is learned by using a machine learning algorithm, and the learned mapping relationship is expressed by using a machine learning model. The process of training a machine learning model is a process of learning a mapping relationship. For example, in signal detection, a received signal containing noise is a sample, and the actual constellation point corresponding to the signal is a label. In machine learning, it is expected that the mapping relationship between the samples and the labels is learned through training, i.e., enabling the machine learning model to learn a signal detector. During training, model parameters are optimized by calculating the error between the model's predicted value and the actual label. Once the mapping relationship is learned, the sample label of each new sample can be predicted by using the learned mapping relationship. The mapping relationship learned through supervised learning may include linear mapping and nonlinear mapping. Learning tasks may be classified into classification tasks and regression tasks based on the type of label.

[0121] Please refer to Figure 1. Figure 1 is a diagram of the neural network iteration process. As shown in Figure 1, n samples are selected to form a batch, and then the batch is input into the neural network to obtain an output result. The output result and the sample labels are then input into a loss function to calculate the loss in the current round. Finally, the respective derivatives of each parameter are used together with the step parameter to perform parameter update. This is the iteration in the training process. Batch means "group" and indicates that the neural network processes data in batches. The batch size indicates the amount of samples in each batch. Therefore, an appropriate sample amount can usually be used for parallel calculation to accelerate the training speed. The amount of data processed at one time is not excessively large.

[0122] A training dataset is a set of training samples. Each training sample is an input for a neural network. The training dataset is used for model training. The training dataset is one of the most important parts of machine learning. The training process of machine learning is essentially to learn some features of the training dataset so that the difference between the output of the neural network and the ideal target value (i.e., label or output truth value) in the training dataset is minimized. Usually, even if the same network structure is used, the weights and outputs of a neural network trained using different training datasets will be different. Therefore, the composition and selection of the training dataset determine to some extent the performance of the trained neural network.

[0123] When an AI model is applied to air interface technology, data from an actual deployed network must be collected to form the training dataset required for model updating / training, regardless of whether the model updating / training is offline or online. A good training dataset helps design a wireless communication AI algorithm to obtain greater performance gains and improve the generalization ability and robustness of the final designed algorithm in multiple scenarios.

[0124] When an AI model is applied to some application scenarios of air interface technology, if the AI ​​model training network element and the training data collection network element are not in the same network element, the training data needs to be exchanged between the AI ​​model training network element and the training data collection network element. Based on the current technical situation, the training data is usually exchanged periodically or continuously, which is likely to cause waste of air interface resources.

[0125] In addition, the process by which the training data collection network element collects training data is not limited by the requirements of the AI ​​model training network element, and invalid collection often occurs. For example, the training data collected by the training data collection network element is not the training data actually required by the AI ​​model training network element, causing some invalid interactions and wasting air interface resources. In addition, when the AI ​​model training network element uses the training data to train the AI ​​model, the training dataset of the AI ​​model is easily contaminated, resulting in multiple problems such as inaccurate gain evaluation, model overfitting, weak generalization ability, and poor scenario adaptability.

[0126] Regarding the above problem, the present application provides a method for obtaining training data in AI model training, which helps to solve or improve the above problem.

[0127] The technical solutions provided in this application may be applied to various communication systems. For example, the communication system may be a fourth-generation (4G) communication system (e.g., a long-term evolution (LTE) system), a fifth-generation (5G) communication system, a worldwide interoperability for microwave access (WiMAX) or wireless local area network (WLAN) system, a satellite communication system, or a future communication system, such as a 6G communication system or a converged system of multiple systems. A 5G communication system may also be referred to as a new radio (NR) system.

[0128] A network element in a communication system may send a signal to or receive a signal from another network element. The signal may include information, signaling, data, etc. A network element may alternatively be referred to as an entity, a network entity, a device, a communication device, a communication module, a node, a communication node, etc. In this application, a network element is used as an example for purposes of illustration.

[0129] A communication system to which the present application is applicable includes a first network element and a second network element, and may optionally further include a third network element, and the quantity of the first network element, the quantity of the second network element, and the quantity of the third network element are not limited.

[0130] Please refer to FIG. 2. FIG. 2 is a diagram of the architecture of a communication system to which an embodiment of the present application can be applied. (a) of FIG. 2 is a diagram of the architecture of a communication system to which an embodiment of the present application can be applied. For example, the communication system includes a network device 110, a terminal device 120, and a terminal device 130. The terminal devices 120 and 130 may access and communicate with the network device 110. Optionally, the network device 110 may be an access network device. In an implementation, the communication system may further include an AI entity. The network device may forward data reported by the terminal device and related to an AI model to the AI ​​entity, and the AI ​​entity performs AI-related operations such as training dataset construction and model training, and provides outputs of the AI-related operations such as a trained AI model, model evaluation, and test results to the network device. In another implementation, the AI ​​entity may also be located inside the network device 110, i.e., in a module of the network device 110. (b) of FIG. 2 is another diagram of the architecture of a communication system to which an embodiment of the present application can be applied. The communication system includes a network device 110, a terminal device 120, a terminal device 130, and a positioning device 140. The positioning device 140 and the network device 110 may communicate with each other by using interface messages. For example, the positioning device 140 may be a location management function (LMF), and the network device 110 may be an access network device, such as a gNB or an eNB. This is not limited to this. For example, if the access network device 110 is a gNB, the gNB and the LMF may exchange information by using NR positioning protocol A (NRPPa) messages.If the access network device 110 is an eNB, the eNB and the LMF may exchange information by using LTE positioning protocol (LPP) messages. Optionally, the terminal device may alternatively directly communicate with the positioning device 140, for example, the interaction between the terminal device 130 and the positioning device 140 shown in (b) of FIG. 2. In FIG. 2, the AI ​​entity may be configured inside the positioning device 140 or may be disposed separately from the positioning device 140. This is not limited. Optionally, the positioning device and the network device may be different modules of the same device or different separate devices.

[0131] In practical applications, one network device may serve one or more terminal devices. One terminal device may also access one or more network devices. The number of terminal devices and the number of network devices included in the wireless communication system are not limited in the embodiments of the present application. In addition, the positioning device 140 in FIG. 2(b) is not limited to an LMF network element, and may alternatively be another network element having a positioning function, and the number of positioning devices is also not limited.

[0132] For example, the network device may be a device with wireless transceiver functionality. The network device may be a device that provides wireless communication function services, and is usually located on the network side, and may include, but is not limited to, a next generation Node B (gNodeB, gNB) in a fifth generation (5G) communication system, a base station in a sixth generation (6G) mobile communication system, a base station in a future mobile communication system, an access point (AP) in a wireless fidelity (Wi-Fi) system, an evolved Node B (eNB) in a long term evolution (LTE) system, a radio network controller (RNC), a Node B (Node B, NB), a base station controller (BSC), a home base station (e.g., a home evolved Node B, or Home Node B, HNB), a baseband unit (BBU), a transmission reception point (TRP), a transmitting point (TP), a base transceiver station (BTS), a base transceiver station (BTS), a base station controller (BSC), a base station controller (BSC), a base station controller (BSC), a baseband unit (BBU), a baseband unit (BBU), a base station controller (BSC ... station controller (BSC), a base station controller (BSC), a base station controller (BSC), a base station controller (BSC), a baseband unit (BBU), a baseband unit (BBU), a base station controller (BSC), a base station controller (BSC), a base station controller (BSC), a base station controller (BSC), a base station controller (BSC), a base station controller (BSC), a base station controller (BSC), a base station controller (BSC), a base station controller (BSC), a base station controller (BSC), a base station controller (BSC), a base station controller (BSC), a base station controller (BSC), a baseband unit (BBU), a baseband unit (BBU), a base BTS), satellites, unmanned aerial vehicles, etc. In a network structure, the network devices may include a central unit (CU) node, a distributed unit (DU) node, a RAN device including a CU node and a DU node, or a RAN device including a CU control plane node, a CU user plane node, and a DU node. Alternatively, the network devices may be radio controllers, relay stations, in-vehicle devices, wearable devices, etc. in a cloud radio access network (CRAN) scenario.Additionally, the base station may be a macro base station, a micro base station, a relay node, a donor node, or a combination thereof. Alternatively, the base station may be a communication module, a modem, or a chip disposed in the above-mentioned device or apparatus. Alternatively, the base station may be a mobile switching center, a device performing base station functions in device-to-device (D2D), vehicle-to-everything (V2X), or machine-to-machine (M2M) communications, a network-side device in a 6G network, a device performing base station functions in a future communication system, etc. The base station may support networks of the same access technology or different access technologies. This is not limited thereto.

[0133] The network devices may be fixed or mobile. For example, the access network device 110 may be static and responsible for wireless transmission and reception from the terminal devices 120 and 130 in one or more cells. The access network device 110 may alternatively be mobile. For example, a helicopter or unmanned aerial vehicle may be configured as a mobile base station, and one or more cells may move based on the location of the mobile base station. It should be understood that in another example, a helicopter or unmanned aerial vehicle may be configured as a device that communicates with the base station 110.

[0134] In the present application, a communication device configured to implement the functions of an access network may be an access network device, may be a network device having some functions of an access network, or may be a device capable of supporting the implementation of the functions of an access network, such as a chip system, a hardware circuit, a software module, or a combination of a hardware circuit and a software module. This device may be installed in or used together with an access network device. In the method of the present application, an example in which the communication device configured to implement the functions of an access network device is an access network device is used for explanation.

[0135] The terminal device may be an entity configured to receive or transmit signals at a user side, such as a mobile phone. The terminal device may include a handheld device with wireless connectivity, another processing device connected to a wireless modem, an in-vehicle device, etc. The terminal device may be a portable, pocket-sized, handheld, computer-integrated, or in-vehicle mobile device. The terminal device 120 may be widely used in various scenarios, such as cellular communications, Wi-Fi systems, D2D, V2X, peer-to-peer (P2P), M2M, machine-type communications (MTC), the Internet of Things (IoT), virtual reality (VR), augmented reality (AR), industrial control, autonomous driving, telemedicine, smart grids, smart furniture, smart offices, smart wearables, smart cities, unmanned aerial vehicles, robots, remote sensing, passive sensing, positioning, navigation and tracking, autonomous delivery and mobility, etc.Some examples of the communication device 120 include a user equipment (UE) of a 3GPP standard, a station (STA) of a Wi-Fi system, a fixed device, a mobile device, a handheld device, a wearable device, a cellular phone, a smartphone, a session initiation protocol (SIP) phone, a notebook computer, a personal computer, a smartbook, a vehicle, a satellite, a global positioning system (GPS) device, a target tracking device, an unmanned aerial vehicle, a helicopter, an airplane, a ship, a remote control device, a smart home device, an industrial device, a personal communication service (PCS), a telephone, a wireless local loop (WLL) station, a personal digital assistant (PDA), a wireless network camera, a tablet computer, a palmtop computer, a mobile internet device (MID), a wearable device such as a smart watch, a virtual reality (VR) device, an augmented reality (AR) device, an industrial control wireless terminals in smart city systems, such as smart fuel dispensers; terminal devices for high-speed rail; and wireless terminals in smart homes, such as smart speakers, smart coffee machines, and smart printers.The terminal device 120 may be a wireless device or a device disposed in a wireless device in the various scenarios described above, such as a communication module, modem, or chip in the device. The terminal device may also be referred to as a terminal, user equipment (UE), mobile station (MS), mobile terminal (MT), etc. Alternatively, the terminal device may be a terminal device in a future wireless communication system. In addition, the terminal device may further include a location reference device, such as an automated guided vehicle (AGV) or a device with similar functionality. The specific technology and the specific device form used by the terminal device are not limited in the embodiments of the present application.

[0136] In the present application, a communication device configured to implement terminal device functions may be a terminal device, may be a terminal device having some functions of the above-mentioned communication device, or may be an apparatus capable of supporting the implementation of the above-mentioned terminal device functions, for example, a chip system. This apparatus may be installed in or used together with a terminal device. In the present application, a chip system may include a chip, or may include a chip and another individual component.

[0137] It should be understood that the amount of devices and the respective types of devices in the communication system shown in Figure 2 are used for example only, and the present application is not limited thereto. In actual applications, the communication system may further include more terminal devices, more access network devices, and more positioning devices, and may further include other network elements, for example, core network devices and / or network elements configured to implement artificial intelligence functions.

[0138] The following describes the technical solutions provided in this application.

[0139] Please refer to Figure 3. Figure 3 is a schematic flowchart of a method for acquiring training data in AI model training according to the present application. In the method of Figure 3, the first network element may be a network element that collects training data for the AI ​​model, and the second network element is an AI model training network element. Optionally, the second network element may be the AI ​​model training network element, and may also be a network element on which AI inference is performed. The first network element and the second network element may be logically deployed separately. In different implementations, the first network element and the second network element may be physically deployed in the same network element or different network elements. This is not limited.

[0140] 310: A first network element receives first information from a second network element, the first information indicating determining validity of candidate training data collected by the first network element, and the validity determination result includes valid or invalid.

[0141] The first network element may determine the validity of the collected candidate training data based on the first information. In other words, the first network element may determine whether the collected candidate training data is valid based on the first information.

[0142] In particular, if the first network element determines, based on the first information, that the collected candidate training data includes valid training data, the first network element determines that the collected candidate training data is valid; if the first network element determines, based on the first information, that the collected candidate training data does not include valid candidate training data, the first network element determines that the collected candidate training data is invalid. In other words, if some candidate training data among the candidate training data collected by the first network element is valid, the determination result is valid. If the determination result is valid, the valid candidate training data becomes the training data currently collected by the first network element. For example, valid candidate training data (hereinafter sometimes referred to as "valid data" for short) may be, but is not limited to, some or all of the collected candidate training data, and are collectively referred to as first training data in this specification. If the candidate training data collected by the first network element does not include valid candidate training data, the determination result is invalid.

[0143] It can be seen that in this embodiment of the present application, after collecting the candidate training data, the first network element determines the validity once at the first network element side.

[0144] In an example, the first information indicates a constraint that is used by the first network element to determine the validity of the collected candidate training data.

[0145] For example, the constraint is quality indicator thresholds and quality indicator criteria; A training data quantity threshold and training data quantity criterion that meets the quality metric criteria; or Maximum duration of candidate training data collection corresponding to one validation Contains one or more of:

[0146] Optionally, the first information is for determining a constraint.

[0147] For example, the first piece of information is quality indicator thresholds, quality indicator criteria, A threshold for the amount of training data that satisfies the quality metric criteria; A measure of the amount of training data that meets the quality metric criteria, or Maximum duration of candidate training data collection corresponding to one validation This indicates one or more of the following information:

[0148] The constraint is assumed to include one or more quality indicators. When the first information is for determining the constraint, there may be multiple implementations.

[0149] Optionally, in an example, the first information indicates thresholds of one or more quality indicators, and criteria for the one or more quality indicators are predefined in a protocol. In this example, the first network element determines the constraint based on the first information and the protocol predefinition.

[0150] Optionally, in another example, the first information indicates thresholds of one or more quality indicators and criteria for determining the one or more quality indicators. In this example, the first network element determines the constraint based on the first information.

[0151] Optionally, in yet another example, the constraint includes a plurality of quality indicators, the first information indicates thresholds of some of the plurality of quality indicators, and the thresholds of the other portion of the quality indicators and the criteria for the plurality of quality indicators are predefined in a protocol. In this example, the first network element determines the constraint based on the first information and the protocol predefinition.

[0152] Optionally, in yet another example, the first information indicates thresholds of some quality indicators and one index information, and the index information is for determining criteria for the some quality indicators, thresholds of other quality indicators in the constraint, and criteria for the other quality indicators. In this example, the first network element determines the constraint based on the first information and the index information.

[0153] Optionally, in yet another example, the first information indicates one index information, and the index information is for determining thresholds of one or more quality indicators and criteria for the one or more quality indicators. In this example, the first network element determines the constraint based on the index information.

[0154] In addition, optionally, the "protocol pre-definition" in the above example may also be another implementation, such as pre-configuration or pre-storage, which is not limited thereto.

[0155] In addition, for specific implementations in which the first information is for determining constraints, please refer to the corresponding implementations in the content section of the Summary of the Invention, and the details will not be described again in this specification.

[0156] In the example, the first information further indicates a maximum number of times k for determining validity, where k is a positive integer.

[0157] When the first information includes a plurality of the above information, it may be understood that the plurality of information may be carried in one message or separately in multiple messages. In other words, the first information may be carried in one message or in multiple messages.

[0158] The maximum duration of candidate training data collection corresponding to one validation determination is hereinafter denoted as Z, where Z is a number greater than 0. The maximum duration of candidate training data collection is also the maximum duration for which candidate training data can be used for validation determination. When the retention duration of candidate training data exceeds the maximum duration Z, the candidate training data becomes invalid and is no longer used for validation determination. Optionally, the maximum duration Z may be the same as the interval between two adjacent validation determinations, or may be greater than or less than the interval between two adjacent validation determinations. If the maximum duration is greater than the interval between two adjacent validation determinations, the candidate training data corresponding to one validation determination may include all or part of the candidate training data corresponding to one or more validation determinations prior to the current validation determination. The interval between two adjacent validation determinations may be fixed, specifically, the validation determinations are performed periodically within a specific time, or may be variable, specifically, the time for validation determinations is not fixed. For example, validation determination counting is performed on candidate training data that meets a threshold requirement. When the amount of candidate training data that meets the threshold requirement satisfies the requirement, the current validity determination is completed and the determination result is valid. When the amount of candidate training data that meets the threshold requirement does not meet the requirement and exceeds the maximum validity determination interval T (i.e., a preset interval threshold), or when the amount of collected candidate training data exceeds a preset threshold (i.e., the maximum amount of collected candidate training data), the current validity determination is also completed and the determination result is invalid. One or more of the specific time-related information of the validity determination, such as the determination time, the start time or duration of the periodic determination, the maximum interval T, and the maximum amount of collected candidate training data, may be fully or partially predefined in the protocol or may be based on configuration.

[0159] From the above description, it is known that in the present application, a second network element sends first information to a first network element, and the first information is for determining the validity of candidate training data collected by the first network element. In practice, the second network element uses the first information to indicate training data requirements to the second network element. In other words, only candidate training data that meets the requirements can be used as training data for training or updating an AI model. It can be seen that only after "filtering" is performed on the candidate training data collected by the first network element, candidate training data that meets the requirements can be used as training data, and the first network element provides the training data to the second network element for use. Therefore, after collecting the candidate training data, the first network element determines the validity of the collected candidate training data based on the first information. If the determination result is invalid, it indicates that the currently collected candidate training data is not required by the second network element, i.e., the currently collected candidate training data does not include candidate training data that meets the requirements. In this case, the first network element may re-collect training data. Therefore, in the process of the first network element collecting training data for the AI ​​model, the training data may not be acquired through a single collection. In a specific implementation, the maximum interval T between one validation determination is equivalent to specifying how frequently the first network element performs validation.

[0160] For example, if the determination result of the i-th validity determination is invalid, the first network element re-collects the candidate training data. After a time period, the first network element performs a validity determination on the candidate training data collected the (i+1)th time, where i is a positive integer. Therefore, in this embodiment of the present application, the number of times the first network element collects candidate training data for the AI ​​model corresponds to or is equal to the number of times the first network element performs a validity determination. In other words, each time the first network element performs a validity determination, it indicates that the candidate training data has been collected once before the current determination. For clarity in the description of the technical solution, the candidate training data collection just before the i-th determination is referred to herein as the i-th collection. Since there are different implementations, in this example, the validity determination performed the (i+1)th time may be on the candidate training data obtained through the (i+1)th collection, or on the candidate training data obtained through the (i+1)th collection and one or more collections before the (i+1)th collection. This is not limited. This implementation may involve the issue of how, after one validation, the first network element handles the candidate training data collected before the current validation.

[0161] In the example, after starting the i-th collection of candidate training data, the first network element performs a validity check on the i-th collected candidate training data after an interval T0 (less than or equal to the maximum interval T), i.e., performs the i-th validation check. The result of the i-th validation check is assumed to be invalid. If the maximum number k of validation checks is not exceeded, the first network element may discard the i-th collected candidate training data and perform the (i+1)-th collection. In this example, each validity check is performed only on the candidate training data collected within the interval T0. If the current collection is invalid, the currently collected candidate training data is discarded. In another example, one validity check may be performed on candidate training data collected within multiple intervals T0. Alternatively, candidate training data for which one validity check has been performed may be collected multiple times. The result of the i-th validation check is assumed to be invalid. If the maximum number k of validity checks is not exceeded, the first network element may retain some of the i-th collected candidate training data. For example, the first network element retains a portion of the candidate training data collected the i-th time that satisfies the criteria for some quality indicators in the constraints. Then, the (i+1)th collection is performed. After the interval T0, the first network element performs a validity determination on the candidate training data collected the (i+1)th time and the portion of the candidate training data collected the i-th time whose retention duration does not exceed the maximum duration Z, i.e., performs the (i+1)th validity determination. Optionally, if the result of the validity determination is always invalid before the maximum number k of validity determinations is exceeded, the first network element may retain the candidate training data that satisfies the criteria for some quality indicators and is among the candidate training data collected each time, and after completing a new collection, perform a validity determination on both the retained historical candidate training data that satisfies the criteria for some quality indicators and the newly collected candidate training data.It can be understood that candidate training data determined to be invalid in one validation is valid for the current validation, but this does not mean that candidate training data determined to be invalid in the current validation cannot be used as training data at all. These specific implementations are not limited in this application.

[0162] In the example, the constraints are based on application scenarios of the AI ​​model, such as, but not limited to, AI model-based CSI feedback or CSI prediction, AI model-based positioning, or AI model-based beam management This scenario includes:

[0163] In different application scenarios, one or more of the quality index, the quality index threshold, the quality index criterion, the training data amount threshold that satisfies the quality index criterion, and the training data amount criterion in the constraint may be different. Examples of different application scenarios are provided below individually.

[0164] 320: A first network element collects candidate training data for an AI model.

[0165] The first network element collects candidate training data for the AI ​​model. Optionally, after receiving the first information, i.e., based on the trigger of the first information, the first network element may start collecting candidate training data for the AI ​​model. Optionally, before or at the time of receiving the first information, the first network element may start collecting candidate training data for the AI ​​model. That is, the order of steps 310 and 320 may not be limited.

[0166] Optionally, since the requirements for training data required by the second network element may change, when the second network element's requirements for training data change, the second network element may send updated first information to the first network element. The update of the first information mainly refers to the update of the constraints determined based on the first information. Upon receiving the updated first information, the first network element determines the validity of the collected candidate training data based on the constraints determined by the updated first information. In the following embodiments, only the first information received by the first network element at a specific time is used as an example to describe the validity determination and subsequent procedures.

[0167] 330: The first network element sends second information to the second network element, where the second information indicates a determination result of the validity of the candidate training data collected by the first network element.

[0168] If the first network element determines, based on the first information, that the collected candidate training data is valid, the first network element sends second information to the second network element, where the second information indicates that the candidate training data collected by the first network element is valid.

[0169] Optionally, in an example, the first network element sends the first training data to the second network element, where the first training data implicitly indicates that the candidate training data collected by the first network element is valid. Optionally, in another example, the first network element sends the first training data to the second network element. In this case, the first network element further sends information, e.g., information a, indicating that the candidate training data collected by the first network element is valid to the second network element. In this example, the first network element sends the first training data and information a to the second network element, where the information indicates that the current collection is valid.

[0170] If the first network element determines, based on the first information, that the collected candidate training data is invalid, the first network element sends second information to the second network element, where the second information indicates that the candidate training data collected by the first network element is invalid. It should be understood that if the candidate training data collected by the first network element is invalid, the first network element sends only an indication that the collected candidate training data is invalid to the second network element, and does not send the collected invalid candidate training data, thereby reducing waste of air interface resources.

[0171] In an example, if the first network element determines that the collected candidate training data is invalid, the first network element discards the currently collected candidate training data. Alternatively, in some implementations described above, the invalid candidate training data from one validation determination may be retained for a subsequent validation determination. Furthermore, if the second network element indicates to the first network element that it should recollect training data for the AI ​​model, the first network element recollects the candidate training data for the AI ​​model.

[0172] Optionally, in step 320, the first network element collecting candidate training data for the AI ​​model may specifically involve the first network element measuring a reference signal from a second network element or a third network element to obtain the candidate training data for the AI ​​model. In other words, the candidate training data includes measurement results obtained by the first network element by measuring a reference signal. In this application, the reference signal is typically a signal for channel measurement. The channel measurement may be used for one or more functions, such as channel state information feedback, beam management, or positioning. The reference signal may include one or more of a channel state information reference signal, a synchronization signal such as a primary synchronization signal and / or a secondary synchronization signal, a physical broadcast signal, a synchronization signal and physical broadcast channel block (SSB), a demodulation reference signal, a phase tracking reference signal, or a positioning reference signal. When the AI ​​model is applied to different scenarios, the reference signal may be different. In the following embodiments, different application scenarios are individually described by using examples. In addition, the third network element is a network element different from the second network element.

[0173] In the example, the first network element may be understood to measure a reference signal from a second network element to obtain a measurement result. Optionally, there may be one or more measurement results. The candidate training data for the AI ​​model collected by the first network element includes one or more measurement results. In this example, before the second network element sends the reference signal to the first network element, the second network element sends air interface transmission configuration information to the first network element. The air interface transmission configuration information corresponds to the air interface transmission configuration, and the air interface transmission configuration information indicates to the first network element that it will collect candidate training data for the AI ​​model based on the air interface transmission configuration. In other words, the second network element sends a reference signal based on the air interface transmission configuration, and the first network element measures the reference signal from the second network element to obtain a measurement result and collect candidate training data based on the air interface transmission configuration. Optionally, the first network element is a UE, and the second network element is an access network device, for example, a base station.

[0174] In another example, a first network element measures a signal from a third network element to obtain measurement results. Optionally, there may be one or more measurement results. The candidate training data for the AI ​​model collected by the first network element includes one or more measurement results. Optionally, the first network element is an access network device, e.g., a base station, and the third network element is a UE. In this example, before measuring a reference signal from the third network element, the first network element configures the third network element to send the reference signal. In particular, the first network element sends air interface transmission configuration information to the third network element, where the air interface transmission configuration information corresponds to the air interface transmission configuration. Similar to the above example, the third network element sends the reference signal based on the air interface transmission configuration, and the first network element measures the reference signal from the third network element to obtain the measurement results and obtain candidate training data based on the air interface transmission configuration.

[0175] In the above example, optionally, the air interface transmission configuration includes: the transmit power of the reference signal, the amount of antenna ports used for the reference signal, the bandwidth of the reference signal, the frequency domain density of the reference signal, or Reference signal period may include one or more of:

[0176] From the procedure shown in FIG. 3, it can be seen that the first network element collects candidate training data for the AI ​​model and determines the validity of the candidate training data based on the first information. This may be understood as filtering the collected candidate training data. When the collected candidate training data is valid, the first network element sends the valid candidate training data to the second network element, which then trains or updates the AI ​​model. In this case, the valid candidate training data is training data. When the candidate training data collected by the first network element is invalid, the first network element indicates to the second network element that the current collection of candidate training data is invalid.

[0177] If the candidate training data collected once is invalid, it may be understood to also indicate that training data was not collected this time, and if the candidate training data collected once is valid, it may be understood to also indicate that training data was collected this time, in which case the valid candidate training data becomes training data, which is referred to herein as first training data, and is provided by the first network element to the second network element.

[0178] In the embodiments of the present application, it should be understood that "the first network element recollects training data for the AI ​​model" also indicates "the first network element recollects candidate training data for the AI ​​model." The first network element attempts to recollect training data only when it does not collect training data. The purpose of recollection is to expect to collect training data for the AI ​​model. However, in the process of collecting training data for the AI ​​model, candidate training data is first collected, and then training data is obtained from the candidate training data through filtering.

[0179] After receiving the second information of the first network element, if the second network element determines, based on the second information, that the current collection of the first network element is invalid, if possible, the second network element determines that training data for the AI ​​model needs to be re-collected.

[0180] In an example, the second network element sends third information to the first network element, where the third information indicates to the first network element to re-collect training data for the AI ​​model. Optionally, in an example, the third information indicates a maximum number of validation determinations k. Optionally, each validation determination includes collection of a new batch of training data, i.e., a new training dataset. Thus, the maximum number of validation determinations may also be referred to as a maximum number of times to perform training dataset collection.

[0181] In this example, when the second network element receives information from the first network element indicating that the candidate training data collected by the first network element is invalid, the second network element sends third information to the first network element to indicate to the first network element to recollect training data for the AI ​​model. In addition, the third information indicates a maximum number k of validity determinations, where k is a positive integer. If the candidate training data corresponding to the first validity determination is invalid, the first network element recollects or continues collecting the candidate training data based on the third information. The recollection or continued collection may be performed multiple times. After each recollection or continued collection, a validity determination is performed. If the determination result is invalid, the first network element may continue to perform the next recollection and the next validity determination until the maximum number k of validity determinations is reached. If the determination results of the first to (k-1)th validity determinations are all invalid and the determination result of the kth validity determination is still invalid, the first network element stops collecting training data.

[0182] In this example, when indicating to the first network element that it will recollect training data by using the third information, the second network element may further indicate a maximum number k of validity determinations to the first network element. In other words, the maximum number k of validity determinations is sent after the second network element determines that training data needs to be recollected. The maximum number k of validity determinations may be included in the third information or may be carried in information other than the third information. Optionally, in another example, as described in step 501, the second network element indicates the maximum number k of validity determinations in the first information sent to the first network element. The second network element indicates the maximum number k of validity determinations to the first network element to limit the first network element's process of recollecting candidate training data; therefore, if no training data (i.e., valid candidate training data) is collected, the first network element will not enter a recollection cycle without a time limit, but will stop collection after the maximum number k of validity determinations is reached, regardless of whether training data has been collected.

[0183] If the first network element determines, based on the first information, that the result of the j-th validation is valid, i.e., that the candidate training data on which the j-th validation is performed includes valid candidate training data, before the maximum number k of validity determinations is exceeded, the first network element sends fourth information to the second network element. The fourth information includes second training data and indicates that the result of the j-th validation is valid. The second training data may particularly include valid candidate training data among the candidate training data on which the j-th validation is performed. j is less than or equal to k and is a positive integer.

[0184] It should be noted that the jth validation may be considered as a validation performed on a set of candidate training data, and all candidate training data included in the set is training data for which the jth validation has been performed. The training data for which the jth validation has been performed is not limited to the candidate training data collected at the jth collection, but may further include, without limitation, candidate training data obtained in one or more collections prior to the jth collection.

[0185] Optionally, if the determination result of the jth validity determination is valid, the first network element sends second training data and information a to the second network element, where the information a indicates that the current collection is valid.

[0186] In addition, in an embodiment of the present application, the maximum number k of validity determinations corresponds to a start moment, and the start moment should be understood as the start moment of the training data collection process corresponding to the maximum number k of validity determinations. For example, the start moment may be the moment when the first network element receives the first information or the third information. In other words, the first network element starts collecting training data for the AI ​​model from the moment when the first information or the third information is received. Optionally, the end moment of the collection process is uncertain. For example, if the determination result of the jth validity determination is valid before the maximum number k of validity determinations is exceeded, the collection process ends, and the first network element sends valid candidate training data (i.e., the first training data) to the second network element, where j is less than or equal to k and j is a positive integer. However, if the determination results of the first to kth validity determinations are all invalid, the moment when the determination result of the kth validity determination is invalid is determined to be the end moment of the collection process.

[0187] According to the method for obtaining training data provided in this application, the training data collection network element performs a validity check on the collected candidate training data and provides valid candidate training data to the AI ​​model training network element, ensuring that the collection network element only provides training data that meets the requirements of the training network element. Training data that does not meet the requirements is filtered out at the collection network element side, so that invalid training data exchange is omitted. This not only saves air interface resources, but also avoids contamination of the training data set at the training network element side and other adverse effects.

[0188] The above describes in detail the main steps of the method for obtaining training data for an AI model. Below, by using examples, we will describe the method for obtaining training data when the AI ​​model is applied to different scenarios.

[0189] Application Scenario 1 AI model-based channel state information (CSI) feedback or CSI prediction For example, in application scenario 1, the AI ​​model training or update is deployed on the access network device side. The access network device sends a downlink reference signal to the UE, and the UE then obtains a measurement result by measuring the downlink reference signal, which is candidate training data. The UE determines the validity of the obtained candidate training data and provides the valid candidate training data to the access network device side for AI model training or update. For example, in this application scenario, the downlink reference signal may be, in particular, CSI-RS. The valid candidate training data provided by the UE to the access network device is a label for the AI ​​model, in particular, CSI.

[0190] In many application scenarios, the access network side needs to acquire downlink CSI to determine one or more of configurations such as resources for scheduling a UE's downlink data channel, modulation and coding scheme (MCS), and precoding. In a time division duplex (TDD) system, due to reciprocity between the uplink channel and the downlink channel, the access network device may acquire uplink CSI by measuring an uplink reference signal to infer downlink CSI, for example, use the uplink CSI as the downlink CSI. In a frequency division duplex (FDD) system, reciprocity between the uplink channel and the uplink channel cannot be guaranteed, and the downlink CSI is acquired by the UE by measuring a downlink reference signal. For example, the UE acquires downlink CSI by measuring signals such as CSI-RS or a synchronization signal and physical broadcast channel block (SSB). The UE generates a CSI report in a manner predefined in a protocol or preconfigured by the access network device, and feeds back downlink CSI to the access network device by using the CSI report, so that the access network device obtains the downlink CSI.

[0191] Please refer to FIG. 4. FIG. 4 is a diagram of an AI model-based CSI feedback mechanism. As shown in FIG. 4, an autoencoder (AE) model includes two submodels, an encoder and a decoder, and AE generally refers to a network structure including the two submodels. The AE model is also called a two-way model, a two-sided model, or a cooperative model. The encoder and decoder of the AE are usually trained together and may be used together. CSI feedback may be implemented based on the AI ​​model of the AE. For example, the UE side measures a downlink reference signal sent by a base station to obtain measured CSI. The UE compresses and quantizes the CSI obtained through measurement by using an encoder and feeds back the compressed and quantized information, e.g., “feedback CSI” shown in FIG. 4, to the base station. The base station restores the “feedback CSI” by using a decoder to obtain restored CSI. For the base station, the input of the decoder is the information about the CSI fed back by the UE, and the CSI obtained by the UE through measurements needs to be used as the truth value (or label) of the recovered CSI for training the decoder.

[0192] In application scenario 1, the AI ​​model deployed on the access network device side may be the decoder shown in Figure 4.

[0193] Please refer to Figure 5. Figure 5 is an example of obtaining training data in AI model-based CSI feedback according to the present application.

[0194] 501: Optionally, an access network device determines that training data for an AI model needs to be collected.

[0195] 502: The access network device sends first information to the UE, where the first information is for determining validity of candidate training data collected by the UE. Optionally, the determination result of the validity may be valid or invalid.

[0196] For example, the first information indicates constraints for determining the validity of candidate training data collected by the UE.

[0197] For the first information, constraints, etc., please refer to the relevant explanations in step 310. The details will not be described again here.

[0198] In application scenario 1, for example, the quality indicators of the measurement results may be the SINR of the training data and the amount of training data. The first information indicates an SINR threshold Q and a training data amount threshold N. The SINR criterion and the training data amount criterion (e.g., SINR is equal to or greater than Q and the amount of training data is equal to or greater than N) may be predefined in the protocol. In another example, the first information indicates the SINR threshold Q, the training data amount threshold N, the SINR criterion, and the training data amount criterion. For example, the first information indicates the threshold Q and the threshold N, and includes an information field indicating the criterion. For example, the information field includes one bit, and this one bit corresponds to the SINR criterion and the training data amount criterion. For example, a value "1" of the one bit indicates that "the SINR of the training data is equal to or greater than Q and the amount of training data is equal to or greater than N," and a value "0" of the one bit indicates that "the SINR of the training data is greater than Q and the amount of training data is greater than N." For example, the information field includes two bits b1b0, where b1 corresponds to the criterion of the SINR and b0 corresponds to the criterion of the amount of training data. For example, when the value of b1 is 1, it indicates that the SINR of the training data is equal to or greater than Q, and when the value of b1 is 0, it indicates that the SINR of the training data is less than Q. Representing the criterion of the amount of training data by b0 is similar. Details will not be described again. In yet another example, the first information indicates a threshold Q for the SINR of the training data and a threshold N for the amount of training data. In addition, the first information indicates a criterion for a part of the quality indicator, and the criterion for the other part of the quality indicator is predefined in the protocol. For example, the first information indicates a threshold Q and a threshold N. In addition, the first information further includes a 1-bit information field. When the value of the 1-bit is 1, it indicates that the SINR is equal to or greater than Q. When the value of the 1-bit is 0, it indicates that the SINR is less than Q.The criterion for determining the amount of training data is predefined in the protocol, for example, "the amount of training data is at least N." It should be understood that the above implementation is only an example for determining the constraint of the first information, and is not limited thereto.

[0199] For example, N may be an integer multiple of the batches during AI model training, or the amount of training data required for AI model convergence.

[0200] 503: The access network device sends a reference signal to the UE.

[0201] The UE obtains one or more measurement results by measuring a reference signal of an access network device. In this application, the measurement results may alternatively be referred to as reference signal measurement results or channel measurement results. The substitution expressions are also applicable to embodiments in different application scenarios. Details will not be described below.

[0202] Optionally, the measurement results include a channel response, for example a channel response matrix.

[0203] Additionally, optionally, the UE may obtain one measurement result through one measurement, in which case the candidate training data includes the one measurement result. Optionally, the UE obtains multiple measurement results through multiple measurements, in which case the candidate training data includes the multiple measurement results.

[0204] For example, the quality indicator of the measurement result may include, but is not limited to, one or more of: first path power, first path arrival delay, timing error group (TEG), average power of time-domain sampling points, phase difference between antenna ports, equivalent SINR of the entire band or sub-band, interference level of the entire band or sub-band, line of light (LOS) probability, inter-site synchronization error, or reliability of the measurement result. This is not limited. The indicator quality may be applicable to Application Scenario 1 or another application scenario described below. This is not limited. It should be understood that these quality indicators may be obtained by performing corresponding processing on the measurement result of the reference signal. The specific processing is not limited herein and may be, for example, any known or future processing.

[0205] It should be understood that before the access network device sends the reference signal to the UE, the access network device further sends air interface transmission configuration information corresponding to the reference signal to the UE. The air interface transmission configuration information indicates a related air interface configuration for sending the reference signal by the access network device. For example, the air interface transmission configuration information may include, but is not limited to, one or more of the following information: a transmission power of the reference signal, an amount of antenna ports used by the access network device to send the reference signal, a bandwidth of the reference signal, a frequency domain density of the reference signal, a duration of the reference signal, etc. Those skilled in the art may understand that the air interface transmission configuration information may further include other related information, which will not be listed one by one in this specification.

[0206] In this scenario, the candidate training data collected by the UE is one or more measurement results or one or more channel measurements obtained by measuring a reference signal from an access network device.

[0207] For example, in a 5G system, the reference signal may be a channel state information reference signal (CSI-RS).

[0208] 504: The UE determines validity of the collected candidate training data based on the first information.

[0209] For example, it is assumed that the candidate training data is a plurality of measurement results obtained by the UE by measuring a reference signal. The plurality of measurement results is candidate training data. The UE determines whether the plurality of measurement results includes valid candidate training data (or valid measurement results) based on the constraint. In the above example, if the constraint is "a quality indicator (e.g., SINR) is greater than or equal to a threshold Q, and the amount of candidate training data whose quality indicator is greater than or equal to the threshold Q is at least N," the UE determines whether the plurality of collected measurement results includes a measurement result whose quality is greater than or equal to the threshold Q. For simplicity, the measurement result whose quality indicator is greater than or equal to the threshold Q is hereinafter referred to as measurement result 1. If the UE determines that the plurality of collected measurement results includes measurement result 1, it further needs to determine whether the amount of measurement result 1 reaches N. If it is determined based on the constraint that valid measurement results have been collected, the UE determines that the currently collected candidate training data is valid. The valid candidate training data (i.e., the first training data, hereinafter also referred to as valid data) is a part of the measurement results that satisfies the constraint. For example, if the amount of measurement results 1 is P, where P is an integer equal to or greater than N, the P measurement results 1 are valid data collected this time, that is, the first training data.

[0210] Conversely, if the UE determines that the multiple collected measurement results do not include any measurement results that satisfy the constraint, for example, if the multiple collected measurement results include measurement result 1 whose SINR is greater than or equal to threshold Q but the amount of measurement result 1 is less than N, or if the SINRs of the multiple collected measurement results are all less than threshold Q, the UE determines that the candidate training data collected this time is invalid.

[0211] 505: The UE sends second information to the access network device based on determining the validity of the collected candidate training data, where the second information indicates a result of the validity determination.

[0212] If possible, the second information indicates that the candidate training data collected by the UE is valid. In an example, the second information may be valid collected candidate training data, for example, P measurement results 1 in the above example. In this example, P measurement results 1 are valid candidate training data, and P measurement results 1 also implicitly indicate that the candidate training data collected by the UE is valid. In another example, the UE sends the second information and valid candidate training data. In this example, the second information indicates that the candidate training data collected by the UE is valid. For example, the second information may include one bit. When the value of the one bit is "1," it indicates that the candidate training data collected by the UE is valid. In addition, the UE sends the valid candidate training data to the access network device. In contrast, the previous example can further reduce signaling overhead on the premise that the candidate training data collected by the UE can be indicated to be valid.

[0213] In another possible case, the second information indicates that the candidate training data collected by the UE is invalid. In an example, the second information may include one bit. When the value of the one bit is "0", it indicates that the candidate training data collected by the UE is invalid.

[0214] For example, the second information may be conveyed by using uplink control information (UCI) signaling. For example, the UCI includes one bit of information, which indicates whether the candidate training data collected by the UE is valid or invalid. Optionally, as shown in the above example, if the UE implicitly indicates that the collected candidate training data is valid by using valid candidate training data, the valid candidate training data may also be sent in the UCI. This is not limited.

[0215] 506: The access network device determines, based on the second information, whether the candidate training data of the UE is valid.

[0216] If possible, the second information indicates that the candidate training data collected by the UE is valid, corresponding to the determination result in step 505. In this case, the access network device further obtains the valid candidate training data collected by the UE from the UE. Furthermore, the access network device trains or updates the AI ​​model based on the valid candidate training data, as shown in step 507.

[0217] 507: The access network device trains the AI ​​model to obtain the AI ​​model or update the AI ​​model.

[0218] In another possible case, the second information indicates that the candidate training data collected by the UE is invalid. In a possible implementation of this case, the access network device either maintains the CSI feedback in the original AI model or switches to CSI feedback in a non-AI model. In an example, maintaining the original AI model may be for a scenario in which a trained AI model has been deployed on the access network device and training data is now collected for the purpose of updating the AI ​​model. Switching to a non-AI model may be for a scenario in which a trained AI model is not on the access network device and training data is now collected for training to obtain an AI model. In this scenario, if valid candidate training data is not obtained during the current collection, the access network device may switch to CSI feedback in a non-AI model. Two possible cases are shown in step 508.

[0219] 508: The access network device implements CSI feedback based on the original AI model or by switching to a non-AI model.

[0220] When the access network device performs step 507 or step 508, the training data collection procedure ends.

[0221] Optionally, in another possible case, the second information indicates that the candidate training data collected by the UE is invalid. After obtaining the second information, the access network device determines to re-collect training data, as shown in steps 509 and 510.

[0222] 509: The access network device determines to recollect training data for the AI ​​model.

[0223] 510: The access network device sends third information to the UE, where the third information indicates to the UE to recollect training data for the AI ​​model.

[0224] Optionally, in a possible implementation, the third information further indicates a maximum number k of validity checks, where k is a positive integer. Optionally, in another possible implementation, the maximum number k of validity checks may also be indicated by the first information. This is not limited. These two implementations are described in detail in the procedure of FIG. 3. The details will not be described again in this specification.

[0225] Optionally, when the training data of the AI ​​model is recollected, the access network device may update the air interface transmission configuration, and correspondingly, the UE recollects candidate training data of the AI ​​model based on the updated air interface transmission configuration.

[0226] 511: Optionally, the access network device sends air interface transmission configuration information to the UE, where the air interface transmission configuration information indicates an updated air interface transmission configuration.

[0227] The air interface transmission configuration information is described above. If the air interface transmission configuration is updated, the air interface transmission configuration information in step 511 indicates the updated air interface transmission configuration. For example, the air interface transmission configuration update may include updates to the transmit power of the reference signal, the number of antenna ports used by the access network device to send the reference signal, the bandwidth of the reference signal, the frequency domain density of the reference signal, the duration of the reference signal, etc. This is not limited to this. For example, the air interface transmission configuration update includes increasing the transmit power of the reference signal and increasing the frequency domain density of the reference signal. In this case, the access network device attempts to send the reference signal to the UE with higher transmit power and higher frequency domain density to enable the UE to acquire candidate training data that satisfies the constraints. The air interface transmission configuration is updated to ensure the accuracy of the AI ​​model-based CSI feedback over the air interface.

[0228] Of course, when the training data of the AI ​​model is re-collected, the original air interface transmission configuration may not be updated. In this case, the UE re-collects the candidate training data in the original air interface transmission configuration, determines the validity of the re-collected candidate training data based on the constraints, and indicates the validity determination result to the access network device.

[0229] 512: Optionally, the UE recollects training data for the AI ​​model.

[0230] It should be understood that in the procedure of recollecting training data, the validity determination of the recollected candidate training data and the indication of the determination result are similar to those in the above procedure of Figure 5. The details will not be described again. It should be understood that in the process of recollecting training data, the UE is constrained by the maximum number k of validity determinations.

[0231] Optionally, the maximum number k of validity determinations may be determined by the access network device based on the urgency of training data collection. For example, the urgency may be the interval between the last update of the AI ​​model and the current time point. For example, if the interval between the last update of the AI ​​model and the current time point is large and exceeds a threshold, the update requirement for the AI ​​model is deemed urgent. A larger interval indicates a higher probability that the channel environment will change, and the adaptability of the AI ​​model to the current channel environment may be reduced. Therefore, the update requirement becomes more urgent. In this case, the maximum number k of validity determinations may be set to a correspondingly larger value, which is expected to enable valid candidate training data to be obtained through multiple re-collections after one invalid collection. If the interval between the last update of the AI ​​model and the current time point is extremely small, for example, smaller than a threshold, the update requirement is deemed not urgent, and the maximum number k of validity determinations may be set to a small value. Optionally, the urgency determination criterion may also be implemented in another manner. This is not limited thereto.

[0232] It can be seen that the method for obtaining training data in AI model training provided in this application can be applied to AI model-based CSI feedback or CSI prediction scenarios, thereby reducing the waste of air interface resources in the AI ​​model training procedure. In addition, a UE may send invalid training data to an access network device, which may cause contamination of the training dataset of the AI ​​model, thereby affecting the training or update of the AI ​​model and causing inaccurate gain evaluation. This solution avoids the above-mentioned effects.

[0233] Application scenario 2 AI model-based positioning scenario Because uplink positioning is different from downlink positioning, the following describes the application of the technical solutions of the present application in uplink positioning and downlink positioning separately.

[0234] 1. Application to uplink positioning Please refer to Figure 6. Figure 6 is a diagram of a technical solution in an AI model-based uplink positioning scenario according to the present application. As shown in Figure 6, in uplink positioning, AI model training or updating is performed on the network side. A 5G system is used as an example. The AI ​​model may be deployed in a positioning device in a core network, for example, an LMF network element. In uplink positioning, the input of the AI ​​model is one or more channel responses (or channel measurement results) corresponding to one or more sounding reference signals, and the output of the AI ​​model is the location of the UE. There may be one or more transmitters, for example, UEs, of the one or more sounding reference signals, and there may be one or more receivers, for example, access network devices.

[0235] As shown in FIG. 6, in uplink positioning, when training an AI model used for positioning, the positioning device acquires, from the access network side, multiple measurement results obtained by one access network device measuring multiple sounding reference signals or by each of multiple access network devices measuring one or more sounding reference signals, and location information of third network elements. The multiple sounding reference signals may include multiple sounding reference signals from one third network element, or may include one or more sounding reference signals from each of multiple third network elements. The location information of the third network element that sends multiple sounding reference signals at different moments, or the location information of multiple third network elements that send one or more sounding reference signals sent at one or more moments, is used as the truth value (i.e., label) of the location information output by the AI ​​model.

[0236] Please refer to Figure 7. Figure 7 is an example of obtaining training data in AI model-based uplink positioning according to the present application.

[0237] 701: Optionally, a positioning device determines that training data for an AI model needs to be collected.

[0238] 702: The positioning device sends first information to the access network, the first information being for determining validity of candidate training data collected by the access network device. Optionally, the determination result may be valid or invalid.

[0239] For example, the first information may be carried in an interface message between the positioning device and the access network device. A 5G system is used as an example. When the positioning device is an LMF and the access network device is a gNB, the first information between the LMF and the gNB may be included in an NRPPa message.

[0240] 703: The access network device sends air interface transmission configuration information (e.g., air interface transmission configuration information #1) to a third network element, where the air interface transmission configuration information indicates an air interface transmission configuration to be used by the third network element to send the sounding reference signal.

[0241] In this embodiment, the third network element is a network element that can provide location information of the third network element. In an example, the third network element may be a location reference device. The location reference device may be considered a special network element and may be typically configured by a network vendor. For example, the network vendor may configure one or more of the location, transmission capability, reception capability, processing capability, etc. of the location reference device. The location reference device may provide location information of the location reference device to an access network device. For example, the third network element may be a reference UE or an automated guided vehicle (AGV). In another example, the third network element may alternatively be a normal UE. In this specification, the normal UE refers to the location reference device. After obtaining location information of the normal UE by using some positioning methods, the normal UE may provide location information to the access network device.

[0242] 704: The access network device measures the sounding reference signal from the third network element to obtain one or more measurement results.

[0243] A 5G system is used as an example. The sounding reference signal sent by the third network element may be a sounding reference signal (SRS).

[0244] In step 704, the access network device measures the sounding reference signal sent by the third network element to obtain one or more measurement results, and there is a correspondence between the one or more measurement results and the location information of the third network element. In an example, the third network element sends the sounding reference signal at location 1, and the access network device obtains measurement result 1 by measuring the sounding reference signal, where measurement result 1 corresponds to location 1. The third network element sends the sounding reference signal at location 2, and the access network device obtains measurement result 2 by measuring the sounding reference signal, where measurement result 2 corresponds to location 2. Optionally, in another example, the absolute location of the third network element does not change, but the surrounding environment of the third network element changes at different times. The access network device measures the sounding reference signal sent by the third network element at different times, and the obtained measurement results may also change. For example, measurement result 1 obtained by the access network device at time 1 corresponds to location 1 of the third network element, and measurement result 2 obtained at time 2 corresponds to location 1 of the third network element. Optionally, in yet another example, there may be multiple third network elements in this embodiment. The access network device individually measures sounding reference signals from the multiple third network elements to obtain multiple measurement results. In other words, each measurement result among the multiple measurement results corresponds to the location of one third network element among the multiple third network elements. Correspondingly, in step 705, the multiple third network elements individually provide their respective location information to the access network device or the positioning device.

[0245] 705: The third network element provides location information of the third network element.

[0246] In this embodiment, one third network element is used as an example for explanation. One piece of location information corresponds to one or more measurement results obtained by one or more access network devices by measuring one or more sounding reference signals sent by the third network element at locations corresponding to the location information. In an uplink positioning scenario, the candidate training data for the AI ​​model are the one or more measurement results and the location information of the third network element corresponding to the one or more measurement results. When the one or more access network devices determine that the collected candidate training data (i.e., one or more measurement results) are valid, the one or more access network devices individually provide the valid candidate training data to the positioning device.

[0247] Optionally, location information of the third network element corresponding to the valid candidate training data may be provided to the positioning device by the third network element via at least one of the one or more access network devices, as shown in step 705a. It should be understood that 705a is an implementation of step 705. The location information of the third network element may be visible or invisible to the at least one access network device.

[0248] Optionally, in this implementation, the third network element provides location information (not shown) of the subframe directly to the positioning device. The positioning device acquires valid candidate training data and location information corresponding to the valid candidate training data from one or more access network devices. It should be understood that there are multiple valid candidate training data and multiple location information of the third network element. The positioning device determines the correspondence between the valid candidate training data and the location information. The positioning device uses the location information as a label for the AI ​​model to train or update the AI ​​model, i.e., train in a new creation process or train in an update process. It should be understood that in this embodiment, the valid candidate training data (i.e., first training data) for the AI ​​model acquired by the positioning device includes one or more measurement results that satisfy the constraints and are among the measurement results acquired by the access network device by measuring the sounding reference signal sent by the third network element, and location information of the third network element that corresponds to each of the measurement results. The location information of the third network element is the output truth value, i.e., the label, of the AI ​​model.

[0249] 706: The access network device determines validity of the collected candidate training data based on the first information.

[0250] It should be noted that in step 706, the access network device specifically determines the validity of the measurements in the candidate training data.

[0251] For example, in application scenario 2, the quality indicator of the measurement result may include, but is not limited to, one or more of: first path power, first path arrival delay, timing error group (TEG), average power of time-domain sampling points, phase difference between antenna ports, equivalent SINR of the entire band or sub-band, interference level information of the entire band or sub-band, line of light (LOS) probability, indication information of inter-site synchronization error, or indication information of reliability of the measurement result. In addition, in the application scenario of positioning, the quality indicator of the label may be the distance between the locations of different samples.

[0252] For example, the quality indicator in the constraint may be the SINR and the amount of training data. In an example, the threshold for the amount of training data is N, where N may be an integer multiple of the batch or the minimum amount of training data required for the AI ​​model convergence. Optionally, in an uplink positioning scenario, the candidate training data for the AI ​​model collected by the access network device further includes a label, and the label is location information. For example, the quality indicator in the constraint may further include a quality indicator of the label. For example, the quality indicator of the label may be the distance between the locations of different samples. This is not limited herein.

[0253] For the validity determination, please refer to the description of step 504. The details will not be described again here.

[0254] 707: The access network device sends second information to the positioning device based on the determination result of the validity of the candidate training data, where the second information indicates the determination result of the validity.

[0255] For example, the second information may be included in an interface message between the access network device and the positioning device. Alternatively, the access network device sends an interface message to the positioning device, where the interface message includes the second information.

[0256] 708: The positioning device determines, based on the second information, whether the candidate training data collected by the access network device is valid.

[0257] If possible, the second information indicates that the candidate training data collected by the access network device is valid. In this case, the positioning device acquires valid candidate training data (i.e., first training data) collected by the access network device. Herein, the first training data specifically includes one or more measurement results that satisfy the constraints and location information of the third network element that corresponds to each of the one or more measurement results. Furthermore, the positioning device trains or updates the AI ​​model based on the valid candidate training data, as shown in step 709.

[0258] 709: The positioning device trains the AI ​​model to obtain the AI ​​model or update the AI ​​model.

[0259] In another possible case, the second information indicates that the candidate training data collected by the access network device is invalid. In this case, the positioning device may either maintain the original AI model or switch to a non-AI model, as shown in step 710.

[0260] 710: The positioning device performs uplink positioning based on the original AI model or by switching to a non-AI model.

[0261] Optionally, in another possible case, the second information indicates that the candidate training data collected by the access network device is invalid. In another possible implementation, the positioning device determines to re-collect the training data of the AI ​​model. In this case, step 711 and step 712 are further included.

[0262] 711: The positioning device determines to recollect training data.

[0263] 712: The positioning device sends third information to the access network device, where the third information indicates to the access network device to recollect training data for the AI ​​model.

[0264] Optionally, the third information further indicates a maximum number k of validity checks, where k is a positive integer. Optionally, the maximum number k of validity checks may alternatively be indicated by the first information. For details, please refer to the relevant description of the procedure shown in Figure 3. Details will not be described again.

[0265] In this application scenario, in an example, the maximum number k of validity determinations may be configured by the positioning device based on the urgency of the requirement of the training data of the AI ​​model. This is similar to that in Application Scenario 1. In an example, the criterion for determining the urgency may be determined based on the result error of estimating the location of the third network element by the current AI model or the interval from the time of the last update of the AI ​​model to the current time. For example, if the error of the result of estimating the location of the third network element by the positioning device based on the current AI model is large, for example, equal to or greater than a specified threshold, the requirement may be determined to be urgent. In this case, the maximum number k of validity determinations may be set to a large value. Conversely, if the error of the result of estimating the location of the third network element based on the current AI model is small, for example, smaller than a specified threshold, the requirement may be determined to be not urgent. In this case, the maximum number k of validity determinations may be set to a small value. The error of the result of estimating the AI ​​model is determined by splitting the training data collected in the previous AI model training into a training set and a validation set. Since the error of the training set is already extremely low, the error of the estimation result of the validation set is used as the criterion for determining whether the AI ​​model has become significantly invalid. In addition, this criterion may also be set based on the interval from the last update of the AI ​​model to the current time. For details, please refer to the description of application scenario 1. The details will not be described again.

[0266] Optionally, when determining to re-collect training data for the AI ​​model, the positioning device may indicate to the access network device to update the air interface transmission configuration used by the access network device to collect the training data, as shown in step 713.

[0267] 713: The access network device sends air interface transmission configuration information (eg, air interface transmission configuration information #2) to the third network element, where the air interface transmission configuration information indicates the updated air interface transmission configuration.

[0268] It should be understood that the air interface transmission configuration update in step 713 is an update to the air interface transmission configuration in step 703. For example, the update includes, but is not limited to, increasing the transmit power of the sounding reference signal, increasing the frequency domain density of the sounding reference signal, etc. It should be understood that the purpose of updating the air interface transmission configuration is for the access network device to attempt to collect candidate training data that meets the constraints and provide the candidate training data to the positioning device.

[0269] 714: The access network device recollects training data for the AI ​​model.

[0270] It can be seen that the method for obtaining training data in AI model training provided in this application can be applied to an AI model-based uplink positioning scenario, thereby reducing the waste of air interface resources in the AI ​​model training procedure. In addition, an access network device may send invalid training data to a positioning device, which may cause contamination of the training dataset of the AI ​​model, thereby affecting the training or update of the AI ​​model and causing inaccurate gain evaluation. This solution avoids the above effects.

[0271] It should be understood that in an uplink positioning scenario, the access network device is an example of a training data collection network element, and the positioning device is an example of an AI model training network element.

[0272] 2. Application to Downlink Positioning Please refer to Figure 8. Figure 8 is a diagram of a technical solution in an AI model-based downlink positioning scenario according to the present application. As shown in Figure 8, in downlink positioning, AI model inference is deployed on the UE side, while AI model training is deployed on a positioning device on the network side, for example, an LMF network element. The AI ​​model deployed on the positioning device uses the corresponding channel response obtained by the UE by measuring a reference signal as input and uses the location of the UE as output. A 5G system is used as an example. The reference signal may be a positioning reference signal and may be sent to the UE by one or more base stations (BSs).

[0273] Please refer to Figure 9. Figure 9 is an example of obtaining training data in AI model-based uplink positioning according to the present application.

[0274] 801: Optionally, the positioning device determines that training data for the AI ​​model needs to be collected.

[0275] The training data for the AI ​​model may come from measurements made by one UE on multiple reference signals, or from measurements made by each of multiple UEs on one or more reference signals. The multiple reference signals may be from one or more access network devices. This embodiment will be described in terms of communications between the positioning device and the UE or a particular UE among the one or more UEs.

[0276] 802: The positioning device sends first information to the UE, where the first information is for determining validity of candidate training data collected by the UE. Optionally, the determination result of the validity may be valid or invalid.

[0277] Optionally, in an example, the positioning device sends the first information to the UE via the access network device, as shown in steps 802a and 802b of Figure 9. In another example, the positioning device may alternatively send the first information to the UE directly through an interface between the positioning device and the UE. The positioning device sends information #1 to the access network device, where information #1 indicates to the access network device to send a positioning reference signal to the UE. Optionally, information #1 may alternatively be the first information. The access network device sends a positioning reference signal to the UE based on the trigger of information #1. The implementation shown in Figure 9 is merely an example.

[0278] 803: The access network device sends a positioning reference signal to the UE.

[0279] The UE measures positioning reference signals from the access network device, or from the access network device and another access network device, to obtain candidate training data, which is in particular one or more measurement results of the positioning reference signals and location information of the UE corresponding to the one or more measurement results.

[0280] It should be understood that before sending the PRS to the UE, the access network device further sends air interface transmission configuration information of the PRS to the UE to indicate the air interface transmission configuration of the PRS.

[0281] A 5G system is used as an example. The positioning reference signal sent to the UE by the access network device may be a PRS. The candidate training data is one or more measurement results obtained by the UE by measuring the PRS and location information of the UE corresponding to the one or more measurement results.

[0282] 804: The UE determines validity of the collected candidate training data based on the first information.

[0283] For example, the UE determines the validity of the collected candidate training data based on the constraint indicated by the first information. In this specification, in particular, the UE determines the validity of one or more measurement results. As with step 504, please refer to step 504 for understanding. Detailed descriptions will be omitted in this specification. In addition, for examples of quality indicators included in the constraints in the downlink positioning scenario, please refer to the description of the uplink positioning scenario. Details will not be described again in this specification.

[0284] 805: The UE sends second information to the positioning device, where the second information indicates the validity determination result.

[0285] In a possible case, the second information indicates that the candidate training data collected by the UE is valid. In an example, the second information may be first training data among the candidate training data collected by the UE. The first training data includes location information of the UE. Alternatively, the first training data is, in particular, measurement results that satisfy the constraint and location information of the UE corresponding to the measurement results. In another possible case, the second information indicates that the candidate training data collected by the UE is invalid.

[0286] Optionally, in step 805, the UE may directly send the second information to the positioning device through the interface between the UE and the positioning device, as shown in FIG. 9 . Alternatively, the UE may send the second information to the access network device, and then the access network device may send the second information to the positioning device. Alternatively, when the candidate training data collected by the UE is valid, the UE sends some information included in the second information, for example, measurement results included in the first training data that satisfy the constraint (i.e., valid measurement results), to the access network device and sends location information of the UE to the positioning device. The access network device then sends the measurement results that satisfy the constraint to the positioning device. Thus, the positioning device obtains the first training data, and the first training data includes valid measurement results and location information of the UE corresponding to the valid measurement results. This is not limited thereto.

[0287] 806: The positioning device determines, based on the second information, whether the collection of the candidate training data collected by the UE is valid.

[0288] If possible, the second information indicates that the candidate training data collected by the UE is valid. In this case, the second information may include the first training data, and the first training data includes location information of the UE. Furthermore, the positioning device trains or updates the AI ​​model based on the first training data, as shown in step 807.

[0289] 807: The positioning device trains the AI ​​model to obtain the AI ​​model or update the AI ​​model.

[0290] In another possible case, the second information indicates that the candidate training data collected by the UE is invalid. In this case, the access network device may maintain the beam management of the original AI model or switch to a non-AI model to perform downlink positioning, as shown in step 808.

[0291] 808: The positioning device performs positioning by keeping the original AI model or by switching to a non-AI model.

[0292] Optionally, in another possible case, the second information indicates that the candidate training data collected by the UE is invalid. In a possible implementation, the positioning device determines to re-collect training data, as shown in step 809.

[0293] 809: The positioning device determines to re-collect training data.

[0294] 810: The positioning device sends third information to the UE, where the third information indicates to the UE to recollect training data for the AI ​​model.

[0295] Optionally, the third information may further indicate a maximum number k of validity checks. Alternatively, the first information indicates a maximum number k of validity checks.

[0296] Optionally, when the training data of the AI ​​model is recollected, the positioning device may indicate to the access network device that the air interface transmission configuration should be updated. For example, the positioning device may send information #2 to the access network device, where the information #2 indicates to the access network device that the training data should be recollected. In response, the access network device may send air interface transmission configuration information corresponding to the updated air interface transmission configuration to the UE, as shown in step 811.

[0297] 811: The access network device sends air interface transmission configuration information to the UE, where the air interface transmission configuration information indicates an updated air interface transmission configuration.

[0298] The UE measures the positioning reference signals sent by the access network device based on the updated air interface transmission configuration and recollects training data for the AI ​​model.

[0299] 812: The UE recollects training data for the AI ​​model.

[0300] Optionally, the UE in this embodiment may be a location reference device or a normal UE, which is not limited. For the location reference device or the normal UE, please refer to the description of step 703. Details will not be described again.

[0301] In addition, it should be understood that the UE in this embodiment is an example of a training data collection network element, and the positioning device is an example of an AI model training network element.

[0302] It can be seen that the method for obtaining training data in AI model training provided in this application can be applied to an AI model-based downlink positioning scenario, thereby reducing the waste of air interface resources in the AI ​​model training procedure. In addition, a terminal device (e.g., a location reference device) may send invalid training data to a positioning device, which may cause contamination of the training data set of the AI ​​model, thereby affecting the training or update of the AI ​​model and causing inaccurate gain evaluation. This solution avoids the above effects.

[0303] Application Scenario 3 AI model-based beam management For example, in Application Scenario 3, AI model training is deployed on the access network device side. The access network device needs to obtain training data (e.g., one or more measurement results of the reference signal) obtained by the UE side by measuring a reference signal, and use the training data obtained from the UE side to train or update the AI ​​model. In this application scenario, the label of the AI ​​model is reference signal information corresponding to the K optimal measurement results. Alternatively, in Application Scenario 3, the reference signal information corresponding to the K optimal measurement results relates to the K beams and may be replaced with information corresponding to the K optimal measurement results, for example, the IDs of the K beams.

[0304] It can be seen that in 5G systems, high-frequency bands above 6 GHz will be introduced for data communication. Compared with mid- and low-frequency bands below 6 GHz, the continuously available bandwidth of the high-frequency band spectrum will be larger and the center frequency will be higher. Therefore, higher transmission speeds and greater system capacity can be achieved. However, due to the weak penetration ability of high-frequency signals (e.g., millimeter waves) and the strong path fading effect, the propagation distance of high-frequency signals is limited, resulting in poor coverage. Thanks to large-scale antenna technology, high-frequency communication systems typically use a large number of antennas for beamforming, which can achieve clear beam gain and compensate for the limited propagation distance caused by high-frequency propagation characteristics. However, to design accurate beamforming, the base station needs to obtain accurate channel information from the terminal. Obtaining channel information for such a large-scale antenna array requires the consumption of a large amount of air interface overhead, which is unacceptable in practical systems. Experiments have shown that high-frequency wireless channels have clear sparsity, i.e., the main energy of the channel is concentrated in a limited number of paths. For example, when there is a direct line-of-sight (LoS) path between a signal transmitter and receiver without any obstacles, the primary energy between the receiver and the transmitter is concentrated on the direct line-of-sight path. When there is a non-line-of-sight (NLOS) obstacle between the transmitter and receiver, the primary energy is concentrated on the path that can be reached through a single reflection. Typically, each path has a different angle of incidence and exit. Therefore, the transmitter and receiver in a high-frequency communication system only need to align their beam directions with the angle of incidence and exit angle of the channel's primary path to obtain the most channel transmission energy and complete the communication.

[0305] For high-frequency communication systems, it is assumed that the transmitter has a total of S antennas and the receiver has R antennas, and the antenna configuration may include linear antennas or planar array antennas. The transmitter and receiver use different precoding weights to multiply the transmitter and receiver antennas to precode the transmitted signal, thereby achieving a beamforming effect. For example, the downlink signal transmission model is as follows:

[0306] Y=VHWX+N The receiver's receive precoding matrix is ​​V and the channel response is H. The transmitter's transmit precoding matrix is ​​W, the transmitted signal is X, and the noise is N. The signal received by the receiver is Y. The transmitter's transmit precoding matrix has the form W = [W1, W2, … W M ] and W i is the precoding weight for the i-th antenna at the transmitter. Similarly, the received precoding matrix at the receiver has the form V = [V1,V2,…V R ] and V i is the precoding weight of the i-th antenna of the receiver. The signal obtained by precoding the transmitted signal X by using W is WX, and WX is the final transmitter signal of the transmitter. WX has a beamforming effect in space. WX can be classified into a reference signal and a data signal according to the difference between the information carried on X. Usually, the reference signal is sent in the beam management process. Possible types of reference signals include SSB, CSI-RS, SRS, phase-tracking reference signal (PTRS), demodulation reference signal (DMRS), etc.

[0307] Because the angle of the main path of a channel can be within a wide range, e.g., 0 to 360 degrees, and each precoding matrix W can only cover a specific angle range in space and corresponds to one beamformed beam, multiple precoding matrices W need to be designed to ensure good signal coverage. Multiple precoding matrices W with different directivity angles form a codebook. Both the transmitter and the receiver maintain their own codebooks. In the beam management process, the transmitter and the receiver implement angle matching between the transmitter and the receiver by traversing and scanning their own codebooks. For example, the transmitter's codebook contains 64 precoding matrices corresponding to 64 beamformed beams. The receiver's codebook contains four precoding matrices corresponding to four beamformed beams. Therefore, a total of 256 (64*4) scanning operations are required to determine the optimal beamformed beam pair for the receiver and the transmitter, resulting in significant scanning overhead and delay.

[0308] In beam management techniques based on AI models, sparse beam scanning can be implemented, and the overhead and delay of beam scanning can be significantly reduced. For high-frequency communication systems, it is assumed that the transmitter has a total of S antennas, the receiver has R antennas, the transmitter codebook has S precoding matrices (S beamformed beams), and the receiver codebook has R precoding matrices (R beamformed beams). For each receive beamformed beam, any transmit beamformed beam can form a transmit / receive beam pair with the receive beamformed beam. Therefore, the process of determining the optimal receive / transmit beam pair can be split as follows: For a receive beam, transmitter beam scanning is performed to determine the matched optimal transmit beam, and then this process is repeated for the remaining R-1 receive beams to determine the globally optimal transmit / receive beam pair. Similarly, for each transmit beam, any receive beam can form a transmit / receive beam pair with the transmit beam. Therefore, the process of determining the optimal transmit / receive beam pair can be split as follows: For a transmit beam, receiver beam scanning is performed to determine the best matched transmit beam, and then this process is repeated for the remaining S-1 transmit beams to determine the globally best transmit / receive beam pair. Therefore, the following uses transmitter beam scanning as an illustrative example.

[0309] See Figure 10. Figure 10 is a diagram of the AI-assisted sparse beam scanning process.

[0310] The precoding matrix in the transmitter's codebook is assumed to correspond to a first value, e.g., 64 beamformed beams. In conventional solutions, all 64 beams need to be scanned to determine the optimal beam. However, in an AI-assisted (i.e., AI model-based) sparse beam scanning solution, only some beams in the codebook, e.g., the second value beams marked in FIG. 10, e.g., 16 beams, need to be scanned. There may be many ways to select some beams from the codebook to scan, and each combination of selected beams is called a sparse beam pattern. The transmitter performs precoding by using the precoding matrix in the sparse beam pattern and sends a reference signal. The receiver inputs the measurement results of the reference signal into a neural network for AI beam prediction, and the neural network outputs indices of K beams, also called the indices of the top K beams. Note that the K beams are K of all 64 beamformed beams in the codebook and are not limited to the K beamformed beams included in the sparse beam pattern. The receiver feeds back the indices of the top K beams to the transmitter. Optionally, if K>1, the transmitter scans only the K beamformed beams and sends beamformed reference signals. The receiver measures the energy of the K reference signals by using an energy detection method and selects the beam with the strongest energy as the optimal beam.

[0311] The following describes the application of the method for acquiring training data in AI model training provided in this application to beam management based on sparse beam scanning technology. In this application, the transmitter is an access network device, for example, a base station. The receiver is a UE. The UE measures a reference signal from the access network device, obtains multiple measurement results, determines top K beam indexes, and feeds back the multiple measurement results and the top K beam indexes to the access network device.

[0312] See Figure 11. Figure 11 is an example of obtaining training data in AI model-based beam management according to the present application.

[0313] 901: Optionally, an access network device determines that training data for an AI model needs to be collected.

[0314] 902: The access network device sends first information to the UE, where the first information is for determining validity of candidate training data collected by the UE. Optionally, the determination result of the validity may be valid or invalid.

[0315] 903: The access network device sends a plurality of reference signals to the UE.

[0316] The UE measures a plurality of reference signals from the access network device to obtain a plurality of measurement results, i.e., candidate training data. Optionally, the plurality of reference signals corresponds to the above-mentioned first value, for example, 64 beamformed beams.

[0317] Optionally, in application scenario 3, the reference signal may be CSI-RS and / or SSB. In this example, CSI-RS and / or SSB are used by the UE to perform channel measurements.

[0318] 904: The UE determines, based on the first information, the validity of the collected candidate training data.

[0319] 905: The UE sends second information to the access network device, where the second information indicates the validity determination result.

[0320] If possible, the second information indicates that the candidate training data collected by the UE is valid. In an example, the second information may be first training data, which includes reference signal information corresponding to K best measurement results among one or more measurement results, where K is an integer equal to or greater than 1. For example, the reference signal information corresponding to the K best measurement results may be understood as the indexes of the top K beams in FIG. 10. In another example, the second information may be first training data, which includes L measurement results among the plurality of measurement results and information on L reference signals corresponding to the L measurement results. The L measurement results are valid measurement results, i.e., measurement results that satisfy the constraints. In another possible case, the second information indicates that the candidate training data collected by the UE is invalid.

[0321] In this application scenario, the reference signal may also be replaced by a "beam." This is not limited.

[0322] 906: The access network device determines, based on the second information, whether the candidate training data collected by the UE is valid.

[0323] If possible, the second information indicates that the candidate training data collected by the UE is valid. In this case, the access network device acquires the first training data from the UE. Furthermore, the access network device trains or updates the AI ​​model based on the first training data, as shown in step 907.

[0324] 907: The access network device trains the AI ​​model based on the first training data to obtain an AI model or update the AI ​​model.

[0325] In another possible case, the second information indicates that the candidate training data collected by the UE is invalid. In this case, the access network device may maintain the beam management of the original AI model or switch to a non-AI model to perform beam management, as shown in step 908.

[0326] 908: The access network device keeps the original AI model unchanged or switches to a non-AI model to perform beam management.

[0327] Optionally, in another possible case, the second information indicates that the candidate training data collected by the UE is invalid. In a possible implementation, the access network device determines to re-collect the training data, as shown in step 909.

[0328] 909: The access network device determines to re-collect training data.

[0329] 910: The access network device sends third information to the UE, where the third information indicates to the UE to recollect training data for the AI ​​model.

[0330] Optionally, the third information further indicates the maximum number k of validity checks. Alternatively, the maximum number k of validity checks may also be indicated by the first information. This is not limited thereto.

[0331] Optionally, when the training data of the AI ​​model is recollected, the access network device may update the air interface transmission configuration, and correspondingly, the UE recollects candidate training data of the AI ​​model based on the updated air interface transmission configuration.

[0332] 911: The access network device sends air interface transmission configuration information to the UE, where the air interface transmission configuration information indicates an updated air interface transmission configuration.

[0333] 912: The UE recollects training data for the AI ​​model.

[0334] Similar to the above two application scenarios, in the process of recollecting candidate training data, the UE is limited by the maximum number k of validity determinations. Optionally, the maximum number k of validity determinations may be determined by the access network device based on the urgency of training data collection. For example, the criterion for determining the urgency may be the time period from the last update of the AI ​​model to the current time, or may be set by the access network device based on the current AI model for the error of the result of estimating the optimal measurement result (or the optimal beam). The error of the result of estimating the AI ​​model is determined by splitting the training data collected in the previous AI model training into a training set and a validation set. Because the error of the training set is already extremely low, the error of the estimation result of the validation set is used as a criterion for determining whether the AI ​​model has become significantly invalid. For example, if the error of the result of the access network device predicting information about the reference signal corresponding to the optimal measurement result when the UE receives the reference signal based on the current AI model has a large error, for example, equal to or greater than a specified threshold, the requirement may be determined to be urgent. In this case, the maximum number k of validity determinations may be set to a large value. On the other hand, if the error of the result of predicting information about the reference signal corresponding to the optimal measurement result when the UE receives the reference signal based on the current AI model is small, for example, smaller than a specified threshold, it may be determined that the requirement is not urgent. In this case, the maximum number k of validity determinations may be set to a small value. In addition, this determination criterion may also be set based on the interval from the time of the last update of the AI ​​model to the current time. For details, please refer to the description of Application Scenario 1. Details will not be described again.

[0335] It can be seen that the method for obtaining training data in AI model training provided in the present application can be applied to an AI model-based beam management scenario, thereby reducing the waste of air interface resources in the AI ​​model training procedure. In addition, this solution avoids contamination of the training dataset of the AI ​​model caused by the UE sending invalid training data to the access network device.

[0336] In the method procedure diagrams in Figures 3 to 11, the sequence numbers of the steps are only intended to clearly explain the technical solutions of the present application and should not constitute limitations on the specific implementation of the method. These steps may be expanded into more steps or combined into fewer steps according to different specific implementations. This is not limited. In addition, the steps shown with dashed lines in Figures 3 to 11 are optional steps.

[0337] The above describes in detail the method for obtaining training data in AI model training provided in the present application. Based on the same technical concept, please refer to Figure 12. The present application provides a communication device 1000.

[0338] 12 , the communication device 1000 includes a processing module 1001 and a communication module 1002. The communication device 1000 may be a terminal device, or may be a communication device, such as a chip or circuit, that is used in a terminal device or by matching with a terminal device and can implement a communication method implemented on the terminal device side. Alternatively, the communication device 1000 may be a network device, or may be a communication device, such as a chip or circuit, that is used on the network device side or by matching with a network device side and can implement a communication method implemented on the network device side. For example, the network device side may be, for example, an access network device or a positioning device in the method embodiments of the present application.

[0339] The communication module may also be referred to as a transceiver module, a transceiver, a transceiver machine, a transceiver device, etc. The processing module may also be referred to as a processor, a processing board, a processing unit, a processing device, etc. Optionally, the communication module is configured to perform transmitting and receiving operations on the terminal device side or the network device side in the above method. A component configured to implement a receiving function in the communication module may be considered a receiving unit, and a component configured to implement a transmitting function in the communication module may be considered a transmitting unit. In other words, the communication module includes a receiving unit and a transmitting unit.

[0340] When the communication apparatus 1000 is used in a terminal device, the processing module 1001 may be configured to implement the processing functions of the terminal device in the embodiments of Figures 3 to 11, and the communication module 1002 may be configured to implement the receiving and transmitting functions of the terminal device in the embodiments of Figures 3 to 11. Alternatively, the communication apparatus may be understood with reference to the third aspect and possible designs of the third aspect in the Summary of the Invention.

[0341] When the communication apparatus 1000 is used in a network device, the processing module 1001 may be configured to implement the processing functions of the network device (e.g., an access network device or a positioning device) in the embodiments of Figures 3 to 11, and the communication module 1002 may be configured to implement the receiving and transmitting functions of the network device in the embodiments of Figures 3 to 11. Alternatively, the communication apparatus may be understood with reference to the fourth aspect and possible designs of the fourth aspect in the Summary of the Invention.

[0342] It should be noted that the first network element or the second network element shown in Figure 3 is specifically a terminal device or a network device (e.g., an access network device or a positioning device), which is described in detail in the above method embodiments in various different application scenarios. It may be understood that the first network element is a terminal device or a network device with reference to specific embodiments. The details will not be described again in this specification.

[0343] In addition, it should be noted that the communication module and / or the processing module may be implemented by using virtual modules. For example, the processing module may be implemented by using a software functional unit or a virtual device, and the communication module may be implemented by using a software function or a virtual device. Alternatively, the processing module or the communication module may be implemented by using an entity device. For example, if the device is implemented by using a chip / chip circuit, the communication module may be an input / output circuit and / or a communication interface, performing an input operation (corresponding to the above-mentioned receiving operation) and an output operation (corresponding to the above-mentioned sending operation). The processing module is an integrated processor, a microprocessor, or an integrated circuit.

[0344] The division into modules in this application is an example and is merely a division into logical functions, and may be divided in other ways in actual implementation. In addition, the functional modules in the examples of this application may be integrated into one processor, and each module may exist physically alone, or two or more modules may be integrated into one module. The integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0345] Based on the same technical concept, please refer to Fig. 13. The present application further provides a communication device 1100. Optionally, the communication device 1100 may be a chip or a chip system. Optionally, in the present application, a chip system may include a chip, or may include a chip and another individual component.

[0346] The communication device 1100 may be configured to implement the functionality of any network element in the communication system described in the above examples. The communication device 1100 may include at least one processor 1110. Optionally, the processor 1110 is coupled to a memory. The memory may be located in the device. Alternatively, the memory may be integrated with the processor. Alternatively, the memory may be located external to the device. For example, the communication device 1100 may further include at least one memory 1120. The memory 1120 stores computer programs, computer programs or instructions, and / or data necessary to implement any one of the above examples. The processor 1110 may execute a computer program stored in the memory 1120 to complete the method in any one of the above examples.

[0347] The communication device 1100 may further include a communication interface 1130, through which the communication device 1100 may exchange information with another device. For example, the communication interface 1130 may be a transceiver, a circuit, a bus, a module, a pin, or another type of communication interface. When the communication device 1100 is a chip-type device or circuit, the communication interface 1130 in the device 1100 may alternatively be an input / output circuit that inputs information (also referred to as received information) and outputs information (also referred to as transmitted information). The processor may be an integrated processor, a microprocessor, an integrated circuit, or a logic circuit. The processor may determine output information based on the input information.

[0348] A coupling in this application may be an indirect coupling or communication connection between devices, units, or modules in an electrical, mechanical, or other form, and is used for exchanging information between the devices, units, or modules. The processor 1110 may operate in cooperation with the memory 1120 and the communication interface 1130. The specific connection medium between the processor 1110, the memory 1120, and the communication interface 1130 is not limited in this application.

[0349] Optionally, as shown in Figure 13, the processor 1110, memory 1120, and communication interface 1130 are connected to each other by using a bus 1140. The bus 1140 may be a peripheral component interconnect (PCI) bus, an extended industry standard architecture (EISA) bus, etc. The bus may be categorized as an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used to represent a bus in Figure 13, but this does not imply that there is only one bus or only one type of bus.

[0350] In this application, a processor may be a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component, which may implement or perform the methods, steps, and logic block diagrams disclosed in this application. A general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed with reference to this application may be implemented directly by a hardware processor, or may be implemented by a combination of hardware and software modules in a processor.

[0351] In this application, memory may be non-volatile memory, such as a hard disk drive (HDD) or a solid-state drive (SSD), or may be volatile memory, such as random access memory (RAM). Memory is any other medium capable of carrying or storing expected program code in the form of instructions or data structures and accessible by a computer, but is not limited to such. Alternatively, memory in this application may be a circuit or any other device capable of implementing a storage function and configured to store program instructions and / or data.

[0352] In a possible implementation, the communication device 1100 may be used on the network device side, for example, as an access network device or a positioning device in the embodiments of the present application. In particular, the communication device 1100 may be a network device or a device capable of supporting a network device in implementing corresponding functions of the network device side in any one of the above examples. The memory 1120 stores computer programs (or instructions) and / or data for implementing functions of the network device side in any one of the above examples. The processor 1110 may execute the computer program stored in the memory 1120 to complete the method performed by the network device side in any one of the above examples. When the communication device is used in an access network device, the communication interface in the communication device 1100 may be configured to interact with a terminal device and send information to or receive information from the terminal device. In addition, optionally, the communication interface in the communication apparatus 1100 may be further configured to interact with a core network device, for example, to interact with a positioning device (e.g., an LMF network element) to send information to or receive information from the positioning device.

[0353] In another possible implementation, the communication device 1100 may be used in a terminal device. In particular, the communication device 1100 may be a terminal device or a device capable of supporting a terminal device in implementing the functions of the terminal device in any one of the above examples. The memory 1120 stores computer programs (or instructions) and / or data for implementing the functions of the terminal device in any one of the above examples. The processor 1110 may execute the computer programs stored in the memory 1120 to complete the method performed by the terminal device in any one of the above examples. When the communication device is used as a terminal device, the communication interface in the communication device 1100 may be configured to interact with a network device side (e.g., an access network device) and send information to or receive information from the network device side.

[0354] The communication device 1100 provided in this example may be used on a network device side (e.g., an access network device or a positioning device) to complete a method performed by the network device side, or may be used on a terminal device to complete a method performed by the terminal device. Therefore, for technical effects that can be achieved by this embodiment, please refer to the above method embodiment. Details will not be described again in this specification.

[0355] Based on the above example, the present application provides a communication system including a network device and a terminal device. In an example, the communication system includes an access network device and a terminal device. In another example, the communication system includes a positioning device, an access network device, and a terminal device. The access network device and the terminal device, or the positioning device, the access network device, and the terminal device, may implement the communication methods provided in the examples shown in Figures 3 to 11.

[0356] All or part of the technical solutions provided in this application may be implemented by using software, hardware, firmware, or any combination thereof. When software is used to implement an embodiment, all or part of the embodiment may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the procedures or functions according to this application are generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, a terminal device, an access network device, or another programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from a computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website, computer, server, or data center to another website, computer, server, or data center via a wired (e.g., coaxial cable, fiber optic, or digital subscriber line (DSL)) or wireless (e.g., infrared, radio, or microwave) method. A computer-readable storage medium may be any available medium accessible to a computer, or a data storage device, e.g., a server or a data center, that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disk drives, or magnetic tapes), optical media (e.g., digital video discs (DVDs)), semiconductor media, etc.

[0357] In this application, cross-references may be made between examples where there is no logical contradiction. For example, cross-references may be made between methods and / or terms in method embodiments, cross-references may be made between functions and / or terms in apparatus embodiments, and cross-references may be made between functions and / or terms in apparatus examples and method examples.

[0358] As used herein, terms such as "component," "module," and "system" are used to refer to computer-related entities, hardware, firmware, a combination of hardware and software, software, or software running on it. For example, a component may be, but is not limited to, a process running on a processor, a processor, an object, an executable, a thread of execution, a program, and / or a computer. As illustrated through the use of figures, both a computing device and an application running on a computing device may be a component. One or more components may reside within a process and / or thread of execution; a component may be located on one computer and / or distributed between two or more computers. In addition, these components may execute from various computer-readable media having various data structures stored thereon. For example, components may communicate by using local and / or remote processing and based on signals having, for example, one or more data packets (e.g., data from two components interacting with another component in a local system, data from two components interacting with another component in a distributed system, and / or data from two components interacting with another component over a network such as the Internet that interacts with other systems by using signals).

[0359] Those skilled in the art may realize that, in combination with the examples described in the embodiments disclosed herein, the units and algorithm steps may be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether a function is implemented by hardware or software depends on the specific application and the design constraints of the technical solution. Those skilled in the art may use different methods to implement the described functions for each specific application, but the implementation should not be considered to go beyond the scope of this application.

[0360] For the sake of convenient and brief description, those skilled in the art can clearly understand that for the detailed work processes of the above systems, devices and units, please refer to the corresponding processes in the above method embodiments, and the details will not be described again in this specification.

[0361] In some embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods may be implemented in other manners. For example, the device embodiments described above are merely schematic. For example, the unit division is merely a logical functional division, and actual implementation may involve other divisions. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not implemented. In addition, the shown or discussed mutual couplings or direct couplings or communication connections may be implemented by using some interfaces. Indirect couplings or communication connections between devices or units may be implemented in electronic, mechanical, or other forms.

[0362] The units described as separate parts may or may not be physically separate, and the parts shown as units may or may not be physical units, and may be located in one location or distributed over multiple network units. Some or all of the units may be selected based on actual requirements to achieve the objectives of the solutions of the embodiments.

[0363] Additionally, the functional units in the embodiments of the present application may be integrated into one processing unit, or each of the units may exist physically alone, or two or more units may be integrated into one unit.

[0364] When a function is implemented in the form of a software functional unit and sold or used as an independent product, the function may be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present application, or a part that contributes to the current technology, or a part of the technical solution may be implemented in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for instructing a computer device (which may be a personal computer, a server, or a network device) to perform all or part of the steps of the method described in the embodiments of the present application. The above storage medium includes any medium that can store program code, such as a USB flash drive, a removable hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0365] The above description is merely a specific implementation of the present application and is not intended to limit the scope of protection of the present application. Any modifications or replacements that are easily understood by those skilled in the art within the technical scope disclosed in the present application shall fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be subject to the scope of protection of the claims.

Claims

1. 1. A method for obtaining training data in artificial intelligence (AI) model training, the method being performed by a first network element or a chip used in the first network element, the method comprising: receiving first information from a second network element, the first information being for determining validity of collected candidate training data, and a result of the validity determination comprising valid or invalid; collecting candidate training data for an AI model; sending second information to the second network element based on the candidate training data and the first information, the second information indicating the determination of validity; A method comprising:

2. 2. The method of claim 1, wherein the second information includes first training data, the second information indicates that the collected candidate training data is valid, and the first training data is valid data among the candidate training data.

3. The method of claim 1 , wherein the second information indicates that the collected candidate training data is invalid.

4. The method of claim 1 , wherein the first information is for determining constraints for determining the validity of the collected candidate training data.

5. The method comprises: determining that the candidate training data is valid if it is determined that the candidate training data includes the first training data that satisfies the constraint; or determining that the candidate training data is invalid if it is determined that the candidate training data does not include the first training data that satisfies the constraint; The method of claim 4 further comprising:

6. If the collected candidate training data is invalid, the method includes: receiving third information from the second network element, the third information indicating to recollect candidate training data for the AI ​​model; The method of any one of claims 3 to 5, further comprising:

7. The method comprises: determining air interface transmission configuration information, the air interface transmission configuration information corresponding to an updated air interface transmission configuration, the air interface transmission configuration information indicating collecting the candidate training data for the AI ​​model based on the updated air interface transmission configuration; The updated information about the air interface transmission configuration may include: the transmit power of the reference signal, the amount of antenna ports used for the reference signal, the bandwidth of the reference signal, the frequency domain density of the reference signal, or Reference signal period 7. The method of claim 6, further comprising one or more of the following updates:

8. The method according to claim 6 or 7, wherein the third information further indicates a maximum number of times k for determining the validity, where k is a positive integer.

9. The method according to claim 6 or 7, wherein the first information further indicates a maximum number of times k for determining the validity, where k is a positive integer.

10. The method comprises: collecting the candidate training data for the AI ​​model based on the updated air interface transmission configuration; When the maximum number k of times of validity determination is reached and it is determined that the result of the kth validity determination is invalid based on the first information, stopping the step of collecting the candidate training data for the AI ​​model. The method of claim 8 or 9, further comprising:

11. The method comprises: sending fourth information to the second network element when it is determined that a result of the jth validity determination is valid based on the first information before the maximum number k of validity determinations is exceeded, the fourth information including second training data, the fourth information indicating that the result of the jth validity determination is valid, the second training data including valid data among the candidate training data on which the jth validity determination was performed, j being less than or equal to k and a positive integer; The method of claim 10 further comprising:

12. The step of collecting the candidate training data for the AI ​​model comprises: measuring a reference signal from the second network element to obtain one or more measurement results, wherein the candidate training data for the AI ​​model includes the one or more measurement results; or measuring a reference signal from a third network element to obtain one or more measurement results, wherein the candidate training data for the AI ​​model includes the one or more measurement results; 12. The method of any one of claims 1 to 11, comprising:

13. the first network element is a terminal device or a chip used in the terminal device, and the second network element is an access network device or a chip used in the access network device; The first network element measures the reference signal from the second network element to obtain the one or more measurements. The method of claim 12.

14. 14. The method of claim 13, wherein the first training data further includes reference signal information or beam information corresponding to K best measurement results of the one or more measurement results, where K is an integer greater than or equal to 1.

15. the first network element is an access network device or a chip used in the access network device, and the second network element is a positioning device or a chip used in the positioning device; the first network element measures a sounding reference signal from the third network element to obtain the one or more measurements; The first training data further includes one or more location information of the third network element. The method of claim 12.

16. The first network element is a terminal device or a chip used in the terminal device, and the second network element is a positioning device or a chip used in the positioning device; the first network element measures a positioning reference signal from the third network element to obtain the one or more measurement results, the third network element being an access network device; The first training data further includes one or more location information of the first network element. The method of claim 12.

17. The constraint is: a quality index threshold and a criterion for determining said quality index; a training data quantity threshold that satisfies a quality index criterion and a criterion for determining the quantity of said training data; or Indication of the maximum duration of training data collection corresponding to one validation 17. The method of any one of claims 4 to 16, comprising one or more of:

18. The first information is quality indicator thresholds, quality indicator criteria, A threshold for the amount of training data that satisfies the quality metric criteria; A measure of the amount of training data that meets the quality metric criteria, or Maximum duration of candidate training data collection corresponding to one validation 18. The method of any one of claims 4 to 17, wherein the method comprises one or more of:

19. The constraint is based on an application scenario of the AI ​​model, and the application scenario of the AI ​​model is AI model-based CSI feedback or CSI prediction, AI model-based positioning, or AI model-based beam management 19. The method of any one of claims 4 to 18, comprising one or more of:

20. 1. A method for obtaining training data in AI model training, the method being performed by a second network element or a chip used in the second network element, the method comprising: sending first information to a first network element, the first information being for an AI model to determine validity of candidate training data collected by the first network element, and a result of the validity determination comprising valid or invalid; receiving second information from the first network element, the second information indicating the determination of validity; A method comprising:

21. 21. The method of claim 20, wherein the second information includes first training data, the second information indicates that the candidate training data collected by the first network element is valid, and the first training data is valid data among the candidate training data.

22. 21. The method of claim 20, wherein the second information indicates that the candidate training data collected by the first network element is invalid.

23. 23. The method of any one of claims 20 to 22, wherein the first information is for determining constraints for determining the validity of the candidate training data collected by the first network element.

24. If the candidate training data includes the first training data that satisfies the constraint, the candidate training data is valid; or If the candidate training data does not contain the first training data that satisfies the constraint, the candidate training data is invalid.

24. The method of claim 23.

25. If the second information indicates that the candidate training data collected by the first network element is invalid, the method further comprises: sending third information to the first network element, the third information indicating to the first network element to recollect candidate training data for the AI ​​model; 25. The method of any one of claims 22 to 24, further comprising:

26. The method comprises: determining air interface transmission configuration information, the air interface transmission configuration information corresponding to an updated air interface transmission configuration, the air interface transmission configuration information indicating to the first network element to collect the candidate training data for the AI ​​model based on the updated air interface transmission configuration; The updated information about the air interface transmission configuration may include: the transmit power of the reference signal, the amount of antenna ports used for the reference signal, the bandwidth of the reference signal, the frequency domain density of the reference signal, or Reference signal period 26. The method of claim 25, further comprising one or more of the following updates:

27. 27. The method according to claim 25 or 26, wherein the third information further indicates a maximum number of times k for determining the validity, where k is a positive integer.

28. 27. The method according to claim 25 or 26, wherein the first information further indicates a maximum number of times k for determining the validity, where k is a positive integer.

29. The method comprises: receiving fourth information from the first network element, the fourth information including second training data, the fourth information indicating that a determination result of a j-th validity determination performed by the first network element is valid, the second training data being valid data among the candidate training data on which the j-th validity determination was performed, j being less than or equal to k, and j being a positive integer; 29. The method of claim 27 or 28, further comprising:

30. The second network element is an access network device or a chip used in the access network device, and the first network element is a terminal device or a chip used in the terminal device, and the method includes:

30. The method of any one of claims 20 to 29, further comprising sending a reference signal to the first network element, the reference signal being used by the first network element to obtain one or more measurements corresponding to the reference signal, and the candidate training data for the AI ​​model including the one or more measurements.

31. 31. The method of claim 30, wherein the first training data further comprises reference signals corresponding to K best measurements of the one or more measurements, where K is an integer greater than or equal to 1.

32. 30. The method of any one of claims 20 to 29, wherein the second network element is a positioning device or a chip used in the positioning device, the first network element is an access network device or a chip used in the access network device, the candidate training data for the AI ​​model includes one or more measurement results and location information of a third network element, and the one or more measurement results are obtained by the first network element by measuring a sounding reference signal sent by the third network element.

33. 30. The method of claim 20, wherein the second network element is a positioning device or a chip used in the positioning device, the first network element is a terminal device or a chip used in the terminal device, the candidate training data for the AI ​​model includes one or more measurement results and location information of the first network element, the one or more measurement results are based on measurements of a positioning reference signal sent by a third network element, and the third network element is an access network device.

34. The constraint is: a quality index threshold and a criterion for determining said quality index; a training data quantity threshold that satisfies a quality index criterion and a criterion for determining the quantity of said training data; or Maximum duration of candidate training data collection corresponding to one validation 34. The method of any one of claims 23 to 33, comprising one or more of:

35. The first information is quality indicator thresholds, quality indicator criteria, A threshold for the amount of training data that satisfies the quality metric criteria; A measure of the amount of training data that meets the quality metric criteria, or Maximum duration of candidate training data collection corresponding to one validation 35. The method of any one of claims 20 to 34, wherein the method comprises one or more of:

36. The constraint is based on an application scenario of the AI ​​model, and the application scenario of the AI ​​model is AI model-based CSI feedback or CSI prediction, AI model-based positioning, or AI model-based beam management 36. The method of any one of claims 23 to 35, comprising one or more of:

37. A communication device comprising a module configured to implement a method according to any one of claims 1 to 19 or a module configured to implement a method according to any one of claims 20 to 36.

38. A communication device, a processor coupled to a memory, the processor configured to invoke computer program instructions stored in the memory to perform the method of any one of claims 1 to 19 or to perform the method of any one of claims 20 to 36, Communication equipment.

39. 37. A communication device comprising a processor and a communication interface, wherein the communication interface is configured to receive data and / or information and to transmit the received data and / or information to the processor, wherein the processor processes the data and / or information, and wherein the communication interface is further configured to output the data and / or information processed by the processor, whereby the communication device performs a method according to any one of claims 1 to 19 or performs a method according to any one of claims 20 to 36.

40. 37. A computer-readable storage medium storing instructions that, when executed on a computer, enable the computer to perform the method of any one of claims 1 to 19 or any one of claims 20 to 36.

41. 37. A computer program product, wherein the computer-readable storage medium stores instructions that, when executed on a computer, enable the computer to perform the method of any one of claims 1 to 19 or any one of claims 20 to 36.

42. A communication system comprising a communication device according to any one of claims 37 to 39.

Images