Ai / ML model / function configuration method and apparatus

Abstract

Provided in the embodiments of the present application are an AI / ML model / function configuration method and apparatus. The method comprises: a terminal device sending first information to a network device, wherein the first information indicates an input / output configuration of an AI / ML model / function expected or supported by the terminal device; and the terminal device receiving second information sent by the network device, wherein the second information comprises an input / output configuration of an AI / ML model / function configured by the network device for the terminal device, and the input / output configuration is related to an input / output dimension and / or is related to a probability distribution.

Description

Configuration method and apparatus of AI / ML model / function TECHNICAL FIELD

[0001] Embodiments of the present application relate to the field of communication technology. BACKGROUND

[0002] At RAN1#118 meeting, companies selected the following research cases (Case 0 / 1 / 2 / 3 / 4 / 5) for Rel-19 (Release 19), and finally decided to focus on Case 2 and Case 3. For these two new research cases, the following consensus was reached for the standardization of model structure.

[0003] At RAN1#116 meeting, in order to solve the cross-vendor collaboration problem, 3GPP (3rd Generation Partnership Project) defined the following 5 options.

[0004] Generally speaking, the input / output dimension of an AI / ML model is fixed, so the AI / ML model is usually designed for specific parameter configurations (ports, feedback overhead, number of subbands / bandwidth). If an AI / ML model is designed for any possible configuration, it will bring great challenges to the training, transmission, storage, and deployment of the AI / ML model. Therefore, in the Rel-18 stage, 3GPP also studied the scalability of AI / ML models with respect to ports, feedback overhead, bandwidth / subband number.

[0005] It should be noted that the above introduction to the technical background is only for the convenience of clearly and completely describing the technical solutions of the present application, and for the convenience of understanding by those skilled in the art. The above technical solutions cannot be considered as known to those skilled in the art just because they are described in the background section of the present application.

[0006] SUMMARY

[0007] The inventors found that for Case 3, there is no specific expansion scheme for the number of input time slots. In addition, for AI / ML models / functions with scalability, the following problems still exist:

[0008] 1) In the model training phase, how does the terminal device and / or network device deliver the scalable AI / ML model and / or the corresponding training dataset;

[0009] 2) In the model inference phase, how does the network device configure the scalable AI / ML model / functionality;

[0010] 3) In the performance monitoring phase, how does the LCM (Life Cycle Management) monitor the scalable AI / ML model / functionality.

[0011] To address at least one or more of the issues, an embodiment of the present disclosure provides a method and apparatus for configuring AI / ML model / functionality.

[0012] According to an aspect of an embodiment of the present disclosure, a method for configuring AI / ML model / functionality is provided, which includes:

[0013] The terminal device sends first information to the network device, the first information indicating the input / output configuration of the desired or supported AI / ML model / functionality;

[0014] The terminal device receives second information sent by the network device, the second information including the input / output configuration of the AI / ML model / functionality configured by the network device for the terminal device;

[0015] The input / output configuration is related to the dimension of input / output and / or related to the probability distribution.

[0016] According to another aspect of an embodiment of the present disclosure, a method for configuring AI / ML model / functionality is provided, which includes:

[0017] The network device receives first information sent by the terminal device, the first information indicating the input / output configuration of the desired or supported AI / ML model / functionality of the terminal device;

[0018] The network device sends second information to the terminal device, the second information including the input / output configuration of the AI / ML model / functionality configured by the network device for the terminal device;

[0019] The input / output configuration is related to the dimension of input / output and / or related to the probability distribution.

[0020] According to still another aspect of an embodiment of the present disclosure, an apparatus for configuring AI / ML model / functionality is provided, configured in a terminal device, which includes:

[0021] a sending unit configured to send, to the network device, first information indicating an input / output configuration of an AI / ML model / function expected or supported by the terminal device;

[0022] a receiving unit configured to receive second information sent by the network device, the second information including an input / output configuration of an AI / ML model / function configured by the network device for the terminal device;

[0023] The input / output configuration is related to a dimension of input / output and / or related to a probability distribution.

[0024] According to another aspect of the embodiments of the present application, an AI / ML model / function configuration apparatus is provided, configured in a network device, and the apparatus includes:

[0025] a receiving unit configured to receive first information sent by a terminal device, the first information indicating an input / output configuration of an AI / ML model / function expected or supported by the terminal device;

[0026] a sending unit configured to send, to the terminal device, second information including an input / output configuration of an AI / ML model / function configured by the network device for the terminal device;

[0027] The input / output configuration is related to a dimension of input / output and / or related to a probability distribution.

[0028] One of the beneficial effects of the embodiments of the present application is that the terminal device and the network device achieve support for AI / ML models / functions with scalability through interaction of input / output configuration related information of AI / ML models / functions, thereby guaranteeing the performance of training, inference and monitoring of AI / ML models / functions with scalability, and improving the performance of the communication system.

[0029] Specific embodiments of the application are disclosed herein, and represented in the accompanying drawings, illustrating the principles of the application in a manner that is best suited to the understanding of its principles and the application of its principles. It should be understood that the embodiments of the application are not limited in scope to the specific embodiments disclosed herein. In the scope of the appended claims and their equivalents, the embodiments of the application include numerous changes, modifications and equivalents.

[0030] Features described and / or illustrated with respect to one implementation can be used in one or more other implementations in the same or similar manner, in combination with or in place of the features in the other implementations.

[0031] It should be emphasized that the term "comprises / comprising" when used in this text is taken to mean the presence of the stated features, integers, steps or components but not the exclusion of one or more other features, integers, steps or components. BRIEF DESCRIPTION OF DRAWINGS

[0032] Elements and features depicted in one drawing or embodiment of the application can be combined with elements and features depicted in one or more other drawings or embodiments. Also, in the drawings, like reference numerals designate corresponding parts throughout the several views, and can be used to indicate corresponding parts in more than one embodiment.

[0033] FIG. 1 is a schematic diagram of a communication system according to an embodiment of the application;

[0034] FIG. 2 is a schematic diagram of an AI / ML model corresponding to use case 3;

[0035] FIG. 3 is a schematic diagram of a configuration method of an AI / ML model / function according to an embodiment of the application;

[0036] FIG. 4 is a schematic diagram of an AI / ML model / function with scalable feedback overhead;

[0037] FIG. 5 is a schematic diagram of an example of time slot grouping;

[0038] FIG. 6 is a schematic diagram of time slot number expansion based on padding operation;

[0039] FIG. 7 is a schematic diagram of time slot expansion based on adaptive operation;

[0040] FIG. 8 is a schematic diagram of another example of time slot grouping;

[0041] FIG. 9 is another schematic diagram of a configuration method of an AI / ML model / function according to an embodiment of the application;

[0042] FIG. 10 is a schematic diagram of a configuration apparatus of an AI / ML model / function according to an embodiment of the application;

[0043] FIG. 11 is another schematic diagram of a configuration apparatus of an AI / ML model / function according to an embodiment of the application;

[0044] FIG. 12 is a schematic block diagram of a system configuration of a network device according to an embodiment of the application;

[0045] FIG. 13 is a schematic block diagram of a system configuration of a terminal device according to an embodiment of the application;

[0046] FIG. 14 is a schematic diagram of a model configuration method according to an embodiment of the application;

[0047] FIG. 15 is a schematic diagram of a model configuration method according to an embodiment of the application;

[0048] FIG. 16 is a schematic diagram of a model configuration apparatus according to an embodiment of the application;

[0049] FIG. 17 is a schematic diagram of a model configuration device according to an embodiment of the present application. DETAILED DESCRIPTION

[0050] The foregoing and other features of the present application will become apparent to those skilled in the art upon consideration of the following description of specific embodiments of the present application, taken in conjunction with the accompanying drawings. In the description of embodiments of the application, specific terminology is employed for the sake of clarity. However, the application is not intended to be limited to the specific terminology so selected. The above-mentioned and other aspects of the present application will become apparent and the application will be clearly understood from the following description, taken in conjunction with the accompanying drawings, wherein:

[0051] In the embodiments of the present application, the terms "first", "second", and the like are used to distinguish different elements from one another, but do not indicate spatial arrangement or temporal order of the elements, and the elements should not be limited by these terms. The term "and / or" includes any one and all combinations of the associated listed terms. The terms "comprise", "include", "have", and the like, mean the presence of the stated feature, element, component, or assembly, but do not exclude the presence or addition of one or more other features, elements, components, or assemblies.

[0052] In the embodiments of the present application, the singular forms "a", "an", and "the" include plural referents unless the context clearly dictates otherwise. The term "the" should be construed to mean "at least one" or "one or more" unless the context clearly indicates otherwise. In addition, the term "based on" should be interpreted as "based, at least in part, on" and the term "based upon" should be interpreted as "based, at least in part, upon" unless the context clearly indicates otherwise.

[0053] In the embodiments of the present application, the term "communication network" or "wireless communication network" can refer to a network conforming to any communication standard, such as New Radio (NR), Long Term Evolution (LTE), LTE-Advanced (LTE-A), Wideband Code Division Multiple Access (WCDMA), High-Speed Packet Access (HSPA), and the like.

[0054] Also, communication between devices in a communication system can be in accordance with communication protocols of any stage, for example, can include but is not limited to the following communication protocols: 1G (generation), 2G, 2.5G, 2.75G, 3G, 4G, 4.5G and 5G, New Radio (NR), 6G and future communication, etc., and / or other currently known or to be developed in the future communication protocols.

[0055] In embodiments of the present application, the term "network device" refers to, for example, a device that accesses a terminal device to a communication network and provides services for the terminal device in a communication system. The network device can include but is not limited to the following devices: base station (BS), access point (AP), transmission reception point (TRP), broadcast transmitter, mobile management entity (MME), gateway, server, radio network controller (RNC), base station controller (BSC), etc.

[0056] Among them, the base station can include but is not limited to: node B (NodeB or NB), evolved node B (eNodeB or eNB), 5G base station (gNB), 6G base station and future base station, etc., in addition to remote radio head (RRH), remote radio unit (RRU), relay or low power node (such as femto, pico, etc.). And the term "base station" can include some or all functions of them, and each base station can provide communication coverage for a specific geographic area. The term "cell" can refer to a base station and / or its coverage area, depending on the context in which the term is used.

[0057] In embodiments of the present application, the term "user equipment" (UE) or "terminal equipment" (TE) refers to, for example, a device that accesses a communication network through a network device and receives network services. The user equipment can be fixed or mobile, and can also be referred to as a mobile station (MS), a terminal, a user, a subscriber station (SS), an access terminal (AT), a station, a mobile terminal (MT), etc.

[0058] The terminal device can include, but is not limited to, a cellular phone, a personal digital assistant (PDA), a wireless modem, a wireless communication device, a handheld device, a machine type communication device, a laptop, a cordless phone, a smartphone, a smartwatch, a digital camera, and the like.

[0059] For another example, in an Internet of Things (IoT) scenario or the like, the user equipment can also be a machine or device for monitoring or measurement, for example, can include, but is not limited to, a machine type communication (MTC) terminal, a vehicle-mounted communication terminal, a device to device (D2D) terminal, a machine to machine (M2M) terminal, a terminal supporting sidelink communication, and the like.

[0060] In addition, the term "network side" or "network device side" refers to the side of the network, which can be a certain base station, or can include one or more network devices as described above. The term "user side" or "terminal side" or "terminal device side" refers to the side of the user or terminal, which can be a certain UE, or can include one or more terminal devices as described above. In this document, "device" can refer to a network device or a terminal device unless otherwise specified.

[0061] In the embodiments of the present application, "at least one" and "one or more" can be interchangeable, "a plurality of" and "more than one" can be interchangeable, and "a plurality of" means at least two, or two or more.

[0062] In the embodiments of the present application, pre-defined means defined by a protocol or determined according to a rule defined by a protocol, without additional configuration. Configuration / indication means direct or indirect configuration / indication by a network device through high layer signaling and / or physical layer signaling. The configuration / indication can be configured / indicated by introducing a high layer parameter in high layer signaling, where the high layer parameter refers to fields and / or information elements / units / members (IE) in high layer signaling, and the like. The physical layer signaling, for example, refers to control information (DCI) carried by a physical downlink control channel or control information carried by a sequence, but is not limited thereto.

[0063] In the embodiments of the present application, "time", "slot" and the like are an expression of a time unit, and in the following description, time or slot is taken as an example for description. The present application is not limited thereto, and according to different implementation scenarios, the time or slot in the text can also be replaced by other time units, such as a slot group, a symbol, a symbol group, and the like.

[0064] In the following description, "if" can be replaced by "when" and "when" can be replaced by "if" without causing confusion.

[0065] The scenarios of the embodiments of the present application are described below by way of examples, but the present application is not limited thereto.

[0066] FIG. 1 is a schematic diagram of a communication system according to an embodiment of the present application, which schematically illustrates a case taking a terminal device and a network device as examples. As shown in FIG. 1, the communication system 100 can include a network device 101, a terminal device 102, and a terminal device 103. For simplicity, FIG. 1 only takes two terminal devices and one network device as examples for illustration, but the embodiments of the present application are not limited thereto.

[0067] In the embodiments of the present application, the network device 101, the terminal device 102, and the terminal device 103 can perform existing services or future implementable services transmission. For example, these services can include, but are not limited to, enhanced mobile broadband (eMBB), massive machine type communication (mMTC), ultra-reliable and low-latency communication (URLLC), and related communication of reduced capability terminal devices, etc.

[0068] Among them, the terminal devices 102 and 103 can be in an RRC_IDLE state, or an RRC_INACTIVE state or an RRC_CONNECTED state, and the terminal devices 102 and 103 can also communicate with the network device 101. For example, taking the terminal device 102 as an example, the terminal device 102 can send data to the network device 101, or can perform data retransmission. The network device 101 can send a paging message to the terminal device 102, and can also send data to the terminal device 102, and the terminal device 102 receives the data sent by the network device 101. In addition, different terminal devices can also communicate with each other, for example, the terminal device 102 and the terminal device 103 can exchange data.

[0069] It is worth noting that FIG. 1 shows that the terminal device 102 and the terminal device 103 are both within the coverage of the network device 101, but the present application is not limited thereto. The terminal device 102 and the terminal device 103 can both be outside the coverage of the network device 101, or one of the terminal device 102 and the terminal device 103 is within the coverage of the network device 101 and the other is outside the coverage of the network device 101.

[0070] In embodiments of the present application, one or more AI / ML models can be configured and run in the network device and / or the terminal device. The AI / ML models can be used for various signal processing functions for wireless communication, such as CSI (Channel State Information) prediction, CSI compression, beam prediction, positioning management, etc.; the present application is not limited thereto.

[0071] FIG. 2 is a schematic diagram of an AI / ML model corresponding to use case 3. As shown in FIG. 2, the CSI information of multiple time instants / slots is jointly compressed to achieve lower compression rate or higher feedback accuracy. The CSI information of multiple time instants / slots can be obtained by prediction, and then fed into the CSI compression model. Due to the correlation between time slots, the performance of the joint CSI compression scheme (case 3) will be better than the independent (non-joint) CSI compression scheme (e.g., case 0).

[0072] However, as mentioned before, currently, there is no extension scheme for the number of input slots for use case 3. In addition, for AI / ML models / functions with scalability, there are some problems that do not have specific solutions. In view of at least one of the above problems or other similar problems, the present application is proposed.

[0073] In embodiments of the present application, CSI compression can also be referred to as “CSI encoding”, “CSI generation”, and the like. The result of the CSI compression, or CSI encoding, or CSI generation, and the like, is referred to as CSI feedback information, CSI reporting information, and the like.

[0074] In embodiments of the present application, CSI decompression can also be referred to as “CSI decoding”, “CSI reconstruction”, “CSI recovery”, “CSI reconstruction”, and the like.

[0075] In embodiments of the present application, the AI / ML model can also be referred to as an AI / ML method, an AI / ML unit, an AI / ML function, or an AI / ML element, and the like. On the UE side, the AI / ML model can also be referred to as an encoder, a CSI generation part, and the like; on the network device side, the AI / ML model can also be referred to as a decoder, a CSI reconstruction part, and the like.

[0076] In the embodiments of the present application, for the convenience of description, "training an AI / ML model or a model supporting an AI / ML function" is collectively referred to as "training an AI / ML model / function", and "deploying an AI / ML model or a model supporting an AI / ML function" is collectively referred to as "deploying an AI / ML model / function". In addition, "input dimension and / or output dimension" can also be referred to as "input / output dimension".

[0077] It should be noted that the operation processing method of the AI / ML model or function of the embodiments of the present application is applicable to use cases including but not limited to CSI compression feedback, but those skilled in the art can understand that the embodiments of the present application are also applicable to other various use cases and / or scenarios applying AI / ML model or function.

[0078] The specific embodiments of the embodiments of the present application will be exemplarily described below in conjunction with the accompanying drawings.

[0079] Embodiments of the first aspect

[0080] The embodiments of the present application provide an AI / ML model / function configuration method, which is described from the side of a terminal device.

[0081] FIG. 3 is a schematic diagram of an AI / ML model / function configuration method according to an embodiment of the present application. As shown in FIG. 3, the method comprises:

[0082] 310: The terminal device sends first information to the network device, and the first information indicates the input / output configuration of the AI / ML model / function expected or supported by the terminal device; and

[0083] 302: The terminal device receives second information sent by the network device, and the second information includes the input / output configuration of the AI / ML model / function configured by the network device for the terminal device.

[0084] In the above embodiments, the input / output configuration is related to the dimension of the input / output and / or related to the probability distribution.

[0085] The dimension of the input / output refers to the input dimension and the output dimension supported by the AI / ML model / function (model or function, referred to as model / function for short), wherein the input dimension includes but is not limited to the number of time slots of input channel information, the number of antenna ports, the number of subbands, the size of bandwidth, etc., and the output dimension includes but is not limited to the size of feedback overhead of output channel information, the number of time slots of output channel information, etc.

[0086] In the embodiments of the present application, if the AI / ML model / function supports multiple input dimensions and / or output dimensions, the AI / ML model / function has scalability of input and / or output; if the AI / ML model / function does not support multiple input dimensions and / or output dimensions or only supports one input dimension and / or output dimension, the AI / ML model / function does not have scalability of input and / or output.

[0087] In addition, the probability distribution refers to the probability distribution of the input value and / or output value of the AI / ML model / function. For the input of the AI / ML model / function, the probability distribution under different channel conditions is different. For a multiple input multiple output (MIMO) communication system, different antenna configurations can cause the probability distribution of the input of the AI / ML model to be different, for example, the mapping mode of the antenna port.

[0088] It is worth noting that the above FIG. 3 only schematically illustrates the embodiments of the present application, but the present application is not limited thereto. For example, some other operations can be added. Those skilled in the art can make appropriate modifications according to the above content, and the present application is not limited to the description of the above FIG. 3.

[0089] According to the above embodiments, the terminal device and the network device interact with the related information of the input and output configuration of the AI / ML model / function, realize the support of the AI / ML model / function with scalability, thereby guaranteeing the performance of the training, inference and monitoring of the AI / ML model / function with scalability, and improving the performance of the communication system.

[0090] In some embodiments of the embodiments of the present application, the method of the embodiments of the present application is applied to the model training stage.

[0091] In the above embodiments, the first information indicates the input and output configuration of the AI / ML model / function expected by the terminal device, and the network device configures the input and output configuration of the AI / ML model / function for the terminal device according to the input and output configuration of the AI / ML model / function expected by the terminal device reported by the terminal device, so that the terminal device can train the AI / ML model or train the function supporting the AI / ML function according to the configuration of the network device.

[0092] In the above embodiments, the first information can be sent through uplink RRC (Radio Resource Control, Radio Resource Control) signaling, or can be sent through UAI (UE Assistance Information, terminal device assistance information) signaling in the uplink RRC signaling, and the present application is not limited thereto.

[0093] In the above embodiments, in some possible implementation manners, the first information can request an AI / ML model / function supporting all or part of input dimensions and / or output dimensions, or request an AI / ML model / function supporting only one input dimension and / or output dimension. The network device can send the terminal device the second information described above, which can include, for example, an AI / ML model / function supporting all or part of input dimensions and / or output dimensions, or an AI / ML model / function supporting only one input dimension and / or output dimension, according to the content requested by the first information.

[0094] According to the above embodiments, if the AI / ML model / function provided by the network device to the terminal device can support extensibility, that is, support multiple input dimensions and / or output dimensions, the terminal device can request an AI / ML model supporting all or part of the dimensions or an AI / ML model / function not supporting extensibility (that is, supporting only one dimension) when making a request for AI / ML model transfer. Accordingly, the network device can transfer an AI / ML model supporting all or part of the dimensions to the terminal device or an AI / ML model not supporting extensibility (that is, one dimension) to the terminal device.

[0095] According to the above embodiments, the terminal device can retrain an AI / ML model or a model supporting an AI / ML function based on the AI / ML model transferred by the network device, for subsequent model inference or performance monitoring or determining other information (such as determining CQI (Channel Quality Indicator), RI (Rank Indication), etc.), or the terminal device can directly use the transferred AI / ML model for subsequent model inference or performance monitoring or determining other information.

[0096] In the above embodiments, the network device can transfer the AI / ML model by transferring the model structure and / or model parameters of the AI / ML model. In addition, the network device can transfer the AI / ML model by the second information described above or by other information. For specific content of the model structure and the model parameters, refer to related technologies, which will not be described here.

[0097] The method of the above embodiments will be described below taking the CSI compression feedback based on an AI / ML model / function as an example. In this example, the network device can provide the terminal device with an AI / ML model / function with scalable feedback overhead.

[0098] FIG. 4 is a schematic diagram of an AI / ML model / function with scalable feedback overhead. Among them, the encoder of the AI / ML model on the UE side is used for CSI compression, and the decoder of the AI / ML model on the network side is used for decompression or reconstruction or recovery of CSI.

[0099] As shown in FIG. 4, when the terminal device requests an AI / ML model that does not support scalability and requires a feedback overhead bit of 120, the network device can send the “Encoder” in FIG. 4 and the “Adaption layer 120bits” AI / ML model on the UE side to the terminal device, or the network device can also send the “Decoder” in FIG. 4 and the “Adaption layer 120bits” AI / ML model on the network side to the terminal device, or the network device can also send the “Decoder” in FIG. 4, the “Adaption layer 120bits” AI / ML model on the UE side and the network side to the terminal device.

[0100] In the above embodiment, in another possible implementation, the first information can request to train a dataset of an AI / ML model / function that supports all or part of the input dimension and / or output dimension, or request to train a dataset of an AI / ML model / function that supports only one input dimension and / or output dimension. The network device can send second information to the terminal device according to the content requested by the first information, which may, for example, include a dataset of an AI / ML model / function that supports all or part of the input dimension and / or output dimension, or include a dataset of an AI / ML model / function that supports only one input dimension and / or output dimension.

[0101] According to the above embodiment, if the AI / ML model / function provided by the network device to the terminal device can support scalability, that is, support multiple input dimensions and / or output dimensions, the terminal device can request a dataset of an AI / ML model that supports all or part of the dimensions when making a request for training data transmission, or request a dataset of an AI / ML model that does not support scalability, that is, supports only one dimension. Correspondingly, the network device can transmit a dataset of an AI / ML model that supports all or part of the dimensions to the terminal device, or transmit a dataset of an AI / ML model that does not support scalability, that is, one dimension, to the terminal device.

[0102] According to the above embodiment, the terminal device can train an AI / ML model or a model supporting an AI / ML function based on the dataset transmitted by the network device, for subsequent model inference or performance monitoring or determining other information (such as determining CQI, RI, etc.).

[0103] The method of the above embodiments is described below still taking the CSI compression feedback based on the AI / ML model / function as an example. In this example, the network device can provide the terminal device with an AI / ML model / function with scalable feedback overhead.

[0104] As shown in FIG. 4, when the terminal device requests an AI / ML model that does not support scalability and requires a feedback overhead bit of 120, the network device can send the terminal device with a data set containing the {X, Z120} sample in FIG. 4, or the network device can also send the terminal device with a data set of the {Z120, X'} sample in FIG. 4, or the network device can also send the terminal device with a data set of the {X, Z120, X'} sample in FIG. 4.

[0105] The above X refers to the input of the encoder, such as the measured CSI information, the true value CSI information, etc.; the above Z120 refers to the output of the encoder, that is, the input of the decoder, such as the PMI (Precoding Matrix Indicator), the compressed CSI information, etc.; and the above X' refers to the output of the decoder, such as the reconstructed CSI information, etc.

[0106] In some other embodiments of the present application, the method of the present application is applied to the model inference stage.

[0107] In the above embodiments, the first information indicates the input-output configuration of the AI / ML model / function supported by the terminal device, and the network device configures the input-output configuration of the AI / ML model / function for the terminal device according to the input-output configuration of the AI / ML model / function supported by the terminal device reported by the terminal device, so that the terminal device can deploy the AI / ML model / function according to the configuration of the network device.

[0108] In the above embodiments, the first information can indicate whether the terminal device supports multiple input dimensions and / or output dimensions of the AI / ML model / function, can indicate the input dimension and / or output dimension of the AI / ML model / function supported by the terminal device, can indicate the AI / ML model / function supported by the terminal device, or can indicate any combination of the above. The network device can configure the AI / ML model / function for the terminal device according to the content indicated by the first information, and send the terminal device with the second information containing the related information of the AI / ML model / function configured for the terminal device.

[0109] In some possible implementation manners, for a certain AI / ML model / function, the terminal device reports, through the first information, whether the AI / ML model / function supports scalability and / or which input / output dimensions are supported. If the terminal device reports that the AI / ML model / function does not support scalability (that is, only supports one input dimension and / or one output dimension), the network device can perform configuration according to the input dimension and / or the output dimension supported by the terminal device.

[0110] For example, for CSI compression feedback based on an AI / ML model / function, if the terminal device reports that the size (output dimension) of CSI feedback overhead supported by the terminal device is K (corresponding to K-bit feedback overhead), the network device can configure the AI / ML model / function with K-bit feedback overhead for the terminal device to perform CSI compression feedback.

[0111] For another example, for CSI compression feedback / prediction based on an AI / ML model / function, if the terminal device reports that the number of antenna ports (input dimension) supported by the terminal device is P, the network device can configure the AI / ML model / function with the number of antenna ports P for the terminal device to perform CSI compression feedback / prediction.

[0112] In some possible implementation manners, for a certain AI / ML model / function, the terminal device reports, through the first information, whether the AI / ML model / function supports scalability and / or which input / output dimensions are supported. When performing configuration of the AI / ML model / function, the network device can perform configuration within the range of the input / output dimensions supported by the terminal device.

[0113] For example, if the terminal device reports an AI / ML model / function that supports scalability, and reports that the input dimension supported by the AI / ML model / function is {X1, X2, X3, X4}, when performing configuration of the AI / ML model / function, the network device can select any value in {X1, X2, X3, X4} to perform configuration.

[0114] The above embodiments are based on CSI compression feedback of a bilateral model, but the present application is not limited thereto, and the above method of the embodiments of the present application is also applicable to CSI prediction, beam management, and the like of a unilateral model.

[0115] In the above embodiments, in some possible implementation manners, the deployed AI / ML model / function is used to process channel information of multiple time slots, and the terminal device can group channel information to be processed according to the number K of time slots supported by the deployed AI / ML model / function for model inference, and provide the network device with channel information processed by the deployed AI / ML model for each group of channel information to be processed.

[0116] In the above embodiments, the number of time slots S' of each group of channel information to be processed is equal to or less than the number of time slots K for model inference. For the group with a number of time slots less than the number of time slots K for model inference, the terminal device can perform padding and / or adaptation on the channel information of the group, so that the number of time slots of the channel information of the group is equal to the number of time slots K for model inference, and then process each group of channel information using the deployed AI / ML model / function. In addition, for the group with a number of time slots equal to the number of time slots K for model inference, the terminal device can directly process each group of channel information using the deployed AI / ML model / function.

[0117] The padding operation described above can be a zero padding operation, a copy operation, or a combination of the above two. The adaptation operation described above can be a linear transformation operation, a nonlinear transformation operation, or a combination of the two. The adaptation operation described above can be obtained based on the AI / ML model / function through training, or can be obtained by a non-AI / ML method, which is not limited in the present application.

[0118] In some possible implementations, the network device can indicate the terminal device to perform the padding operation and / or the adaptation operation through display signaling.

[0119] For example, the terminal device can receive the indication information (referred to as first indication information) sent by the network device, which indicates the terminal device to perform the padding operation and / or the adaptation operation. The implementation of the first indication information is not limited in the present application.

[0120] In the above example, the network device can also indicate the specific dimension of the padding operation, that is, the first indication information described above can also indicate the dimension of the padding operation performed by the terminal device. For example, the network device can indicate the terminal device to perform the padding operation in the third and fourth dimensions.

[0121] The above is only an example, and the present application does not limit the padding operation and / or the adaptation operation, which can also be predefined, that is, no explicit signaling is required for indication.

[0122] The channel information to be processed described above is, for example, CSI information, and the processing performed on each group of channel information to be processed is, for example, CSI information compression. That is, the terminal device can compress the CSI information of multiple time slots using the deployed AI / ML model / function and feed back to the network device. The present application is not limited thereto, and the channel information to be processed described above can also be other.

[0123] The method of the above embodiments will be described below taking the joint compression feedback of the AI / ML model / function on the CSI information of S time slots as an example.

[0124] In some possible implementations, if the number of time slots S of the compressed CSI information is larger than the number of time slots K for AI / ML model inference, the CSI information can be grouped in groups of K consecutive time slots, that is, group#1: [CSI(1), CSI(2)..., CSI(K)], group#2: [CSI(1+K), CSI(2+K)..., CSI(2K)], and so on.

[0125] FIG. 5 is a schematic diagram of an example of time slot grouping.

[0126] As shown in FIG. 5, S = 9 and K = 4, and then the CSI information of 9 time slots is allocated to three groups, wherein the number of time slots of the CSI information of the first two groups is 4, and the number of time slots of the CSI information of the last group is 1.

[0127] FIG. 6 is a schematic diagram of time slot expansion based on padding operation.

[0128] As shown in FIG. 6, for the last group, if the number of time slots S' of the CSI information is less than the number of time slots K for AI / ML model inference, padding operation can be performed on the group so that the number of time slots of the CSI information of the last group is equal to K, for example, the CSI information of the AI / ML model input is filled with zero values in the time domain dimension, or the CSI information of other time slots is copied. For the AI / ML model / function on the network side, corresponding truncation operation is also performed to obtain the CSI information of the correct number of time slots.

[0129] Taking FIG. 5 as an example, the number of time slots of the CSI information of the last group is 1, which is less than 4, and therefore padding operation (filling with 0 or copying) can be performed on the CSI information of the group, and the network side performs corresponding truncation operation when reconstructing the CSI information after receiving the feedback CSI information.

[0130] FIG. 7 is a schematic diagram of time slot expansion based on adaptive operation.

[0131] As shown in FIG. 7, for the last group, when the number of time slots S' of the CSI information is less than the number of time slots K for AI / ML model inference, the S' time slots of the CSI information can also be transformed into K time slots of the CSI information through adaptive operation. These transformations can be linear or nonlinear, can be obtained based on AI / ML model / function training, or can be preset based on non-AI / ML methods, and the like.

[0132] In some possible implementation manners, when the number of slots S of the compressed CSI information is greater than the number of slots K of AI / ML model inference, the CSI information of the S slots can be grouped, and the number of slots of each group of CSI information is as even as possible.

[0133] FIG. 8 is a schematic diagram of another example of slot grouping.

[0134] As shown in FIG. 8, S=9 and K=4, that is, the CSI information of 9 slots is evenly distributed to three groups, and the number of slots of the CSI information in each group is 3. Therefore, the CSI information of each group needs to be padded (0 or copied) or adaptively operated to form 4-slot CSI information.

[0135] The above two grouping methods are only examples, and the application is not limited thereto. Other strategies can also be used for grouping. For the group in which the number of slots of the CSI information is less than the number of slots of model inference, the padding operation and / or adaptive operation are performed, so that the number of slots of the CSI information of each group is equal to the number of slots of model inference, so that the deployed AI / ML model / function can perform compressed feedback on the CSI information of the multiple slots.

[0136] The above embodiments are applicable to CSI prediction of a single-sided model, and the application is not limited thereto. The above embodiments can also be used for other use cases of a single-sided model or other use cases of a double-sided model.

[0137] In the above embodiments, in some possible implementation manners, the deployed AI / ML model / function is used to process MIMO channel information, and the terminal device can process the MIMO channel information to be processed according to the number of transmission / reception ports Nt / Nr of model inference supported by the deployed AI / ML model / function, and then feed back to the network device.

[0138] For example, when the number of transmission / reception ports of the MIMO channel information to be processed is less than the number of transmission / reception ports Nt / Nr of model inference, the terminal device performs padding operation or adaptive operation on the MIMO channel information, so that the number of transmission / reception ports of the MIMO channel information is equal to the number of transmission / reception ports Nt / Nr of model inference. The related content of the padding operation and the adaptive operation has been described in the foregoing, and will not be described here again.

[0139] The MIMO information to be processed is, for example, a MIMO channel matrix, and the processing performed on the MIMO channel information to be processed is, for example, prediction of a future channel matrix. The application is not limited thereto, and the MIMO information can also be other.

[0140] In yet some embodiments of the embodiments of the present application, the terminal device can also perform performance monitoring on all or part of the input dimensions and / or output dimensions supported by the deployed AI / ML model / function. The input dimensions and / or output dimensions of the AI / ML model / function that need to be monitored can be configured or indicated by the network device, or can be predefined.

[0141] In the above embodiments, the terminal device reports its support for the extensibility (multiple input dimensions and / or output dimensions) of the AI / ML model / function through the first information, the network device configures the AI / ML model / function supporting the extensibility for the terminal device, and configures the terminal device through the second information. When the terminal device and / or the network device perform performance monitoring on the AI / ML model / function, performance monitoring can be performed on all or part of the supported dimensions.

[0142] In some possible implementations, the terminal device configures or indicates the dimensions that the terminal device specifically needs to perform performance monitoring. For example, in FIG. 4, the network device indicates the terminal device to only need to perform performance monitoring on the case where the feedback overhead is 120 bits.

[0143] In another possible implementation, the network device indicates the terminal device to specifically need to perform performance monitoring in a predefined manner. For example, for AI / ML-based CSI compression feedback, only the AI / ML model with 60-bit feedback overhead is used for CSI feedback, and performance monitoring is only performed on the case where the feedback overhead is 60 bits (no performance monitoring is needed on the AI / ML model with 120-bit feedback overhead).

[0144] The above embodiments are only exemplarily described, but the present application is not limited thereto, and appropriate modifications can be made on the basis of the above embodiments. For example, the above embodiments can be used alone, or one or more of the above embodiments can be combined.

[0145] According to the embodiments of the present application, the terminal device and the network device realize support for the AI / ML model / function with extensibility through the interaction of the input and output configuration information of the AI / ML model / function, thereby guaranteeing the performance of training, inference and monitoring of the AI / ML model / function with extensibility, and improving the performance of the communication system.

[0146] Embodiments of the second aspect

[0147] The embodiments of the present application provide an AI / ML model / function configuration method applied to a network device, which corresponds to the method of the embodiments of the first aspect, and the same content as the embodiments of the first aspect will not be described herein.

[0148] FIG. 9 is a schematic diagram of a method for configuring an AI / ML model / function according to an embodiment of the present application. As shown in FIG. 9, the method comprises the following steps.

[0149] 910: The network device receives first information sent by the terminal device, the first information indicating an input / output configuration of an AI / ML model / function expected or supported by the terminal device.

[0150] 920: The network device sends second information to the terminal device, the second information comprising an input / output configuration of an AI / ML model / function configured by the network device for the terminal device.

[0151] In the embodiments of the present application, the input / output configuration is related to the dimension of the input / output and / or related to the probability distribution. The related meanings of the dimension of the input / output and the probability distribution have been described in the embodiments of the first aspect, and will not be repeated here.

[0152] In some embodiments of the embodiments of the present application, the method of the embodiments of the present application is applied to a model training stage, the first information indicates an input / output configuration of an AI / ML model / function expected by the terminal device, and the network device configures an input / output configuration of an AI / ML model / function for the terminal device according to the input / output configuration of the AI / ML model / function expected by the terminal device, and sends the above-mentioned second information to the terminal device.

[0153] According to the above-mentioned embodiments, the terminal device can train an AI / ML model according to the above-mentioned second information, or train a model supporting an AI / ML function according to the above-mentioned second information.

[0154] In the above-mentioned embodiments, the first information can be sent through uplink RRC signaling, or can be sent through UAI (UE Assistance Information) signaling in uplink RRC, and the present application is not limited thereto.

[0155] In some possible implementation manners, the first information indicating an input / output configuration of an AI / ML model / function expected by the terminal device can be that the first information indicates that the terminal device requests an AI / ML model / function supporting all or part of the input dimension and / or the output dimension (i.e., an AI / ML model / function supporting extensibility), or indicates that the terminal device requests an AI / ML model / function supporting only one input dimension and / or output dimension (i.e., an AI / ML model / function not supporting extensibility).

[0156] In the above embodiments, the network device can send the second information to the terminal device according to the first information, and in this example, the second information includes an AI / ML model / function supporting all or part of the input dimensions and / or output dimensions, or an AI / ML model / function supporting only one input dimension and / or output dimension.

[0157] In the above embodiments, the network device may, for example, send the model structure and / or model parameters of the AI / ML model / function to the terminal device.

[0158] In the above embodiments, based on the AI / ML model / function delivered by the network device, the terminal device can retrain the AI / ML model / function (for example, retrain the AI / ML model or retrain the model supporting the AI / ML function), or the terminal device can also directly use the AI / ML model / function delivered by the network device.

[0159] In other possible implementations, the first information indicates the input / output configuration of the AI / ML model / function expected by the terminal device, which can be that the first information indicates that the terminal device requests a dataset for training an AI / ML model / function supporting all or part of the input dimensions and / or output dimensions (that is, an AI / ML model / function supporting extensibility), or the first information indicates that the terminal device requests a dataset for training an AI / ML model / function supporting only one input dimension and / or output dimension (that is, an AI / ML model / function not supporting extensibility).

[0160] In the above embodiments, the network device can send the second information to the terminal device according to the first information, and in this example, the second information includes a dataset for training an AI / ML model / function supporting all or part of the input dimensions and / or output dimensions, or a dataset for training an AI / ML model / function supporting only one input dimension and / or output dimension.

[0161] The above describes the behavior of the network device and the terminal device in the model training process, and specific examples and specific behavior of the network device can refer to the embodiments of the first aspect, which will not be described here.

[0162] In other embodiments of the embodiments of the present application, the method of the embodiments of the present application is applied to the model inference stage, the first information indicates the input / output configuration of the AI / ML model / function supported by the terminal device, the network device configures the input / output configuration of the AI / ML model / function for the terminal device according to the input / output configuration of the AI / ML model / function supported by the terminal device, and sends the second information to the terminal device.

[0163] According to the above embodiments, the terminal device can deploy the AI / ML model / function according to the second information to use the deployed AI / ML model / function to deliver relevant information to the network device.

[0164] In some possible implementations, the deployed AI / ML model / function is used to process channel information of multiple time slots, and the terminal device groups the channel information to be processed according to a number K of time slots supported by the deployed AI / ML model / function for model inference, and processes each group of channel information to be processed using the deployed AI / ML model and provides the network device with the processed channel information.

[0165] In the above embodiments, the channel information to be processed is, for example, CSI information, and the processing performed on each group of channel information to be processed is, for example, CSI information compression.

[0166] In the above embodiments, the number S' of time slots of each group of channel information to be processed is equal to or less than the number K of time slots for model inference, and for a group with a number of time slots less than the number K of time slots for model inference, the terminal device can perform padding and / or adaptive operation on the channel information of the group so that the number of time slots of the channel information of the group is equal to the number K of time slots for model inference, so that the deployed AI / ML model can be used to process the channel information of the multiple time slots.

[0167] In the above embodiments, the grouping manner is not limited, and can refer to the grouping manner of the embodiments of the first aspect, or other possible grouping manners can be used, so that the number S' of time slots of each group of channel information to be processed is equal to or less than the number K of time slots for model inference, so that the terminal device performs padding and / or adaptive operation on a group with a number of time slots less than the number K of time slots for model inference.

[0168] In the above embodiments, the padding operation can include, for example, zero padding and / or copy padding, and the adaptive operation can include, for example, linear transformation and / or nonlinear transformation. In addition, the adaptive operation can be obtained based on AI / ML model / function training, and / or based on non-AL / ML method.

[0169] In the above embodiments, the network device can also indicate the terminal device to perform the above padding operation and / or adaptive operation through signaling. For example, the network device can send first indication information to the terminal device to indicate the terminal device to perform padding operation and / or adaptive operation. The implementation of the first indication information is not limited in the present application.

[0170] In the above embodiments, the network device can also indicate the dimension for the terminal device to perform the padding operation. For example, the network device further indicates the dimension for the terminal device to perform the padding operation through the first indication information. The present application is not limited to this, and the network device can also indicate the dimension for the terminal device to perform the padding operation through other signaling.

[0171] In some possible implementation manners, the deployed AI / ML model / function is used to process MIMO channel information, and the terminal device performs model inference on the MIMO channel information to be processed according to the number of transmission / reception ports Nt / Nr supported by the deployed AI / ML model / function, and feeds back the processed MIMO channel information to the network device.

[0172] In the above embodiments, when the number of transmission / reception ports of the MIMO channel information to be processed is less than the number of transmission / reception ports Nt / Nr for model inference, the terminal device can perform padding operation or adaptive operation on the MIMO channel information, so that the number of transmission / reception ports of the MIMO channel information is equal to the number of transmission / reception ports Nt / Nr for model inference, so that the deployed AI / ML model / function can process the MIMO channel information.

[0173] In the above embodiments, for the padding operation and the adaptive operation, reference can be made to the embodiments of the first aspect, which will not be described herein again.

[0174] In the above embodiments, the MIMO channel information to be processed may, for example, be a MIMO channel matrix, and the processing performed on the MIMO channel information to be processed may, for example, be prediction of a future channel matrix. Reference can be made to the embodiments of the first aspect, which will not be described herein again.

[0175] In some possible implementation manners, the first information indicates the input / output configuration of the AI / ML model / function supported by the terminal device, which may, for example, indicate whether the terminal device supports multiple input dimensions and / or output dimensions of the AI / ML model / function, and / or indicate the input dimension and / or the output dimension of the AI / ML model / function supported by the terminal device, and / or indicate the AI / ML model / function supported by the terminal device.

[0176] In the above embodiments, the network device can configure the AI / ML model / function for the terminal device according to the first information, and send the second information to the terminal device. In this example, the second information may, for example, be input / output dimension information of the AI / ML model / function.

[0177] The above describes the behaviors of the network device and the terminal device in the model inference process. Specific examples and the specific behaviors of the network device can be referred to the embodiments of the first aspect, which will not be described herein again.

[0178] In some embodiments of the present application, the terminal device can perform performance monitoring on all or part of the input dimensions and / or output dimensions supported by the deployed AI / ML model / function.

[0179] In the above embodiments, the input dimensions and / or output dimensions of the AI / ML model / function that need to be monitored can be configured or indicated by the network device, or can be predefined. For details, please refer to the embodiments of the first aspect, which will not be described here.

[0180] The above embodiments are only exemplary descriptions of the embodiments of the present application, but the present application is not limited thereto, and appropriate modifications can be made on the basis of the above embodiments. For example, the above embodiments can be used alone, or one or more of the above embodiments can be combined.

[0181] According to the embodiments of the present application, the terminal device and the network device interact with the input and output configuration related information of the AI / ML model / function, realize the support of the AI / ML model / function with scalability, and thus guarantee the performance of the training, inference and monitoring of the AI / ML model / function with scalability, and improve the performance of the communication system.

[0182] Embodiments of the third aspect

[0183] The embodiments of the present application provide an AI / ML model / function configuration apparatus. The apparatus can be a terminal device, or can be a certain component or assembly configured in the terminal device, which corresponds to the method applied to the terminal device side in the embodiments of the first aspect, and the same content as the embodiments of the first aspect will not be described here.

[0184] FIG. 10 is a schematic diagram of an AI / ML model / function configuration apparatus according to an embodiment of the present application. As shown in FIG. 10, the AI / ML model / function configuration apparatus 1000 according to an embodiment of the present application includes a sending unit 1010 and a receiving unit 1020.

[0185] The sending unit 1010 is configured to send first information to the network device, the first information indicating the input and output configuration of the AI / ML model / function expected or supported by the terminal device; and the receiving unit 1020 is configured to receive second information sent by the network device, the second information including the input and output configuration of the AI / ML model / function configured by the network device for the terminal device, the input and output configuration being related to the dimensions of the input and output and / or being related to the probability distribution.

[0186] In some embodiments, as shown in FIG. 10, the apparatus 1000 further includes a processing unit 1030. In some embodiments, as shown in FIG. 10, the apparatus 1000 further includes a processing unit 1030.

[0187] In some embodiments, the first information indicates an input-output configuration of an AI / ML model / function expected by the terminal device, and the processing unit 1030 trains an AI / ML model or a model supporting an AI / ML function according to the second information.

[0188] In the above embodiments, the first information can be sent through uplink RRC signaling, or can be sent through UAI (UE Assistance Information) signaling in uplink RRC.

[0189] In some possible implementations, the first information indicates an input-output configuration of an AI / ML model / function expected by the terminal device, including:

[0190] The first information indicates that the terminal device requests an AI / ML model / function supporting all or part of input dimensions and / or output dimensions, or the first information indicates that the terminal device requests an AI / ML model / function supporting only one input dimension and / or output dimension.

[0191] The network device sends the above-mentioned second information to the terminal device according to the first information, the second information including an AI / ML model / function supporting all or part of input dimensions and / or output dimensions, or including an AI / ML model / function supporting only one input dimension and / or output dimension.

[0192] In the above embodiments, the network device can send the model structure and / or model parameters of the configured AI / ML model to the terminal device through the second information.

[0193] In the above embodiments, the processing unit 1030 can retrain an AI / ML model or a model supporting an AI / ML function according to the AI / ML model / function sent by the network device, or the processing unit 1030 can directly use the AI / ML model / function sent by the network device.

[0194] In other possible implementations, the first information indicates an input-output configuration of an AI / ML model / function expected by the terminal device, including:

[0195] The first information indicates that the terminal device requests a dataset for training an AI / ML model / function supporting all or part of input dimensions and / or output dimensions, or the first information indicates that the terminal device requests a dataset for training an AI / ML model / function supporting only one input dimension and / or output dimension.

[0196] The network device sends the second information to the terminal device according to the first information, the second information including a dataset for training an AI / ML model / function supporting all or part of input dimensions and / or output dimensions, or including a dataset for training an AI / ML model / function supporting only one input dimension and / or output dimension.

[0197] In some embodiments, the first information indicates an input-output configuration of an AI / ML model / function supported by the terminal device, and the processing unit 1030 deploys the AI / ML model / function according to the second information.

[0198] In some possible implementations, the deployed AI / ML model / function is used to process channel information of multiple time slots, and the processing unit 1030 groups the channel information to be processed according to a number K of time slots for model inference supported by the deployed AI / ML model / function, and processes each group of channel information to be processed using the deployed AI / ML model and provides the network device with the processed channel information.

[0199] The channel information to be processed is, for example, CSI information, and the processing of each group of channel information to be processed is, for example, CSI information compression.

[0200] In the above embodiments, the number S' of time slots of each group of channel information to be processed is equal to or less than the number K of time slots for model inference, and for a group with a number of time slots less than the number K of time slots for model inference, the processing unit 1030 performs padding and / or adaptive operation on the channel information of the group so that the number of time slots of the channel information of the group is equal to the number K of time slots for model inference.

[0201] The padding includes zero padding and / or copy padding, and the adaptive operation includes linear transformation and / or nonlinear transformation.

[0202] In some embodiments, the adaptive operation is obtained based on training of the AI / ML model / function, and / or based on a non-AL / ML method.

[0203] In some embodiments, the receiving unit 1020 receives first indication information sent by the network device, the first indication information indicating that the terminal device performs the padding and / or the adaptive operation.

[0204] In the above embodiments, the first indication information can also indicate a dimension of the padding performed by the terminal device.

[0205] In some possible implementation manners, the deployed AI / ML model / function is used for processing MIMO channel information, and the processing unit 1030 supports model inference of a number of sending / receiving ports Nt / Nr according to the deployed AI / ML model / function.

[0206] In the above embodiment, when the number of sending / receiving ports of the MIMO channel information to be processed is less than the number of sending / receiving ports Nt / Nr of the model inference, the processing unit 1030 performs padding operation or adaptive operation on the MIMO channel information, so that the number of sending / receiving ports of the MIMO channel information is equal to the number of sending / receiving ports Nt / Nr of the model inference.

[0207] In the above embodiment, the MIMO channel information to be processed may, for example, be a MIMO channel matrix, and the processing of the MIMO channel information to be processed may, for example, be prediction of a future channel matrix.

[0208] In some possible implementation manners, the first information indicates an input / output configuration of the supported AI / ML model / function, and includes:

[0209] The first information indicates whether the terminal device supports multiple input dimensions and / or output dimensions of the AI / ML model / function, and / or indicates the input dimensions and / or output dimensions of the AI / ML model / function supported by the terminal device, and / or indicates the AI / ML model / function supported by the terminal device.

[0210] The network device configures the AI / ML model / function for the terminal device according to the above first information, and sends the above second information to the terminal device.

[0211] In some possible implementation manners, the processing unit 1030 performs performance monitoring on all or part of the input dimensions and / or output dimensions supported by the deployed AI / ML model / function.

[0212] In the above embodiment, the input dimensions and / or output dimensions of the AI / ML model / function to be monitored may be configured or indicated by the network device, or may be predefined.

[0213] Embodiments of the present application also provide an apparatus for configuring an AI / ML model / function. The apparatus may, for example, be a network device, or may be one or more components or assemblies configured in the network device, and corresponds to the method applied to the network device side in the embodiments of the second aspect. The same content as the embodiments of the second aspect will not be described herein again.

[0214] FIG. 11 is a schematic diagram of an AI / ML model / function configuration apparatus according to an embodiment of the present application. As shown in FIG. 11, the AI / ML model / function configuration apparatus 1100 according to an embodiment of the present application includes a receiving unit 1110 and a sending unit 1120.

[0215] The receiving unit 1110 is configured to receive first information sent by a terminal device, the first information indicating an input / output configuration of an AI / ML model / function expected or supported by the terminal device; and the sending unit 1120 is configured to send second information to the terminal device, the second information including an input / output configuration of an AI / ML model / function configured by a network device for the terminal device; the input / output configuration is related to a dimension of input / output and / or related to a probability distribution.

[0216] In some embodiments, as shown in FIG. 11, the apparatus 1100 further includes a processing unit 1130.

[0217] In some embodiments, the first information indicates an input / output configuration of an AI / ML model / function expected by the terminal device, and the terminal device trains an AI / ML model or a model supporting an AI / ML function according to the second information.

[0218] In the above embodiments, the first information can be sent through uplink RRC signaling, or can be sent through UAI (UE Assistance Information) signaling in uplink RRC.

[0219] In some possible implementations, the first information indicates an input / output configuration of an AI / ML model / function expected by the terminal device, and includes:

[0220] The first information indicates that the terminal device requests an AI / ML model / function supporting all or part of input dimensions and / or output dimensions, or the first information indicates that the terminal device requests an AI / ML model / function supporting only one input dimension and / or output dimension.

[0221] In the above embodiments, the sending unit 1120 sends the above-mentioned second information to the terminal device according to the first information, the second information including an AI / ML model / function supporting all or part of input dimensions and / or output dimensions, or including an AI / ML model / function supporting only one input dimension and / or output dimension.

[0222] In the illustrated embodiments, the sending unit 1120 can send the model structure and / or model parameters of the configured AI / ML model / function to the terminal device through the above-mentioned second information.

[0223] In the above embodiments, the terminal device can retrain the AI / ML model or the model supporting the AI / ML function according to the AI / ML model / function sent by the network device, or the terminal device can directly use the AI / ML model / function sent by the network device.

[0224] In some possible implementation ways, the first information indicates the input / output configuration of the AI / ML model / function expected by the terminal device, including:

[0225] The first information indicates that the terminal device requests to train a dataset supporting the AI / ML model / function of all or part of the input dimension and / or the output dimension, or the first information indicates that the terminal device requests to train a dataset supporting the AI / ML model / function of only one input dimension and / or output dimension.

[0226] In the above embodiments, the sending unit 1120 sends, to the terminal device according to the first information, second information including a dataset for training the AI / ML model / function supporting all or part of the input dimension and / or the output dimension, or a dataset for training the AI / ML model / function supporting only one input dimension and / or output dimension.

[0227] In some embodiments, the first information indicates the input / output configuration of the AI / ML model / function supported by the terminal device, and the terminal device deploys the AI / ML model / function according to the second information.

[0228] In some possible implementation ways, the deployed AI / ML model / function is used to process channel information of multiple time slots, and the terminal device can group the channel information to be processed according to the number K of time slots for model inference supported by the deployed AI / ML model, and provide the network device with each group of channel information to be processed after processing by the deployed AI / ML model.

[0229] In the above embodiments, the channel information to be processed is, for example, CSI information, and the processing of each group of channel information to be processed is, for example, CSI information compression.

[0230] In the above embodiments, the number S' of time slots of each group of channel information to be processed is equal to or less than the number K of time slots for model inference, and for the grouping less than the number K of time slots for model inference, the terminal device can perform padding and / or adaptive operation on the channel information of the grouping, so that the number of time slots of the channel information of the grouping is equal to the number K of time slots for model inference.

[0231] The padding can include, for example, zero padding and / or copy padding, and the adaptive operation can include, for example, linear transformation and / or nonlinear transformation.

[0232] In addition, the above-mentioned adaptive operation can be obtained based on AI / ML model / function training, and / or obtained based on a non-AL / ML method.

[0233] In some embodiments, the sending unit 1120 sends first indication information to the terminal device, the first indication information indicating the terminal device to perform the above-mentioned padding operation and / or adaptive operation.

[0234] Optionally, the first indication information can also indicate the terminal device to perform the above-mentioned padding operation in the dimension.

[0235] In another possible implementation, the deployed AI / ML model / function is used to process MIMO channel information, and the terminal device can process the MIMO channel information to be processed according to the number of sending / receiving ports Nt / Nr supported by the deployed AI / ML model / function for model inference and then feed back to the network device.

[0236] In the above-mentioned embodiments, when the number of sending / receiving ports of the MIMO channel information to be processed is less than the number of sending / receiving ports Nt / Nr for model inference, the terminal device can perform padding operation or adaptive operation on the MIMO channel information to be processed, so that the number of sending / receiving ports of the MIMO channel information to be processed is equal to the number of sending / receiving ports Nt / Nr for model inference.

[0237] The above-mentioned MIMO channel information to be processed may, for example, be a MIMO channel matrix; and the above-mentioned processing on the MIMO channel information to be processed may, for example, be predicting a future channel matrix.

[0238] In yet another possible implementation, the first information indicates the input / output configuration of the AI / ML model / function supported by the terminal device, including:

[0239] The first information indicates whether the terminal device supports multiple input dimensions and / or output dimensions of the AI / ML model / function, and / or indicates the input dimension and / or output dimension of the AI / ML model / function supported by the terminal device, and / or indicates the AI / ML model / function supported by the terminal device.

[0240] In the above-mentioned embodiments, the processing unit 1130 configures the AI / ML model / function for the terminal device according to the above-mentioned first information, and the sending unit 1120 sends the above-mentioned second information to the terminal device.

[0241] In yet another embodiment, the processing unit 1130 performs performance monitoring on all or part of the input dimension and / or output dimension supported by the deployed AI / ML model / function.

[0242] In the above embodiments, the input dimension and / or output dimension of the AI / ML model / function that needs to be monitored can be configured or indicated by the network device, or can be predefined.

[0243] The above embodiments are only exemplary for the embodiments of the present application, but the present application is not limited thereto, and appropriate modifications can be made on the basis of the above embodiments. For example, the above embodiments can be used alone, or one or more of the above embodiments can be combined.

[0244] It is worth noting that the above only describes the components or modules related to the present application, but the present application is not limited thereto. The apparatuses 1000 and 1100 can also include other components or modules, and the specific content of these components or modules can be referred to the related art.

[0245] In addition, for simplicity, only the connection relationship or signal path between the components or modules is exemplarily shown in FIGS. 10 and 11, but it should be clear to those skilled in the art that various related technologies such as bus connection can be used. The above components or modules can be implemented by hardware facilities such as processors, memories, transmitters, receivers, etc.; the present application is not limited thereto.

[0246] According to the above embodiments, the terminal device and the network device realize the support for the AI / ML model / function with scalability by interacting the input and output configuration related information of the AI / ML model / function, thereby guaranteeing the performance of the training, inference and monitoring of the AI / ML model / function with scalability, and improving the performance of the communication system.

[0247] Embodiments of the fourth aspect

[0248] The embodiments of the present application provide a communication system, including a terminal device and a network device.

[0249] For example, the structure of the communication system can refer to FIG. 1. As shown in FIG. 1, the communication system 100 includes a network device 101 and terminal devices 102 and 103.

[0250] In some embodiments, the network device 101 performs the functions of the network device in the embodiments of the second aspect, and correspondingly, the terminal devices 102 and 103 perform the functions of the terminal device in the embodiments of the first aspect. Since the functions of the network device and the terminal device have been described in the embodiments of the first aspect to the second aspect, the contents are incorporated herein, and will not be repeated here.

[0251] The embodiments of the present application also provide a network device.

[0252] Figure 12 is a schematic block diagram of a system structure of a network device according to an embodiment of the present application. As shown in Figure 12, the network device 1200 can include a processor 1210 and a memory 1220, wherein the memory 1220 is coupled to the processor 1210. The memory 1220 can store various data. In addition, the memory 1220 can store a program 1230 for information processing, and execute the program 1230 under the control of the processor 1210.

[0253] In one embodiment, the network device 1200 is configured to implement the method according to the embodiments of the second aspect, for example, the functions of the apparatus 1100 in the embodiments of the third aspect, which can be integrated into the processor 1210, or configured separately from the processor 1210, for example, the apparatus 1100 can be configured as a chip connected to the processor 1210, and the functions of the apparatus are implemented through the control of the processor 1210.

[0254] In addition, as shown in Figure 12, the network device 1200 can further include a transceiver 1240, an antenna 1250, and the like. The functions of the above components are similar to those in the prior art, which will not be described here. It should be noted that the network device 1200 does not necessarily include all the components shown in Figure 12. In addition, the network device 1200 can include components not shown in Figure 12, which can be referred to the prior art.

[0255] The embodiments of the present application further provide a terminal device.

[0256] Figure 13 is a schematic block diagram of a system structure of a terminal device according to an embodiment of the present application. As shown in Figure 13, the terminal device 1300 can include a processor 1310 and a memory 1320, wherein the memory 1320 is coupled to the processor 1310. It should be noted that the figure is exemplary; other types of structures can also be used to supplement or replace the structure to implement telecommunication functions or other functions.

[0257] In one embodiment, the terminal device 1300 is configured to implement the method according to the embodiments of the first aspect, including the functions of the apparatus 1000 in the embodiments of the third aspect, which can be integrated into the processor 1310, or configured separately from the processor 1310, for example, the apparatus 1000 can be configured as a chip connected to the processor 1310, and the functions of the apparatus are implemented through the control of the processor 1310.

[0258] As shown in FIG. 13, the terminal device 1300 can further include a communication module 1330, an input unit 1340, a display 1350, and a power supply 1360. It is worth noting that the terminal device 1300 does not necessarily include all the components shown in FIG. 13; in addition, the terminal device 1300 can also include components not shown in FIG. 13, which can be referred to related technologies.

[0259] As shown in FIG. 13, the processor 1310, which is also sometimes referred to as a controller or operation control, can include a microprocessor or other processor device and / or logic device, which receives input and controls the operation of various components of the terminal device 1300.

[0260] The memory 1320, for example, can be one or more of a buffer, a flash memory, a hard drive, a removable media, a volatile memory, a non-volatile memory, or other suitable device. Various data can be stored, and programs for performing related information can also be stored. The processor 1310 can execute the programs stored in the memory 1320 to achieve information storage or processing, etc. The functions of other components are similar to the existing ones, and will not be described here. The components of the terminal device 1300 can be implemented by dedicated hardware, firmware, software, or a combination thereof, without departing from the scope of the present application.

[0261] The embodiments of the present application also provide a computer readable program, which, when executed in a network device, causes a computer to execute the method of the embodiments of the second aspect in the network device.

[0262] The embodiments of the present application also provide a storage medium storing a computer readable program, which causes a computer to execute the method of the embodiments of the second aspect in the network device.

[0263] The embodiments of the present application also provide a computer readable program, which, when executed in a terminal device, causes a computer to execute the method of the embodiments of the first aspect in the terminal device.

[0264] The embodiments of the present application also provide a storage medium storing a computer readable program, which causes a computer to execute the method of the embodiments of the first aspect in the terminal device.

[0265] The embodiments of the present application also provide a computer program product comprising a computer program, wherein the computer program is executed by a processor to implement the method of the embodiments of the first aspect or the second aspect.

[0266] Embodiments of the fifth aspect

[0267] The embodiments of the present application also provide a model configuration method, which is described from the side of the first device.

[0268] FIG. 14 is a schematic diagram of a model configuration method according to an embodiment of the present application. As shown in FIG. 14, the method comprises:

[0269] 1410: The first device receives a first data set and / or one or more indication information related to the first data set sent by the second device.

[0270] 1420: The first device performs model training on the second model part using the first data set and / or the indication information.

[0271] In the above embodiment, the first data set is used by the first device to perform model training on the second model part (UE part model). For example, the first data set can be a data set that the second device expects the first device to train the second model part. The first data set can only include a training data set, but the present application is not limited thereto, and the first data set can also include a validation data set, a test data set, etc.

[0272] In some embodiments, the indication information includes model indication information and / or data set indication information related to the first data set.

[0273] In the above embodiment, the data set indication information can be used to indicate information related to the first data set. For example, the data set indication information can include, but is not limited to, at least one of the following:

[0274] identification information of the first data set;

[0275] size information of the first data set;

[0276] sequence number information of the first data set in a first data set group;

[0277] feature information corresponding to the first data set.

[0278] In the above embodiment, the second device can transmit multiple data sets to the first device, and the multiple data sets can be distinguished by a data set group and different sequence numbers in the data set group.

[0279] In the above embodiment, the feature information corresponding to the first data set can include at least one of first feature information, second feature information, and third feature information.

[0280] The first feature information is used to indicate that the first data set is used for at least one of training, model monitoring, and performance monitoring. For example, it can be used to indicate whether the first data set is used for training, model monitoring, or performance monitoring, and in addition, it can also be used to indicate whether the first data set is used for validation, or for testing, etc.

[0281] The second feature information is used to indicate at least one of an input part of the corresponding model, an output part of the model, and a configuration related to input and output of the model. For example, it can be used to indicate whether the first data set corresponds to an input part of the model or an output part of the model, or to indicate that the first data set corresponds to a configuration related to input and output of the model.

[0282] The third feature information is used to indicate whether the first data set is used for parameter update of the model or is used for structure and parameter update of the model.

[0283] In the above embodiments, the model indication information is used to indicate information related to the model, for example, the model indication information can include at least one of first model indication information, second model indication information, and third model indication information.

[0284] The first model indication information can be used to indicate a first model part (NW part model) and / or a second model part related to the first model part. For example, the first model indication information can include or correspond to at least one of the following:

[0285] Identification information corresponding to the first model part;

[0286] Identification information corresponding to a second model part compatible with the first model part;

[0287] Identification information corresponding to a bilateral model to which the first model part and the second model part belong.

[0288] The above first model indication information can further include identification information of a model structure, identification information of a model version, identification information of a model parameter, etc.

[0289] In the above embodiments, the first model part can correspond to a network part model (NW part model) of a network device, a reconstruction model part of a bilateral model, etc. (for example, a reconstructor part, a decoder). The second model part can correspond to a counterpart part model compatible with the first model part, such as a terminal part model (UE part model), a constituent model part of a bilateral model (for example, a generator part, an encoder), etc.

[0290] The second model indication information is used to indicate at least one of a data format, data quantization, and data preprocessing related to the first data set. For example, the second model indication information can include or correspond to the above-mentioned data format information, data quantization information, data preprocessing information, etc. related to the first data set.

[0291] The third model indication information is used to indicate the performance of the second model part related to the first device. For example, the third model indication information can include or correspond to performance-related indication information of the second model part related to the first device, such as performance information that needs to be met when developing the second model part based on the first data set, such as GCS (Generalized cosine similarity), SGCS (Squared GCS), MSE (Mean square error), NMSE (Normalized mean square error), distance information, and the like.

[0292] In the above embodiments, the processing of the second device is corresponding to the first device, for example, it can send the first data set and / or one or more indication information related to the first data set to the first device.

[0293] The embodiments of the present application also provide a model configuration method, which is described from the side of the first device.

[0294] FIG. 15 is a schematic diagram of a model configuration method according to an embodiment of the present application. As shown in FIG. 15, the method includes the following steps.

[0295] 1510: The first device receives the second data set and / or one or more indication information related to the second data set sent by the second device.

[0296] 1520: The first device performs model training, performance verification and / or testing on the second model part by using the second data set and / or the indication information.

[0297] In the above embodiments, the second data set is used by the first device to perform performance verification and / or testing on the second model part. For example, the second data set can be data that the network device expects the terminal device to perform performance verification and / or testing on the second model part, and the present application is not limited thereto. The second data set can also be a data set used by the network device to train the terminal device to perform training on the second model part.

[0298] In some embodiments, the indication information includes model indication information and / or data set indication information related to the second data set.

[0299] In the above embodiments, the data set indication information can be used to indicate information related to the second data set. For example, the data set indication information can include but is not limited to at least one of the following:

[0300] Identification information of the second data set;

[0301] Size information of the second data set;

[0302] sequence number information in the second data set group;

[0303] feature information corresponding to the second data set.

[0304] In the above embodiment, the second device can transmit multiple data sets to the first device, and the multiple data sets can be distinguished by the data set group and different sequence numbers in the data set group.

[0305] In the above embodiment, the feature information corresponding to the second data set can include at least one of first feature information, second feature information, and third feature information.

[0306] The first feature information is used to indicate that the second data set is used for at least one of verification, testing, model monitoring, and performance monitoring. For example, it can be used to indicate whether the second data set is used for verification, or testing, or for model monitoring, or for performance monitoring.

[0307] The second feature information is used to indicate that the second data set is at least one of an input part of a model, an output part of a model, and a configuration related to model input and output. For example, it can be used to indicate that the second data set corresponds to an input part of a model, or an output part of a model, or to indicate that the second data set corresponds to a configuration related to model input and output.

[0308] The third feature information is used to indicate whether the second data set is used for parameter update of a model or for structure and parameter update of a model.

[0309] In the above embodiment, the model indication information is used to indicate information related to a model, for example, the model indication information can include at least one of first model indication information, second model indication information, and third model indication information.

[0310] The first model indication information can be used to indicate a first model part and / or a second model part related to the first model part. For example, the first model indication information can include or correspond to at least one of the following:

[0311] identification information corresponding to the first model part;

[0312] identification information corresponding to a second model part compatible with the first model part;

[0313] identification information corresponding to a bilateral model to which the first model part and the second model part belong.

[0314] The above first model indication information can further include identification information of a model structure, identification information of a model version, identification information of a model parameter, etc.

[0315] In the above embodiments, the first model part can correspond to a network part model of the network device, a reconstruction model part of the bilateral model, etc. (e.g., a reconstructor part, a decoder). The second model part can correspond to a contralateral part model compatible with the first model part, such as a terminal part model, a constituting model part of the bilateral model (e.g., a generator part, an encoder), etc.

[0316] The second model indication information is used to indicate at least one of a data format, data quantization, and data preprocessing related to the second data set. For example, the second model indication information can include or correspond to the above-mentioned data format information, data quantization information, data preprocessing information, etc. related to the second data set.

[0317] The third model indication information is used to indicate the corresponding performance when the first device uses the second data set to verify and / or test the model performance of the second model part. For example, the third model indication information can include or correspond to indication information related to the corresponding performance when the first device uses the second data set to verify and / or test the model performance of the second model part, for example, the performance required to be met when testing the second model part using the second data set, such as SGCS information, NMSE information, etc.

[0318] In the above embodiments, the processing of the second device corresponds to that of the first device, for example, it can send the above-mentioned second data set and / or one or more indication information related to the second data set to the first device.

[0319] In the above various embodiments, the first device can be a terminal device or a terminal side device, including but not limited to a mobile phone type terminal, a machine type terminal, or other types of terminal devices. The terminal side device corresponds to a server of a terminal device, a third party server, etc. The above-mentioned terminal device or terminal side device can have the training ability, running ability, model monitoring ability, etc. of the AI / ML model.

[0320] In the above various embodiments, the second device can be a network device or a network side device, including but not limited to a base station, or a CU (centralized unit) unit on the side of the base station, or a DU (distributed unit) unit on the side of the base station, or other independent functional units on the side of the base station. The network side device can correspond to a network element on the side of the core network. The above-mentioned various network or network side devices or functional units have the training ability, running ability, model monitoring ability of the AI / ML model.

[0321] It is worth noting that the above various embodiments only exemplarily illustrate the embodiments of the present application, but the present application is not limited thereto, and appropriate modifications can be made on the basis of the above various embodiments. For example, the above various embodiments can be used alone, or one or more of the above various embodiments can be combined.

[0322] The method of the above embodiments is described below by specific examples.

[0323] Embodiment 1

[0324] Reference is made to the model separation training process in the CSI compression feedback with AI / ML. Here, the base station device has trained the decoder on the base station side and the corresponding encoder with the first data set. The base station device transmits the first data set and the related indication information to the terminal or the server on the terminal side. The terminal side trains its encoder with this data set and completes the update of the encoder.

[0325] In order to ensure that the trained encoder has reliable performance, the base station device can also transmit a second data set to the terminal side, requiring the terminal side to perform model verification or model testing based on this data set. The model performance of the test needs to meet the performance-related indication information configured by the base station device to the terminal, such as SGCS information, NMSE information, distance information, etc.

[0326] When the performance information meets, the terminal device can report the performance meeting information to the network device.

[0327] In addition, the training and update of the encoder can be only for the update of the parameters of the model, or for the update of the whole model. These can be known through the related information indication of the network side.

[0328] At the same time, the training process of the encoder also needs to reach the consistency of the related model input and output format, type, etc. configuration information with the decoder, which can be realized through the configuration information interaction of the related data set.

[0329] Specifically, when the performance of the terminal side encoder meets the performance of the second data set, the terminal device can use the model identifier associated with the first data set or the second data set for subsequent model identification process, that is, the terminal device can use the first model indication information associated with the first data set or the second data set to report to the network device. In this way, the network device can know that the terminal device has the model that meets its performance needs and the model identifier. Or equivalently, the terminal device can use the identifier information of the first data set or the second data set for reporting, to indicate that the model based on the first data set or the second data set can meet the performance requirements.

[0330] Embodiment 2

[0331] Another cooperative training procedure for the model in the CSI compression feedback with reference to AI / ML. The terminal or terminal device receives the first data set or the second data set and the related indication information from the network device, and trains the encoder. If the performance of the training cannot meet the performance requirement configured by the network side, the terminal device sends the related performance indication information to the base station device. The base station device can further configure the data set for training the encoder.

[0332] In another specific embodiment, the terminal device or terminal side device receives the data set and model parameters or all model information from the network device, and related indication information. The terminal device or terminal side device updates the model using the model parameters, and completes further performance testing and evaluation using the data set.

[0333] In the above specific embodiments, the encoder on the terminal side is trained using the data set from the network device and the encoder is updated, or the model parameters or model information from the network device are directly received to update the encoder.

[0334] After the update is completed, performance evaluation related to the model is required. The performance evaluation indicators and the data set for performance evaluation are specified, configured or given by the network device. The terminal device needs to report the result information of the performance evaluation to the network device.

[0335] In addition, for the update of the encoder, if the update is based on the data set from the network, the updated encoder can use the model identifier of the associated decoder or encoder or the bilateral model as the new model identifier of the encoder, or use the version identifier according to the regulation or network configuration.

[0336] In addition, for the update of the encoder, if the update is based on the model, model structure or model parameter transmitted from the network, the updated encoder can use the transmitted model identifier, model structure identifier or model parameter identifier as the new model identifier of the encoder, or use the version identifier according to the regulation or network configuration.

[0337] Embodiment of the sixth aspect

[0338] The embodiments of the present application also provide a model configuration device configured in the first device. The principle of solving the problem of the device is the same as that of the embodiment shown in FIG. 14 of the embodiment of the fifth aspect, and the repeated parts are not described again.

[0339] FIG. 16 is a schematic diagram of a model configuration device according to an embodiment of the present application. As shown in FIG. 16, the device 1600 includes:

[0340] a receiving unit 1610 configured to receive a first data set and / or one or more pieces of indication information related to the first data set sent by the second device;

[0341] a processing unit 1620 configured to perform model training on the second model part by using the first data set and / or the indication information.

[0342] The related content of the first data set, the indication information, the second model part, and the first model part has been described in the embodiments of the fifth aspect, and the content is incorporated herein, and thus will not be described again.

[0343] The embodiments of the present application also provide a model configuration apparatus configured to be arranged in the first device, and the principle of solving the problem of the apparatus is the same as that of the embodiment shown in FIG. 15 of the embodiments of the fifth aspect, and thus the repeated content will not be described again.

[0344] FIG. 17 is a schematic diagram of a model configuration apparatus according to an embodiment of the present application. As shown in FIG. 17, the apparatus includes:

[0345] a receiving unit 1710 configured to receive a second data set and / or one or more pieces of indication information related to the second data set sent by the second device;

[0346] a processing unit 1720 configured to perform model training, performance verification, and / or testing on the second model part by using the second data set and / or the indication information.

[0347] The related content of the second data set, the indication information, the second model part, and the first model part has been described in the embodiments of the fifth aspect, and the content is incorporated herein, and thus will not be described again.

[0348] The embodiments of the present application also provide a terminal device including a processor and a memory storing a computer program, wherein the processor is configured to execute the computer program to implement the method described in the embodiments of the fifth aspect. For example, the terminal device can include the apparatus 1600 and / or 1700 described in the embodiments of the sixth aspect. The system configuration of the terminal device can refer to FIG. 13, and thus will not be described again.

[0349] The apparatuses and methods described above can be implemented by hardware, or by a combination of hardware and software. The present application relates to a computer readable program, which, when executed by a logic component, enables the logic component to implement the apparatuses or constituent components described above, or enables the logic component to implement the various methods or steps described above. The present application also relates to a storage medium for storing the above program, such as a hard disk, a magnetic disk, an optical disk, a DVD, a flash memory, and the like.

[0350] The method / apparatus described in conjunction with the embodiments of the present application can be directly embodied as hardware, a software module executed by a processor, or a combination thereof. For example, one or more of the functional blocks shown in the figures and / or one or more combinations of the functional blocks can correspond to individual software modules of a computer program flow, and can also correspond to individual hardware modules. These software modules can correspond to individual steps shown in the figures, respectively. These hardware modules can be implemented by, for example, fixing the software modules with a field programmable gate array (FPGA).

[0351] The software modules can be located in the RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, mobile disk, CD-ROM, or any other form of storage medium known in the art. One storage medium can be coupled to the processor, so that the processor can read information from the storage medium and write information to the storage medium; or the storage medium can be an integral part of the processor. The processor and the storage medium can be located in an ASIC. The software modules can be stored in the memory of the mobile terminal, or in a memory card that can be inserted into the mobile terminal. For example, if the device (such as a mobile terminal) uses a MEGA-SIM card or a large-capacity flash memory device, the software modules can be stored in the MEGA-SIM card or the large-capacity flash memory device.

[0352] For one or more of the functional blocks described in the figures and / or one or more combinations of the functional blocks, a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, or any appropriate combination thereof can be implemented to perform the functions described in the present application. For one or more of the functional blocks described in FIG. 9 or FIG. 10 and / or one or more combinations of the functional blocks, a combination of computing devices can also be implemented, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors in communication with a DSP, or any other such configuration.

[0353] The present application has been described above with reference to specific embodiments. However, it should be clear to those skilled in the art that these descriptions are exemplary and are not limiting to the scope of protection of the present application. Those skilled in the art can make various modifications and changes to the present application according to the spirit and principles of the present application, and these modifications and changes are also within the scope of the present application.

[0354] In addition to the above embodiments disclosed in the present embodiment, the following notes are also disclosed:

[0355] 1. An AI / ML model / function configuration apparatus configured in a first device, wherein the apparatus comprises:

[0356] a receiving unit configured to receive a first data set and / or one or more indication information related to the first data set transmitted by a second device;

[0357] a processing unit configured to perform model training on a second model part by using the first data set and / or the indication information;

[0358] wherein the first data set is used for the first device to perform model training on the second model part.

[0359] 2. The apparatus according to any one of the preceding appendices, wherein,

[0360] the indication information comprises model indication information and / or data set indication information related to the first data set;

[0361] the data set indication information is used to indicate information related to the first data set;

[0362] the model indication information comprises at least one of first model indication information, second model indication information and third model indication information;

[0363] the first model indication information is used to indicate a first model part and / or a second model part related to the first model part;

[0364] the second model indication information is used to indicate at least one of data format, data quantization and data pre-processing related to the first data set;

[0365] the third model indication information is used to indicate performance of a second model part related to the first device.

[0366] 3. The apparatus according to any one of the preceding appendices, wherein the data set indication information comprises at least one of:

[0367] identification information of the first data set;

[0368] size information of the first data set;

[0369] sequence number information of the first data set in a first data set group;

[0370] feature information corresponding to the first data set.

[0371] 4. The apparatus according to any one of the preceding appendices, wherein the feature information corresponding to the first data set comprises at least one of first feature information, second feature information and third feature information;

[0372] the first feature information is used to indicate that the first data set is used for at least one of training, model monitoring and performance monitoring.

[0373] The second feature information is used to indicate at least one of the following: an input part of a corresponding model, an output part of a model, and a configuration related to model input and output.

[0374] The third feature information is used to indicate whether the first data set is used for parameter update of a model or for structure and parameter update of a model.

[0375] 5. The apparatus according to append 2, wherein the first model indication information comprises at least one of the following:

[0376] identification information of the first model part;

[0377] identification information of the second model part compatible with the first model part;

[0378] identification information of a bilateral model to which the first model part and the second model part belong.

[0379] 6. An apparatus for configuring an AI / ML model / function, configured in a first device, wherein the apparatus comprises:

[0380] a receiving unit configured to receive a second data set and / or one or more indication information related to the second data set sent by a second device;

[0381] a processing unit configured to perform model training, performance verification and / or testing on a second model part by using the second data set and / or the indication information;

[0382] wherein the second data set is used for the first device to perform performance verification and / or testing on the second model part.

[0383] 7. The apparatus according to append 6, wherein,

[0384] the indication information comprises model indication information and / or data set indication information related to the second data set;

[0385] the data set indication information is used to indicate information related to the second data set;

[0386] the model indication information comprises at least one of the following: first model indication information, second model indication information and third model indication information;

[0387] the first model indication information is used to indicate a first model part and / or a second model part related to the first model part;

[0388] the second model indication information is used to indicate at least one of the following: data format, data quantization and data preprocessing related to the second data set;

[0389] The third model indication information is used to indicate the related performance when the first device uses the second data set to perform model performance verification and / or monitoring on the second model part.

[0390] 8. The apparatus according to any of the preceding embodiments, wherein the data set indication information comprises at least one of:

[0391] identification information of the second data set;

[0392] size information of the second data set;

[0393] sequence number information of the second data set in a second data set group;

[0394] feature information corresponding to the second data set.

[0395] 9. The method according to any of the preceding embodiments, wherein the feature information corresponding to the second data set comprises at least one of first feature information, second feature information and third feature information;

[0396] the first feature information is used to indicate that the second data set is used for at least one of verification, testing, model monitoring and performance monitoring;

[0397] the second feature information is used to indicate that the second data set is at least one of an input part of a corresponding model, an output part of a model and a configuration related to model input and output;

[0398] the third feature information is used to indicate whether the second data set is used for parameter update of a model or for structure and parameter update of a model.

[0399] 10. The apparatus according to any of the preceding embodiments, wherein the first model indication information comprises at least one of:

[0400] identification information of the first model part;

[0401] identification information of the second model part compatible with the first model part;

[0402] identification information of a bilateral model to which the first model part and the second model part belong.

Claims

1. An apparatus for configuring an AI (Artificial Intelligence) / ML (Machine Learning) model / function, configured in a terminal device, wherein, The apparatus comprises: a sending unit that sends first information to a network device, the first information indicating an input-output configuration of an AI / ML model / function expected or supported by the terminal device; a receiving unit that receives second information sent by the network device, the second information including an input-output configuration of an AI / ML model / function configured by the network device for the terminal device; wherein the input-output configuration relates to dimensions of input-output and / or relates to probability distribution.

2. The apparatus of claim 1, wherein, The first information indicates an input-output configuration of an AI / ML model / function expected by the terminal device, and the apparatus further comprises: a processing unit that trains an AI / ML model according to the second information, or trains a model supporting an AI / ML function according to the second information.

3. The apparatus of claim 2, wherein, The first information indicates an input-output configuration of an AI / ML model / function expected by the terminal device, and the apparatus further comprises: The first information indicates that the terminal device requests an AI / ML model / function supporting all or part of input dimensions and / or output dimensions, or the first information indicates that the terminal device requests an AI / ML model / function supporting only one input dimension and / or output dimension; The network device sends the second information to the terminal device according to the first information, the second information including an AI / ML model / function supporting all or part of input dimensions and / or output dimensions, or including an AI / ML model / function supporting only one input dimension and / or output dimension.

4. The apparatus of claim 3, wherein, The processing unit trains the AI / ML model, or trains a model supporting an AI / ML function, including: The processing unit re-trains the AI / ML model or the model supporting an AI / ML function according to the AI / ML model / function sent by the network device, or the processing unit directly uses the AI / ML model / function sent by the network device.

5. The apparatus of claim 2, wherein, The first information indicates an input-output configuration of an AI / ML model / function expected, including: The first information indicates that the terminal device requests a dataset for training an AI / ML model / function supporting all or part of input dimensions and / or output dimensions, or the first information indicates that the terminal device requests a dataset for training an AI / ML model / function supporting only one input dimension and / or output dimension; The network device sends the second information to the terminal device according to the first information, the second information including a dataset for training an AI / ML model / function supporting all or part of input dimensions and / or output dimensions, or including a dataset for training an AI / ML model / function supporting only one input dimension and / or output dimension. The first information indicates an input-output configuration of an AI / ML model / function supported by the terminal device, and the apparatus further comprises:

6. The apparatus of claim 1, wherein, a processing unit that deploys an AI / ML model / function according to the second information. The deployed AI / ML model / function is used to process channel information of multiple time slots, and the apparatus further comprises:

7. The apparatus of claim 6, wherein, ​ a processing unit, according to a number of time slots K for model inference supported by the deployed AI / ML model / function, grouping channel information to be processed, and processing each group of channel information to be processed using the deployed AI / ML model, and providing the network device with the processed channel information.

8. The apparatus of claim 7, wherein, the channel information to be processed is channel state information (CSI) information; the processing of each group of channel information to be processed is CSI information compression.

9. The apparatus of claim 7, wherein, a number of time slots S’ of each group of channel information to be processed is equal to or less than the number of time slots K for model inference; for a group of channel information to be processed with a number of time slots less than the number of time slots K for model inference, the processing unit performs padding operation and / or adaptive operation on the group of channel information to be processed, so that a number of time slots of the group of channel information to be processed is equal to the number of time slots K for model inference.

10. The apparatus of claim 9, wherein, the padding operation includes zero padding operation and / or copy padding operation; the adaptive operation includes linear transformation operation and / or non-linear transformation operation; the adaptive operation is based on AI / ML model / function training, and / or based on non-AL / ML method.

11. The apparatus of claim 9, wherein, the receiving unit receives first indication information sent by the network device, the first indication information indicating that the terminal device performs the padding operation and / or adaptive operation, and / or the first indication information indicating a dimension of the padding operation performed by the terminal device. the deployed AI / ML model / function is used for processing MIMO (Multiple Input Multiple Output) channel information, and the apparatus further comprises:

12. The apparatus of claim 6, wherein, a processing unit, according to a number of transmission / reception ports Nt / Nr for model inference supported by the deployed AI / ML model / function, processing MIMO channel information to be processed, and feeding back the processed MIMO channel information to the network device.

13. The apparatus of claim 12, wherein, when a number of transmission / reception ports of the MIMO channel information to be processed is less than the number of transmission / reception ports Nt / Nr for model inference, the processing unit performs padding operation or adaptive operation on the MIMO channel information to be processed, so that the number of transmission / reception ports of the MIMO channel information to be processed is equal to the number of transmission / reception ports Nt / Nr for model inference.

14. The apparatus of claim 13, wherein, the MIMO channel information to be processed is a MIMO channel matrix; the processing of the MIMO channel information to be processed is prediction of future channel matrix. the first information indicates an input / output configuration of the supported AI / ML model / function, including:

15. The apparatus of claim 6, wherein, ​ The first information indicates whether the terminal device supports multiple input dimensions and / or output dimensions of the AI / ML model / function, and / or indicates the input dimensions and / or output dimensions of the AI / ML model / function supported by the terminal device, and / or indicates the AI / ML model / function supported by the terminal device. The network device configures the AI / ML model / function for the terminal device according to the first information, and sends the second information to the terminal device.

16. The apparatus of claim 1, wherein, The apparatus further includes: a processing unit that performs performance monitoring on all or part of the input dimensions and / or output dimensions supported by the deployed AI / ML model / function.

17. The apparatus of claim 16, wherein The input dimensions and / or output dimensions of the AI / ML model / function that need to be monitored are configured or indicated by the network device, or are predefined.

18. The apparatus of claim 1, wherein The input dimensions include at least one of the following: a number of time slots of input channel information; a number of antenna ports; a number of subbands; a bandwidth size; The output dimensions include at least one of the following: a feedback overhead size of output channel information; a number of time slots of output channel information.

19. An apparatus for configuring an AI / ML model / functionality, configured in a network device, wherein, The apparatus includes: a receiving unit that receives first information sent by a terminal device, the first information indicating input / output configurations of an AI / ML model / function expected or supported by the terminal device; a sending unit that sends second information to the terminal device, the second information including input / output configurations of an AI / ML model / function configured by the network device for the terminal device; wherein the input / output configurations are related to dimensions of input / output and / or related to probability distribution.

20. A communication system including a first network device and a terminal device, wherein The first network device includes a memory and a processor, the memory storing a computer program, and the processor being configured to execute the computer program to implement the following method: receive first information sent by a terminal device, the first information indicating input / output configurations of an AI / ML model / function expected or supported by the terminal device; send second information to the terminal device, the second information including input / output configurations of an AI / ML model / function configured by the network device for the terminal device; wherein the input / output configurations are related to dimensions of input / output and / or related to probability distribution. The terminal device includes a memory and a processor, the memory storing a computer program, and the processor being configured to execute the computer program to implement the following method: send first information to a network device, the first information indicating input / output configurations of an AI / ML model / function expected or supported by the terminal device; receive second information sent by the network device, the second information including input / output configurations of an AI / ML model / function configured by the network device for the terminal device; wherein the input / output configurations are related to dimensions of input / output and / or related to probability distribution.

Images