Device capability discovery method and wireless communication device

Abstract

The invention provides a device capability discovery method and a wireless communication device. In one embodiment, the wireless communication device sends its capability messages to a source device having a pool of machine learning (ML) models. The capability message displays whether the wireless communication device is capable of executing a plurality of ML models. The wireless communication device downloads and activates one or more ML models in a subset of the ML model pool as needed. The subset of the ML model pool matches the capability message of the wireless communication device.

Description

Technical Field

[0001] The present disclosure relates to the field of communication systems, and more specifically, to a device capability discovery method and a wireless communication device. Background Art

[0002] Wireless communication systems, such as third generation (3G) mobile phone standards and technologies are well known. These 3G standards and technologies were developed by the Third Generation Partnership Project (3GPP). The third generation of wireless communication is generally developed to support macrocell mobile phone communications. Communication systems and networks have been developed in the direction of broadband and mobile systems. In a cellular wireless communication system, a user equipment (UE) is connected to a radio access network (RAN) via a radio link. The RAN includes a set of base stations (BSs) that provide radio links for UEs located in the base station coverage area and includes a core network (CN) interface that provides overall network control. As is known, the RAN and CN each perform functions related to the entire network. The Third Generation Partnership Project developed the so-called Long Term Evolution (LTE) system, i.e., the Evolved Universal Mobile Telecommunication System Territorial Radio Access Network (E-UTRAN), for mobile access networks, in which one or more macrocells are supported by base stations called eNodeBs or eNBs (evolved NodeBs). Recently, LTE is evolving further towards so-called 5G or New Radio (NR) systems, in which one or more cells are supported by a base station called a gNB. Technical issues

[0003] In 3GPP Rel-18, the "Artificial Intelligence (AI) / Machine Learning (ML) for New Radio Air Interface" Study Item (SI) is about to start development. The AI / ML is applied to 3GPP telecommunication systems and multiple use cases are being studied, including enhanced Channel State Information (CSI) feedback, beam management and positioning.

[0004] Typically, the beam selection is based on the measurement of the channel state information-reference signal (CSI-RS) / synchronization signal block (SSB). This process consumes a lot of reference signals and delays. Therefore, predictive beam switching is proposed to reduce the delay. Applying ML to beam management is under study.

[0005] Telecom devices (e.g., UE) can run multiple AI / ML models for different use cases. Multiple AI / ML models can be deployed for a single use case. How to manage multiple AI / ML models in telecom devices remains unclear.

[0006] On the other hand, generalization of AI / ML models for UE is under development. A generalized AI / ML model is referred to as an AI / ML model that is applicable to all unseen data subsets. No single generalized AI / ML model is universally qualified to fit all described scenarios or use cases of UE.

[0007] Therefore, a device capability discovery method is desired. Technical Solution

[0008] The objective of the present disclosure is to propose a wireless communication device (eg, user equipment (UE) or base station) and a device capability discovery method based on machine learning.

[0009] In a first aspect, an embodiment of the invention provides a device capability discovery method for machine learning (ML), which can be performed in a wireless communication device, including: sending a capability message of the wireless communication device to a source device having a pool of ML models, wherein the capability message indicates whether the wireless communication device is capable of executing a plurality of ML models; and One or more ML models are downloaded from a subset in the ML model pool, wherein the subset in the ML model pool matches the capability message of the wireless communication device.

[0010] In a second aspect, an embodiment of the invention provides a wireless communication device, comprising a processor configured to call and run a computer program stored in a memory, so that the device equipped with the processor performs the disclosed method.

[0011] In a third aspect, an embodiment of the invention provides a device capability discovery method for ML, which can be performed in a wireless communication device, including: Sending a capability message of the wireless communication device to a base station, wherein the wireless communication device has a ML model pool, wherein the capability message indicates whether the wireless communication device is capable of executing a plurality of ML models; receiving an indication indicating a subset of ML models, wherein the subset of ML models matches the capabilities message of the wireless communication device; and One or more ML models deployed in the wireless communication device are activated in response to the ML model subset.

[0012] In a fourth aspect, an embodiment of the invention provides a wireless communication device, which includes a processor configured to call and run a computer program stored in a memory so that a device on which the processor is installed performs the disclosed method.

[0013] The disclosed method can be implemented in a chip. The chip may include a processor configured to call and run a computer program stored in a memory so that a device equipped with the chip executes the disclosed method.

[0014] The disclosed method can be programmed as computer-executable instructions stored in a non-transitory computer-readable medium. When the non-transitory computer-readable medium is loaded into a computer, the processor of the computer is instructed to execute the disclosed method.

[0015] The non-transitory computer readable medium may include at least one of the following group: a hard disk, a CD-ROM, an optical storage device, a magnetic storage device, a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an EPROM, an electrically erasable programmable read-only memory, and a flash memory.

[0016] The disclosed method can be programmed as a computer program product, which enables a computer to perform the disclosed method.

[0017] The disclosed method can be programmed as a computer program, which causes a computer to perform the disclosed method. Beneficial effects

[0018] Multiple AI / ML models can be run in a group of communication devices, such as UE and / or gNB. Some embodiments of the present disclosure provide a mechanism to enable UE to support multiple AI / ML models. For UE, running multiple ML models is more universal and reasonable.

[0019] Some embodiments of the present disclosure provide a relationship between model complexity and UE capabilities. Using the relationship between model complexity and UE capabilities, the UE and gNB can easily support various AI / ML models with various complexity levels. A simplified rule or protocol is provided to associate model complexity with UE capabilities, rather than dealing with the hardware details of the UE or individual model details such as the number of parameters, model size, etc.

[0020] The UE capabilities become configurable for deployment of the at least one AI / ML model. The relationships / correlations between detailed UE capabilities are studied. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments or related technologies of the present disclosure, the following drawings will be described in the embodiments and briefly introduced. Obviously, the drawings are only some embodiments of the present disclosure. Those of ordinary skill in the art can obtain other drawings based on these drawings without paying the above premise.

[0022] Figure 1 A schematic diagram illustrating an example wireless communication system including a user equipment (UE), a base station, and network entities is shown.

[0023] Figure 2 A schematic diagram illustrating a system for performing a device capability discovery method is shown.

[0024] Figure 3 A schematic diagram illustrating an embodiment of the disclosed method is shown.

[0025] Figure 4 A schematic diagram illustrating another embodiment of the disclosed method is shown.

[0027] Figure 5 A schematic diagram illustrating yet another embodiment of the disclosed method is shown.

[0028] Figure 6 A schematic diagram showing the mapping between generalized AI / ML models and scenario-specific AI / ML models is shown.

[0029] Figure 7 A schematic diagram showing a system for wireless communication according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0030] The embodiments of the present disclosure and the technical problems, structural features, achieved goals and effects will be described in detail below with reference to the accompanying drawings. Specifically, the terms in the embodiments of the present disclosure are only for the purpose of describing the specific embodiments, and are not intended to limit the present disclosure.

[0031] Embodiments of the present disclosure relate to artificial intelligence (AI) and machine learning (ML) for New Radio (NR) air interface and address the issues of device capability discovery and model generalization.

[0032] The present disclosure is about the UE capabilities from the perspective of complexity. The multiple AI / ML model aspects are studied. The model generalization and complexity are discussed. Complexity level rules are proposed. In this way, the AI / ML model complexity related to the UE capabilities is clarified.

[0033] These AI / ML models may belong to the same use case or different use cases. Therefore, a related question is how to define the UE capabilities to support the multiple AI / ML models.

[0034] How to support the generalized AI / ML model needs to be studied. How to support the generalized capabilities of the generalized AI / ML model needs to be answered.

[0035] Another problem is related to model complexity. How to describe the relationship between the model complexity and UE capabilities has not yet been defined.

[0036] In another case, when the computing power, memory and storage are occupied by the deployed AI / ML models, some UE capabilities should not be fixed. They can be flexibly configured and even enabled / disabled.

[0037] When resources are limited, configure the deployed AI / ML models according to the priority order and enable / disable them.

[0038] A generalized AI / ML model is an AI / ML model that is trained to be applicable to all unseen datasets. In machine learning, generalization is a definition used to demonstrate how well a trained model can classify or predict unseen data. The training ability of a generalized AI / ML model can be referred to as generalization capability. An appropriate way to evaluate the training ability of a generalized AI / ML model is to compare the performance of the generalized AI / ML model with the performance of a scenario-specific AI / ML model. A scenario-specific is an ML model that is trained and tested using the same set of datasets without model generalization.

[0039] For simplicity, AI / ML model, AI / ML model and model are used interchangeably in the description. In the description of the embodiments of the present disclosure, model switching includes turning off or deactivating one model and turning on or activating another model. The third node may include an application server, a gNB or a UE.

[0040] refer to Figure 1 The telecommunication system including UE 10a, base station 20a, base station 20b and network entity device 30 performs the disclosed method according to the embodiments of the present disclosure. Figure 1It is shown for illustration rather than limitation, and the system may include more UE, base station (BS) and core network (CN) entities. In the figure, the connection between the device and the device components is shown as lines and arrows. The UE 10a may include a processor 11a, a memory 12a and a transceiver 13a. The base station 20a may include a processor 21a, a memory 22a and a transceiver 23a. The base station 20b may include a processor 21b, a memory 22b and a transceiver 23b. The network entity device 30 may include a processor 31, a memory 32 and a transceiver 33. Each of the processors 11a, 21a, 21b and 31 may be configured to implement the proposed functions, processes and / or methods described in this description. The layers of the radio interface protocol may be implemented in the processors 11a, 21a, 21b and 31. Each of the memories 12a, 22a, 22b and 32 operatively stores various programs and information to operate the connected processors. Each of the transceivers 13a, 23a, 23b and 33 is operatively coupled to a connected processor and transmits and / or receives a radio signal. Each of the base stations 20a and 20b may be an evolved Node B (eNB), a gNB or one of other radio nodes.

[0041] Each of the processors 11a, 21a, 21b and 31 may include a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), other chipsets, logic circuits and / or data processing devices. Each of the memories 12a, 22a, 22b and 32 may include a read-only memory (ROM), a random access memory (RAM), a flash memory, a memory card, a storage medium and / or other storage devices. Each of the transceivers 13a, 23a, 23b and 33 may include a baseband circuit and a radio frequency (RF) circuit to process RF signals. When the embodiments are implemented in software, the techniques described herein may be implemented with modules, processes, functions, entities, etc. that perform the functions described herein. The modules may be stored in a memory and executed by the processor. The memory may be implemented within a processor or outside the processor, where those may be communicatively coupled to the processor in various ways known in the art.

[0042] The network entity device 30 may be a node in CN. CN may include LTE CN or 5G core network (5GC), which may include user plane function (UPF), session management function (SMF), mobility management function (AMF), unified data management (UDM), policy control function (PCF), control plane (CP) / user plane (UP) separation (CUPS), authentication server (AUSF), network slice selection function (NSSF) and the network exposure function (NEF).

[0043] refer to Figure 2 , the system 100 for the device capability discovery method based on ML includes a data collection unit 101, a model training unit 102, an actuator 103 and a model reasoning unit 104. Please note that, Figure 2 The device capability discovery method is not necessarily limited to the present example. The device capability discovery method is applicable to any design based on ML. The general steps include data collection and / or model training and / or model reasoning and / or actuator.

[0044] The data collection unit 101 is a function of providing input data to the model training unit 102 and the model reasoning unit 104. AI / ML algorithm-specific data preparation (e.g., data preprocessing and cleaning, formatting, and conversion) is not performed in the data collection unit 101.

[0045] Examples of input data may include measurements from UEs or different network entities, feedback from actuator 103, and output from AI / ML models.

[0046] The training data is data required to be used as input of the AI / ML model training unit 102 .

[0047] The inference data is data required to be used as input of the AI / ML model inference unit 104 .

[0048] The model training unit 102 is a function for performing the ML model training, verification and testing. If necessary, the model training unit 102 is also responsible for data preparation (e.g., data preprocessing and cleaning, formatting and conversion) based on the training data delivered by the data collection unit 101.

[0049] Model deployment / update between units 102 and 104 involves deploying or updating an AI / ML model (e.g., a trained machine learning model 105a or 105b) to the model inference unit 104. The model training unit 102 trains the machine learning model 105a using a plurality of data units as training data, and generates a trained machine learning model 105b from the machine learning model 105a.

[0050] The model inference unit 104 is a function of providing an AI / ML model inference output (e.g., a prediction or decision). The AI / ML model inference output is the output of the machine learning model 105b. If necessary, the model inference unit 104 is also responsible for data preparation (e.g., data preprocessing and cleaning, formatting and conversion) based on the inference data delivered by the data collection unit 101.

[0051] The output shown in the diagram between unit 103 and unit 104 is the inference output of the AI / ML model generated by the model inference unit 104.

[0052] The actuator 103 is a function that receives the output from the model reasoning unit 104 and triggers or executes corresponding actions. The actuator 103 can trigger actions for other entities or for itself.

[0053] Feedback between unit 103 and unit 101 is information that may be needed to derive training or inference data or performance feedback.

[0054] refer to Figure 3 , Figure 4 and Figure 5 , the example of UE 10 in the description may include one of the UE 10a. The example of gNB 20 in the description may include the base station 20a or 20b. Note that even though the gNB is described as an example of a base station hereinafter, the disclosed method may be implemented in any other type of base station (e.g., an eNB or a base station for 5G or later). An uplink (UL) transmission of a control signal or data may be a transmission operation from a UE to a base station. A downlink (DL) transmission of a control signal or data may be a transmission operation from a base station to a UE. The disclosed method is described in detail as follows. The UE 10 and the base station (e.g., gNB 20) perform the device capability discovery method based on ML.

[0055] Figure 3 An embodiment of the disclosed method is shown. At least one wireless communication device performs an ML-based device capability discovery method. In one embodiment, the at least one wireless communication device may include a UE. In another embodiment, the at least one wireless communication device may include a base station. In yet another embodiment, the at least one wireless communication device may include a combination of a UE and a base station.

[0056] refer to Figure 3 , the example of UE 10 in the description may include one of the UE 10a. The example of gNB 20 in the description may include the base station 20a or 20b. Note that even though the gNB is described as an example of a base station hereinafter, the disclosed method may be implemented in any other type of base station (e.g., an eNB or a base station for 5G or later). An uplink (UL) transmission of a control signal or data may be a transmission operation from a UE to a base station. A downlink (DL) transmission of a control signal or data may be a transmission operation from a base station to a UE. The disclosed method is described in detail as follows. The UE 10 and the base station (e.g., gNB 20) perform the device capability discovery method based on ML.

[0057] Figure 3 An embodiment of the disclosed method is shown. At least one wireless communication device performs a device capability discovery method. In one embodiment, the at least one wireless communication device may include a UE. In another embodiment, the at least one wireless communication device may include a base station. In yet another embodiment, the at least one wireless communication device may include a combination of a UE and a base station.

[0058] refer to Figure 3 , the UE 10 sends a capability message of the wireless communication device to a source device having an ML model pool, wherein the capability message indicates whether the wireless communication device is capable of executing multiple ML models (S010). The wireless communication device includes a UE (e.g., UE 10). The source device includes a base station (e.g., gNB 20). The capability message may include a model type. In one embodiment, the model type may include one or more of the following: The type of ML model trained to provide CSI feedback; The type of ML model trained to provide beam predictions in the time domain; The type of ML model trained to provide beam predictions in the spatial domain; and ●The type of ML model trained to provide positioning assistance information or UE location.

[0059] In one embodiment, the model type includes one or more of the following: ● Types of generalized ML models; and ● Types of ML models for specific scenarios.

[0060] In one embodiment, the model type includes one or more of the following: ● One-sided ML model, which means the ML model works in either the UE or the gNB; and ● Bilateral ML model, which means the ML model consists of two parts. One part works on the UE side and the other Part of it works on the gNB side. The output of the UE side model is the input of the gNB side model.

[0061] In one embodiment, the capability message includes a maximum number of ML models supported by the wireless communication device. The maximum number of ML models supported by the wireless communication device is associated with the model type.

[0062] In one embodiment, the capabilities message includes a set of capabilities of the wireless communication device for determining a complexity level of the wireless communication device.

[0063] In one embodiment, the capability message includes the complexity level of the wireless communication device. The complexity level of the wireless communication device is associated with the model type. The set of capabilities of the wireless communication device includes one or more of the following: The wireless communication device's CPU, memory size, storage, floating point operations per second (FLOPS), power, buffer size, and bus bandwidth.

[0064] The UE 10 activates one or more ML models according to the UE capabilities in the capability message of the wireless communication device (S012).

[0065] refer to Figure 4 In one embodiment, the UE 10 sends a capability message 200 of the wireless communication device to a source device (e.g., gNB 20) having an ML model pool, wherein the capability message indicates whether the wireless communication device is capable of executing multiple ML models (S010a). The source device (e.g., gNB 20) receives the capability message 200 (S012a) and determines a subset of the ML model pool that matches the capability message of the wireless communication device (S013a).

[0066] The source device (e.g., gNB 20) sends an indication 201 of the subset to the UE 10 (S014a). The subset includes one or more ML models. The indication 201 may include a list of the one or more ML models in the subset.

[0067] The UE 10 downloads one or more ML models from the subset in the ML model pool, wherein the subset in the ML model pool matches the capability message of the wireless communication device ( S016a ).

[0068] refer to Figure 5 In one embodiment, the UE 10 sends a capability message 200 of the wireless communication device to a source device (e.g., a gNB 20). The UE 10 has an ML model pool. The capability message shows whether the wireless communication device is capable of executing multiple ML models (S010b). The source device (e.g., a gNB 20) receives the capability message 200 (S012b) and determines a subset of ML models that matches the capability message of the wireless communication device (S013b).

[0069] The source device (e.g., gNB 20) sends an indication 202 of the subset to the UE 10 (S014b). The subset includes one or more ML models. The indication 202 may include a list of the one or more ML models in the subset.

[0070] The UE 10 activates one or more ML models deployed in the wireless communication device in response to the ML model subset ( S016b ).

[0071] As an alternative, the ML model pool is located in a third node. The third node may be a server, or may be another gNB or UE. The UE 10 reports its ability to support multiple ML models. The gNB 20 notifies the UE and the third node of the following:

[0072] i) downloading of the ML model by the UE; and

[0073] ii) The specific model name and complexity level.

[0074] After completing the download of the ML model to the UE 10, the gNB activates the ML model. The UE capabilities to support multiple AI / ML models:

[0075] Whether the UE 10 can support multiple AI / ML models is a UE capability type of the UE 10. The UE capability may be indicated by the UE 10 through a signal. For example, the content of the signal may include {yes / no} or {0 / 1}, where "yes" or "0" indicates that the UE 10 can execute multiple AI / ML models, and "no" or "1" indicates that the UE 10 cannot execute multiple AI / ML models.

[0076] In one embodiment, an AI / ML model name or an AI / ML model type is introduced and can be carried in the UE capability message. For simplicity, the AI / ML model name and the AI / ML model type are referred to as model types. For example, the AI / ML model name or the AI / ML model type can be displayed as CSI_feedback_model, beam_prediction_time, beam_prediction_space, or positioning. CSI_feedback_model represents a model type trained to provide CSI feedback; beam_prediction_time represents a model type trained to provide beam prediction in the time domain; beam_prediction_space represents a model type trained to provide beam prediction in the spatial domain; and positioning represents a model type trained to provide positioning. ● The number of supported AI / ML models may be sent together with the model type for enhanced CSI feedback. An example of the message may be {CSI_Feedback_Model, n1}. ● The number of supported AI / ML models may be sent along with the model type for beam prediction in the time domain in a message. An example of the message may be {beam_prediction_time, n2}. ● The number of supported AI / ML models may be sent along with the model type for beam prediction in the spatial domain in a message. An example of the message may be {beam_prediction_spatial, n3}. ● The supported AI / ML models may be sent along with the model type located in the message An example of the message may be {location, n4}.

[0077] In the example, each of {n1, n2, n3, n4} is a non-negative integer indicating the supported number of models of a particular type. The message with the model types supported by the UE 10 may be classified as a UE capability message type.

[0078] In some examples, if at least one of {n1, n2, n3, n4} is left blank or configured as "0", it means that the UE or gNB does not support the corresponding type of ML model.

[0079] Alternatively, the model type can be mentioned implicitly. For example, the model type can be implicitly indicated during the model deployment phase.

[0080] For example, the UE 10 may send a list of models supported by the UE 10 to the gNB 20 in a message. For example, the message may include: List of supported ML models { {Supported_ML_Models_nr,n} …}.

[0081] The message with the list of models supported by the UE 10 may be classified as a UE capability message type.

[0082] In the example, supported_ML_models_nr represents a list of all the models supported by the UE 10, and n represents the number of AI / ML models.

[0083] Alternatively, the UE 10 is capable of providing the two UE capability messages.

[0084] In one embodiment, the gNB 20 and the UE 10 support multiple AI / ML models.

[0085] The gNB 20 may indicate the AI / ML models it supports in at least one of the following messages: ●Synchronization signal block (SSB); Master Information Block (MIB); ● System Information Block Zero (SIB0); ● System Information Block 1 (SIB1); ● Radio Resource Control (RRC) signaling; and ●Downlink Control Information (DCI).

[0086] The UE 10 may explicitly report the capabilities it supports in a UE capability message.

[0087] The UE 10 may implicitly report its supported capabilities. For example, the UE 10 may implicitly report its supported capabilities using a dedicated frequency band number or a random access channel (RACH) sequence. 5.2 Support for generalized AI / ML models and scenario-specific AI / ML models

[0088] In one embodiment, a generalized AI / ML model corresponds to multiple scenario-specific AI / ML models. In order to clearly understand the performance of the generalized AI / ML model, the UE 10 can run at least two AI / ML models simultaneously, including a generalized AI / ML model and an AI / ML model for a specific scenario.

[0089] A generalized AI / ML model can be an AI / ML model trained with dataset A and tested with dataset B, where set A and set B do not contain each other, or set A and set B are collected from different settings respectively.

[0090] The AI / ML model for a specific scenario may be a non-generalized AI / ML model. The AI / ML model is trained using dataset A and tested using dataset C. The data of dataset A and dataset C are collected from the same setting.

[0091] Whether the performance of the generalized AI / ML model meets the relevant requirements is crucial for the telecommunication equipment deployed with the generalized AI / ML model. The performance of the generalized AI / ML model can be determined based on monitoring of the generalized AI / ML model. Therefore, in some examples, whether the UE 10 can support the generalized AI / ML model is a UE capability type.

[0092] Furthermore, in an embodiment of the present disclosure, the telecommunication device performs AI / ML model management based on the relationship between the generalized AI / ML model and the AI / ML model for a specific scenario.

[0093] According to the mapping between the first generalized ML model and the ML model of the first specific scenario, the ML model of the first specific scenario is used as a backup ML model of the first generalized ML model, and the ML model of the first specific scenario is activated in response to the deactivation of the first generalized ML model.

[0094] The mapping between the first generalized ML model and the first scenario-specific ML model may be determined based on an association between a model identifier of the first generalized ML model and a model identifier of the first scenario-specific ML model.

[0095] The mapping between the first generalized ML model and the ML model for the first specific scenario may be determined based on an association between a parameter range of the first generalized ML model and a parameter range of the ML model for the first specific scenario. For example, the ML model for the first specific scenario is applicable to a first parameter range, the ML model for the second specific scenario is applicable to a second parameter range, the first generalized ML model is applicable to the first parameter range and the second parameter range, and the ML model for the first specific scenario and the ML model for the second specific scenario are used as backup ML models for the first generalized ML model.

[0096] Figure 6The mapping between the generalized AI / ML model G1 and the AI / ML models S1, S2, S3 and S4 of the specific scenarios is shown. The generalized AI / ML model G1 is associated with the models S1, S2, S3 and S4 of the four specific scenarios. The models S1, S2, S3 and S4 of the specific scenarios are applicable to different parameter ranges, such as antenna angle, antenna geometry, antenna port, beam width, signal-to-noise ratio (SNR) range or channel rank, Doppler shift, time delay. The generalized AI / ML model G1 is applicable to all the parameter ranges. When the generalized AI / ML model fails within the parameter range, model switching is triggered. In the model switching, the UE 10 deactivates the generalized AI / ML model G1 failure that fails within the parameter range, and activates the AI / ML model work of the specific scenario working within the same parameter range. The AI / ML model work of the specific scenario working within the same parameter range is the backup model of the generalized AI / ML model G1 within the parameter range.

[0097] Figure 6 The mapping between the generalized AI / ML model G2 and the AI / ML models S4, S5 and S6 for specific scenarios is shown. The generalized AI / ML model G2 is associated with four specific scenario models S4, S5 and S6. The specific scenario models S4, S5 and S6 are used for different parameter ranges, such as antenna angle, antenna geometry, antenna port, beam width, signal-to-noise ratio (SNR) range or channel rank (RANK), Doppler shift, time delay. The generalized AI / ML model G2 is applicable to all the above parameter ranges. When the generalized AI / ML model G2 fails within a certain parameter range, model switching will be triggered. In the model switching, the user equipment (UE) 10 deactivates the generalized AI / ML model that fails within the parameter range and activates the AI / ML model for the specific scenario working within the same parameter range. The AI / ML model for the specific scenario working within the same parameter range is the backup model of the generalized AI / ML model G2 within the parameter range.

[0098] The missing model is indicated as null. If the generalized AI / ML model fails within the parameter range associated with the missing model indicated as null, and there is no backup model within the parameter range. In this case, if the generalized AI / ML fails, it is necessary to fall back to non-AI methods. Mapping method:

[0099] The mapping between the generalized AI / ML model and the AI / ML model for a specific scenario can be obtained by linking the model identifier (ID) of the generalized AI / ML model and the model ID of the AI / ML model for the specific scenario.

[0100] Alternatively, the mapping is obtained based on the parameter range of the training data and / or the test data. For example, the mapping between the generalized AI / ML model and the AI / ML model of the specific scenario can be obtained by associating the parameter range of the generalized AI / ML model with the parameter range of the AI / ML model of the specific scenario. For example, when the parameter range of the generalized AI / ML model contains the parameter range of the AI / ML model of the specific scenario, or when the parameter range of the AI / ML model of the specific scenario contains the parameter range of the generalized AI / ML model, the parameter range of the generalized AI / ML model is associated with the parameter range of the AI / ML model of the specific scenario. The mapping is obtained based on the association between the parameter range of the generalized AI / ML model and the parameter range of the AI / ML model of the specific scenario. According to the mapping, the UE 10 uses the AI / ML model of the specific scenario as a backup model of the generalized AI / ML model.

[0101] The distinction between the generalized AI / ML model and the AI / ML model for specific scenarios.

[0102] In the first scheme, the marking / indication of the AI / ML model type in the UE capability message is represented by an explicit attribute in the AI / ML model. For example, the field {G / S} in the UE capability message of the UE 10 indicates whether the UE 10 can execute a generalized AI / ML model type and / or a scenario-specific AI / ML model type, where G represents a generalized AI / ML model and S represents a scenario-specific AI / ML model. In some examples, the {G / S} is explicitly bound to the model ID. For example, G1000 indicates that the ML model number "1000" is a generalized model, and "S2000" indicates that model "2000" is a non-generalized model (scenario-specific model).

[0103] In a second scheme, the marking / indication of the AI / ML model type is represented by an implicit scheme. In one embodiment, a first range of ML model identifiers is assigned to the generalized ML model type, and a second range of ML model identifiers is assigned to the ML model type for the specific scenario. For example, a first AI / ML model type (e.g., one of the generalized AI / ML model type or the AI / ML model type for the specific scenario) is represented by a model ID range, and a second AI / ML model type (e.g., the generalized AI / ML model type or the other of the AI / ML model type for the specific scenario) is represented by another model ID range. For example, model ID 00001 to model ID 10000 represent AI / ML models for a specific scenario, while model IDs greater than 10000 represent generalized AI / ML models.

[0104] In another embodiment, the UE 10 supports the two schemes. Complexity Level

[0105] The UE capabilities supporting the ML model are indicated by a combination of one or more elements in the set Comp: {Central Processing Unit (CPU), memory size, storage, floating-point operations per second (FLOPs), power, buffer size, bus bandwidth, and others}. The CPU in the set may include the model name or the capabilities of the CPU of the UE 10.

[0106] In one example, one or more elements in the set Comp are selected as UE capabilities and reported directly to the gNB (next generation Node B) 20. The gNB 20 will decide on the UE capabilities of the UE 10 to support certain AI / ML models. For example, if the processing time of a certain number of FLOPs executed by the UE 10 is within the delay requirement of the model, the model can be deployed at the UE 10.

[0107] In another example, one or more elements in a set Comp are selected and converted to a complexity level.

[0108] For example, there could be 3 levels: ●Level 1, low complexity; ● Level 2, medium complexity; and ●Level 3, high complexity.

[0109] Please note that the above complexity levels are described as examples and are not limited to three levels. Level 4 or 5, or even greater, can be defined. The complexity level can start from 0. In some examples, the complexity level is expressed as level a, level b, or level c, etc.

[0110] The complexity level of the UE 10 may be determined according to a formula or rule for determining the complexity level. In one embodiment, the complexity level of the wireless communication device may be obtained by comparing a threshold with a weighted sum of elements in the capability set of the wireless communication device.

[0111] In one embodiment, the complexity level of the wireless communication device may be obtained by comparing a threshold with the most important element in the capability set of the wireless communication device.

[0112] For example, the complexity level is determined by a weighted sum of the elements in the set {FLOPs, number of parameters, model size, buffer size, power, bus bandwidth, and others}. The formula of the weighted sum is a rule for determining the complexity level.

[0113] For example, if a1 FLOPs + number of parameters a2 + model size a3 > threshold_1, the UE 10 has complexity level 1 and is suitable for executing the model of level 1. In the above example, a1, a2 and a3 are weight coefficients. The weighted sum formula is a rule for determining the complexity level of the UE 10.

[0114] If a1 FLOPs+number of parameters a2+model size a3>threshold_2, the UE 10 has complexity level 2 and is suitable for executing the model of level 2. The weighted sum formula is a rule for determining the complexity level of the UE 10.

[0115] Alternatively, more factors or elements may be introduced to determine the threshold.

[0116] Alternatively, the most important elements among {FLOPs, number of parameters, model size, buffer size, power, bus bandwidth, and others} are selected and compared with thresholds respectively to determine the complexity level. For example, only the FLOPs are selected as the most important factor and compared with thresholds respectively to determine the complexity level.

[0117] For example, the formula or rule for determining the complexity level is explicitly notified to the UE 10 through the gNB 20 and / or the third node. For example, the formula or rule for determining the complexity level is carried in a downlink control information (DCI) signal, a radio resource control (RRC) signal, or a medium access control (MAC) control element (CE).

[0118] In another example, the formula or rule for determining the complexity level is implicitly notified to the UE 10 or the gNB 20. For example, the complexity level is given to the UE 10 during the manufacturing stage of the UE 10.

[0119] In some embodiments, the support for complexity levels is a UE capability.

[0120] In some embodiments, the model type is introduced into the UE capability message.

[0121] In the UE capability message, the number and complexity level of the AI / ML models supported by the UE 10 are marked with a model type. For example, the number and complexity level of the AI / ML models supported by the UE 10 are marked with a model type CSI_feedback_model in the UE capability message {CSI_feedback_model, n, complexity_level}. The "complexity_level" represents the complexity level, which can be low, medium or high, or level 0, 1, 2..., or level a, b, c... The model with the model type CSI_feedback_model includes an auto-encoder and an auto-decoder for channel state information (CSI) feedback.

[0122] For example, the complexity_level represents a complexity level associated with an autoencoder of the CSI_feedback_model.

[0123] In another example, the complexity_level represents a complexity level associated with an auto-decoder of the CSI_feedback_model.

[0124] In another example, the complexity_level represents both the auto-encoder and the auto-decoder of the CSI_feedback_model.

[0125] For example, the number and complexity level of AI / ML models supported by the UE 10 are tagged with the model type beam_prediction_time in the UE capability message {beam_prediction_time, n, complexity_level}.

[0126] For example, the number and complexity level of AI / ML models supported by the UE 10 are marked with the model type beam_prediction_space in the UE capability message {beam_prediction_space, n, complexity_level}.

[0127] For example, the number and complexity level of the AI / ML models supported by the UE 10 are marked with the model type location in the UE capability message {location, n, complexity_level}.

[0128] In the above examples, n represents the number of the AI / ML models.

[0129] For example, the UE 10 may send a list of models supported by the UE 10 to the gNB 20 in a message. For example, the message may include: List of supported ML models {{supported_ml_model_number,complexity_level}…}.

[0130] The complexity level is contained in the message.

[0131] The model type may be implicitly indicated, for example, during the model deployment phase. Alternatively, the UE 10 may be capable of providing the two UE capability messages.

[0132] In some embodiments, the complexity level is a property of each AI / ML model. The property accompanies the AI / ML model in a model image or a data packet of the AI / ML model.

[0133] For example, the complexity level is associated with one or more elements in the set Comp. In another example, the complexity level is associated with the computing power in the model inference 107 and may include one or more elements in the set {FLOPs, number of parameters, model size, and others}.

[0134] In one embodiment, when the complexity level of the first downloaded ML model matches the complexity level of the wireless communication device, the UE 10 reports that the deployment of the first downloaded ML model among the downloaded one or more ML models is successful.

[0135] When the complexity level of the first downloaded ML model matches the complexity level of the wireless communication device, the UE 10 reports a deployment failure of the first downloaded ML model among the downloaded one or more ML models.

[0136] During model deployment, the UE 10 determines whether the complexity level of the model is the same as the complexity level of the UE 10 in the configuration of the UE 10. If the complexity level of the model is the same as the complexity level of the UE 10 in the configuration of the UE 10, the UE 10 will report that the deployment is successful. Otherwise, the UE 10 will report that the deployment failed due to an unsupported model complexity level. Alternatively, as a configuration, the alignment of the model complexity level of the monitoring model with the model complexity level of the monitored model is one of the factors determining whether the deployment is successful.

[0137] exist Figure 5 In the above example, when the complexity level of the first activated ML model matches the complexity level of the wireless communication device, the UE 10 reports that the deployment of the first activated ML model among the activated one or more ML models is successful.

[0138] Furthermore, when the complexity level of the first activated ML model matches the complexity level of the wireless communication device, the UE 10 reports a deployment failure of the first activated ML model among the activated one or more ML models.

[0139] In some embodiments, if there is no complexity level configuration available for the UE 10, the UE 10 may determine whether the UE 10 can support the complexity level of the deployed AI / ML model. When the UE 10 can support the complexity level of the deployed AI / ML model, the UE 10 reports deployment success (i.e., the deployment success). Otherwise, the UE 10 reports deployment failure due to unsupported complexity (i.e., the deployment failure). Relevance of UE capabilities.

[0140] A set of UE capabilities may be associated with another set of UE capabilities. In one embodiment, two sets of UE capabilities are given as examples and are described in detail below. One of the two sets of UE capabilities is referred to as a first set of UE capabilities, and the other set is referred to as a second set of UE capabilities. In one embodiment, in response to enabling the second set of UE capabilities of the wireless communication device, the first set of UE capabilities of the wireless communication device is disabled. In another embodiment, in response to enabling the second set of UE capabilities of the wireless communication device, the first set of UE capabilities of the wireless communication device is enabled. Each of the first set of UE capabilities and the second set of UE capabilities of the wireless communication device includes one or more of the following contents associated with one of the downloaded one or more ML models:

[0141] Training, inference, monitoring, online training, data collection, post-processing, fine-tuning, and pre-processing. In response to enabling the second set of UE capabilities, disabling the first set of UE capabilities.

[0142] For example, due to limited resources on the UE 10, if the resources on the UE 10 have been configured for training of model A, the UE 10 does not have sufficient resources for post-processing of model B.

[0143] For example, if a model executed by the UE 10 is in inference mode, the model cannot be configured for online training on the UE 10 .

[0144] For example, the deployment or activation of the first model may take up a certain amount of resources, which may prevent the second model from being post-processed (fine-tuned) and / or trained online.

[0145] In another example, although the UE 10 has the UE capability to support two AI / ML models, the deployment of a model with a higher complexity level may prevent the deployment of another model with a lower complexity level.

[0146] For example, the first set of UE capabilities may include at least one of {training, inference, monitoring, online training, data collection, post-processing, fine-tuning, pre-processing}. For example, the second set of UE capabilities may include at least one of {training, inference, monitoring, online training, data collection, post-processing, fine-tuning, pre-processing}.

[0147] Alternatively, in response to configuration / enabling related to the second set of UE capabilities of model B, the first set of UE capabilities of model A is disabled.

[0148] In one embodiment, the UE 10 has two groups of UE capabilities, including a second group of UE capabilities and a third group of UE capabilities. In response to the enabling of the second set of UE capabilities, a third set of UE capabilities is enabled.

[0149] In some embodiments, the fine-tuning of the model in the UE 10 is enabled together with model training. If on-device training is enabled for the model, the fine-tuning is a lightweight operation compared to training and can be enabled together with training.

[0150] For example, the first set of UE capabilities may include at least one of {training, inference, monitoring, online training, data collection, post-processing, fine-tuning, pre-processing}. For example, the third set of UE capabilities may include at least one of {training, inference, monitoring, online training, data collection, post-processing, fine-tuning, pre-processing}.

[0151] Alternatively, in response to enabling the third set of UE capabilities of model B, the first set of UE capabilities of model A is enabled.

[0152] Alternatively, the UE 10 has three groups of UE capabilities, including a first group of UE capabilities, a second group of UE capabilities and a third group of UE capabilities.

[0153] In response to enabling the second set of UE capabilities, disabling the first set of UE capabilities.

[0154] In response to enabling the second set of UE capabilities, a third set of UE capabilities is enabled.

[0155] For example, the first group of UE capabilities may include at least one of {training, inference, monitoring, online training, data collection, post-processing, fine-tuning, pre-processing}. The second group of UE capabilities may include at least one of {training, inference, monitoring, online training, data collection, post-processing, fine-tuning, pre-processing}. The third group of UE capabilities may include at least one of {training, inference, monitoring, online training, data collection, post-processing, fine-tuning, pre-processing}.

[0156] Alternatively, in response to enabling the second set of UE capabilities of model B, disabling the first set of UE capabilities of model A. In response to enabling the second set of UE capabilities of model B, enabling the third set of UE capabilities of model A. UE status report:

[0157] When UE 10 moves from an original cell served by an original gNB (e.g., one of the base stations 20a or 20b) to a new cell served by a new gNB (e.g., the other of the base stations 20a or 20b). The new gNB needs to know the status of the UE 10. The wireless communication device includes a user equipment (UE). The source device includes a base station. The UE 10 performs a handover operation from the original base station to the new base station, and the UE reports the capability message to the new base station.

[0158] The capability message includes the status of the UE. The status of the UE shows which ML model has been deployed or activated by the UE, and how many ML models of a specific complexity level can be deployed to the UE. The capability message can be sent to the new base station via a physical uplink control channel (PUCCH) and / or a physical uplink shared channel (PUSCH). For example, if the UE 10 supports the deployment of two AI / ML models and one model is currently running / activated, the UE 10 only has limited resources for the other AI / ML model.

[0159] The UE 10 should report these conditions to the new gNB 20, and the new gNB can deploy and activate the appropriate AI / ML model accordingly. For example, the new gNB can deactivate the currently deployed AI / ML model and deploy a new model with high complexity. Please note that the "low" and "high" here correspond to different complexities (or complexity levels), respectively.

[0160] Therefore, the UE 10 reports its status in the UE capabilities.

[0161] For example, UE capabilities may include: {Supported ML models, {{2, low}, {1, high}}}

[0162] Wherein 2 and 1 are the number of the models supported by the UE 10. In the above example, 2 represents two low-complexity models, or 1 represents one high-complexity model.

[0163] The UE 10 shall report this condition to the new gNB 20 in a UE capability message. For example, the capability message may be in the following format:

[0164] {Currently deployed model / Currently activated model {1, low}}.

[0165] In the above example, {1, low} indicates that the UE 10 has a currently deployed model of low complexity level.

[0166] Alternatively, the original gNB sends this condition to the new gNB in ​​a UE capability message via the X2 / Xn interface. For example, the UE capability message may be in the following format: {Supported ML models, {{2, low}, {1, high}}}.

[0167] For example, the UE capability message may adopt the following format:

[0168] {Currently deployed model / Currently activated model {1, low}}.

[0169] Please note that in the above embodiment, the status of the UE 10 is reported to the new gNB in ​​a UE capability message.

[0170] Alternatively, the UE 10 reports its status in a UE capability message and sends the UE capability message to the new gNB via a physical uplink control channel (PUCCH) and / or a physical uplink shared channel (PUSCH).

[0171] Alternatively, the UE 10 replies its status in response to a request from the new gNB and / or the original gNB.

[0172] Figure 7 7 is a block diagram of an example system 700 for wireless communications according to an embodiment of the present disclosure. The embodiments described herein may be implemented into the system using any suitably configured hardware and / or software. Figure 7 The system 700 is shown, including a Radio Frequency (RF) circuit 710, a baseband circuit 720, a processing unit 730, a memory / storage device 740, a display 750, a camera 760, a sensor 770, and an input / output (I / O) interface 780, which are coupled to each other as shown.

[0173] The processing unit 730 may include circuits, such as but not limited to one or more single-core or multi-core processors. The processor may include any combination of general-purpose processors and special-purpose processors, such as a graphics processor and an application processor. The processor may be coupled to the memory / storage device and configured to execute instructions stored in the memory / storage device to enable various applications and / or operating systems running on the system.

[0174] The radio control functions may include, but are not limited to, signal modulation, encoding, decoding, radio frequency offset, etc. In some embodiments, the baseband circuit may provide communications compatible with one or more radio technologies. For example, in some embodiments, the baseband circuit may support communications with 5G New Radio (NR), Long Term Evolution (LTE), Evolved Universal Terrestrial Radio Access Network (EUTRAN) and / or other wireless metropolitan area networks (WMAN), wireless local area networks (WLAN), and wireless personal area networks (WPAN). An embodiment in which the baseband circuit is configured to support radio communications of more than one wireless protocol may be referred to as a multi-mode baseband circuit. In various embodiments, the baseband circuit 720 may include circuits for operating signals that are not strictly considered to be in the baseband frequency. For example, in some embodiments, the baseband circuit may include circuits for operating signals having an intermediate frequency, which is between the baseband frequency and the radio frequency.

[0175] In various embodiments, the system 700 may be a mobile computing device, such as, but not limited to, a laptop computing device, a tablet computing device, a netbook, an ultrabook, a smart phone, etc. In various embodiments, the system may have more or fewer components and / or a different architecture. Where appropriate, the methods described herein may be implemented as a computer program. The computer program may be stored on a storage medium, such as a non-transitory storage medium.

[0176] The described embodiments of the present disclosure are a combination of techniques / processes that may be adopted in 3GPP specifications to create a final product.

[0177] If the software functional unit is implemented, used and sold as a product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution proposed in the present disclosure can be basically or partially implemented in the form of a software product. Alternatively, a part of the technical solution that is beneficial to the prior art can be implemented in the form of the software product. The software product in the computer is stored in a storage medium, including a plurality of commands for a computing device (such as a personal computer, a server or a network device) to run all or some of the steps disclosed in the embodiment of the present disclosure. The storage medium includes a USB disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a floppy disk or other types of media capable of storing program code.

[0178] The present disclosure provides a device capability discovery method based on machine learning. The invention provides an embodiment for solving problems in ML model deployment.

[0179] Multiple AI / ML models can be run in a group of telecommunication devices (e.g., UE and / or gNB). Some embodiments of the present disclosure provide a mechanism to enable UE to support multiple AI / ML models. For UE, running multiple ML models is more general and reasonable.

[0180] Some embodiments of the present disclosure provide a relationship between model complexity and UE capabilities. Using the relationship between model complexity and UE capabilities, the UE and gNB can easily support different AI / ML models with various complexity levels. Instead of dealing with the hardware details of the UE or individual model details such as the number of parameters, model size, etc., a simplified rule or protocol is provided to associate model complexity with UE capabilities.

[0181] The UE capabilities become configurable for deployment of the at least one AI / ML model. The relationships / correlations between detailed UE capabilities are studied.

[0182] While the present disclosure has been described in connection with what is considered to be the most practical and preferred embodiments, it is to be understood that the disclosure is not limited to the disclosed embodiments but is intended to cover various arrangements embodied without departing from the scope of the broadest interpretation of the appended claims.

Claims

1. A device capability discovery method for machine learning (ML), which can be performed in a wireless communication device, comprising: sending a capability message of the wireless communication device to a source device having a pool of machine learning (ML) models, wherein the capability message indicates whether the wireless communication device is capable of executing a plurality of ML models; and One or more ML models are downloaded and activated from a subset in the ML model pool, wherein the subset in the ML model pool matches the capability message of the wireless communication device. The method of claim 1 , wherein the capability message includes a model type.

3. The method according to claim 2, wherein the model type comprises one or more of the following: The type of ML model trained to provide channel state information (CSI) feedback; The type of ML model trained to provide beam predictions in the time domain; The type of ML model trained to provide beam predictions in the spatial domain; and The type of ML model trained to provide localization.

4. The method according to claim 2, wherein the model type comprises one or more of the following: Generalizing ML model types; and ML model types for specific scenarios. 5 . The method of claim 4 , wherein a first range of ML model identifiers is assigned to the generalized ML model type, and a second range of ML model identifiers is assigned to the scenario-specific ML model type.

6. The method according to claim 4, wherein according to a mapping between a first generalized ML model and an ML model for a first specific scenario, the ML model for the first specific scenario is used as a backup ML model for the first generalized ML model, and the ML model for the first specific scenario is activated in response to deactivation of the first generalized ML model.

7. The method according to claim 6, wherein a mapping between the first generalized ML model and the first scenario-specific ML model is determined based on an association between a model identifier of the first generalized ML model and a model identifier of the first scenario-specific ML model.

8. The method of claim 6, wherein a mapping between the first generalized ML model and the ML model for the first specific scenario is determined based on an association between a parameter range of the first generalized ML model and a parameter range of the ML model for the first specific scenario.

9. The method according to claim 8, wherein the ML model for the first specific scenario is used for a first parameter range, the ML model for the second specific scenario is used for a second parameter range, the first generalized ML model is used for the first parameter range and the second parameter range, and the ML model for the first specific scenario and the ML model for the second specific scenario are used as backup ML models of the first generalized ML model.

10. The method of claim 2, wherein the capability message includes a maximum number of ML models supported by the wireless communication device.

11. The method of claim 10, wherein the maximum number of ML models supported by the wireless communication device is associated with the model type.

12. The method of claim 1, wherein the capability message comprises a set of capabilities of the wireless communication device for determining a complexity level of the wireless communication device; or The capabilities message includes the complexity level of the wireless communication device.

13. The method of claim 12, wherein the complexity level of the wireless communication device is associated with the model type.

14. The method according to claim 12, further comprising: When the complexity level of the first downloaded ML model matches the complexity level of the wireless communication device, reporting that the deployment of the first downloaded ML model among the downloaded one or more ML models is successful; as well as When the complexity level of the first downloaded ML model matches the complexity level of the wireless communication device, reporting a deployment failure of the first downloaded ML model among the downloaded one or more ML models.

15. The method of claim 12, wherein the capability set of the wireless communication device comprises one or more of the following: The central processing unit (CPU), memory size, storage, floating point operations per second (FLOPs), power, buffer size, and bus bandwidth of the wireless communication device.

16. The method of claim 12, wherein the complexity level of the wireless communication device is obtained by comparing a threshold to a weighted sum of elements in the capability set of the wireless communication device.

17. The method of claim 12, wherein the complexity level of the wireless communication device is obtained by comparing a threshold with a most significant element of the capability set of the wireless communication device.

18. The method of claim 1 , wherein in response to enabling a second set of UE capabilities of the wireless communication device, disabling a first set of UE capabilities of the wireless communication device; or In response to enabling the second set of UE capabilities of the wireless communication device, enabling a first set of UE capabilities of the wireless communication device.

19. The method of claim 18, wherein each of the first set of UE capabilities and the second set of UE capabilities of the wireless communication device includes one or more of the following associated with one of the downloaded one or more ML models: Training, inference, monitoring, online training, data collection, post-processing, fine-tuning, and pre-processing.

20. The method of claim 1, wherein the wireless communication device comprises a user equipment (UE).

21. The method of claim 18, wherein the source device comprises a base station.

22. The method according to claim 21, wherein the UE performs a handover operation from an original base station to a new base station, and the UE reports the capability message to the new base station.

23. The method of claim 22, wherein the capability message includes a status of the UE, the status of the UE indicating which ML model has been deployed or activated by the UE and how many ML models of a specific complexity level can be deployed to the UE.

24. The method according to claim 23, wherein the capability message is sent to the new base station through a physical uplink control channel (PUCCH) and / or a physical uplink shared channel (PUSCH).

25. A wireless communication device, comprising: A processor configured to call and run a computer program stored in a memory so that a device equipped with the processor executes the method according to any one of claims 1 to 24.

26. A chip, comprising: A processor configured to call and run a computer program stored in a memory so that a device equipped with the chip executes the method according to any one of claims 1 to 24.

27. A computer-readable storage medium having a computer program stored therein, wherein the computer program causes a computer to execute the method according to any one of claims 1 to 24.

28. A computer program product comprising a computer program, wherein the computer program causes a computer to execute the method of any one of claims 1 to 24.

29. A computer program, wherein the computer program causes a computer to execute the method of any one of claims 1 to 24.

30. A device capability discovery method for machine learning (ML), executable in a wireless communication device, comprising: Sending a capability message of the wireless communication device to a source device, the wireless communication device having a machine learning (ML) model pool, wherein the capability message indicates whether the wireless communication device is capable of executing a plurality of ML models; receiving an indication indicating a subset of ML models, wherein the subset of ML models matches the capabilities message of the wireless communication device; as well as In response to the ML model subset, one or more ML models deployed in the wireless communication device are activated.

31. The method of claim 30, wherein the capability message includes a model type.

32. The method of claim 31 , wherein the model type comprises one or more of the following: The type of ML model trained to provide CSI feedback; The type of ML model trained to provide beam predictions in the time domain; The type of ML model trained to provide beam predictions in the spatial domain; and The type of ML model trained to provide localization.

33. The method of claim 31 , wherein the model type comprises one or more of the following: Generalizing ML model types; and ML model types for specific scenarios.

34. The method of claim 33, wherein a first range of ML model identifiers is assigned to the generalized ML model type, and a second range of ML model identifiers is assigned to the scenario-specific ML model type.

35. The method of claim 33, wherein, according to a mapping between a first generalized ML model and an ML model for a first specific scenario, the ML model for the first specific scenario is used as a replacement ML model for the first generalized ML model, and the ML model for the first specific scenario is activated in response to deactivation of the first generalized ML model.

36. The method of claim 35, wherein a mapping between the first generalized ML model and the first scenario-specific ML model is determined based on an association between a model identifier of the first generalized ML model and a model identifier of the first scenario-specific ML model.

37. The method of claim 35, wherein a mapping between the first generalized ML model and the ML model for the first specific scenario is determined based on an association between a parameter range of the first generalized ML model and a parameter range of the ML model for the first specific scenario.

38. The method of claim 37, wherein the ML model for the first specific scenario is used for a first parameter range, the ML model for the second specific scenario is used for a second parameter range, the first generalized ML model is used for the first parameter range and the second parameter range, and the ML model for the first specific scenario and the ML model for the second specific scenario are used as backup ML models for the first generalized ML model.

39. The method of claim 31 , wherein the capabilities message comprises a maximum number of ML models supported by the wireless communication device.

40. The method of claim 39, wherein the maximum number of ML models supported by the wireless communication device is associated with the model type.

41. The method of claim 30, wherein the capabilities message comprises a set of capabilities of the wireless communication device for determining a complexity level of the wireless communication device; or The capabilities message includes the complexity level of the wireless communication device.

42. The method of claim 41, wherein the complexity level of the wireless communication device is associated with the model type.

43. The method of claim 41, further comprising: When the complexity level of the first activated ML model matches the complexity level of the wireless communication device, reporting that the deployment of the first activated ML model among the activated one or more ML models is successful; as well as When the complexity level of the first activated ML model matches the complexity level of the wireless communication device, reporting a deployment failure of the first activated ML model among the activated one or more ML models.

44. The method of claim 41, wherein the capability set of the wireless communication device comprises one or more of: The central processing unit (CPU), memory size, storage, floating point operations per second (FLOPs), power, buffer size, and bus bandwidth of the wireless communication device.

45. The method of claim 41, wherein the complexity level of the wireless communication device is obtained by comparing a threshold to a weighted sum of elements in the capability set of the wireless communication device.

46. ​​The method of claim 41, wherein the complexity level of the wireless communication device is obtained by comparing a threshold to a most significant element of the capability set of the wireless communication device.

47. The method of claim 30, wherein in response to enabling a second set of UE capabilities of the wireless communication device, disabling a first set of UE capabilities of the wireless communication device; or In response to enabling the second set of UE capabilities of the wireless communication device, enabling a first set of UE capabilities of the wireless communication device.

48. The method of claim 47, wherein each of the first set of UE capabilities and the second set of UE capabilities of the wireless communication device includes one or more of the following associated with one of the downloaded one or more ML models: Training, inference, monitoring, online training, data collection, post-processing, fine-tuning, and pre-processing.

49. The method of claim 30, wherein the wireless communication device comprises a user equipment (UE).

50. The method of claim 47, wherein the source device comprises a base station.

51. The method according to claim 50, wherein the UE performs a handover operation from an original base station to a new base station, and the UE reports the capability message to the new base station.

52. The method of claim 51 , wherein the capability message includes a status of the UE, the status of the UE indicating which ML model has been deployed or activated by the UE and how many ML models of a specific complexity level can be deployed to the UE.

53. The method according to claim 52, wherein the capability message is sent to the new base station through a physical uplink control channel PUCCH and / or a physical uplink shared channel PUSCH.

54. A wireless communication device, comprising: A processor configured to call and run a computer program stored in a memory so that a device equipped with the processor executes the method according to any one of claims 30 to 53.

55. A chip, comprising: A processor configured to call and run a computer program stored in a memory so that a device equipped with the chip executes a method according to any one of claims 30 to 53.

56. A computer-readable storage medium having a computer program stored therein, wherein the computer program causes a computer to execute the method of any one of claims 30 to 53.

57. A computer program product comprising a computer program, wherein the computer program causes a computer to perform the method of any one of claims 30 to 53.

58. A computer program, wherein the computer program causes a computer to perform the method of any one of claims 30 to 53.

Images