Life cycle management method of artificial intelligence and machine learning models in wireless networks

Abstract

The present disclosure relates generally to wireless communication networks, and in particular to configuration and provisioning of artificial intelligence (AI) and / or machine learning (ML) functions and models of both terminal devices and network nodes in wireless communication networks. For example, such AI / ML functions and models can be located at the wireless terminal side or the wireless network side. Wireless terminals and networks can perform a collaborative procedure to determine, configure, activate, or deactivate a set of AI / ML functions and models for adaptive prediction and inference for network provisioning via a series of trigger messages.

Description

Technical Field

[0001] This disclosure generally relates to wireless communication networks, and more particularly to the configuration and provision of artificial intelligence (AI) and / or machine learning (ML) functions and models for both terminal devices and network nodes in wireless communication networks. Background Technology

[0002] In wireless communication systems, determining adaptive network configurations of communication resources, especially within the air communication interface, can require lengthy and resource-intensive measurement / reporting processes and / or significant computational power. Such communication resource configurations may include, but are not limited to, beam management, channel state information (CSI) feedback compression and decompression, and wireless terminal localization / orientation. The correlation between various network conditions and these adaptive resource configurations can be learned through artificial intelligence (AI) techniques and models, and used to assist in the provisioning of wireless communication resources. Therefore, an efficient mechanism is desired to configure, select, and provision various AI models deployed on terminal devices or wireless network nodes based on the capabilities of terminal devices and the wireless network conditions between these terminal devices and the wireless network. Summary of the Invention

[0003] This disclosure generally relates to wireless communication networks, and more particularly to the configuration and provisioning of artificial intelligence (AI) and / or machine learning (ML) functions and models for both terminal devices and network nodes within wireless communication networks. For example, such AI / ML functions and models may reside on the wireless terminal side or the wireless network side. The wireless terminal and the network may perform a collaborative process to determine, configure, activate, or deactivate a set of AI / ML functions and models via a series of triggering messages for adaptive prediction and inference provisioned by the network.

[0004] In one example implementation, a method is disclosed performed by a user equipment (UE) communicating with a network (NW) in a wireless communication network. The method may include: collaborating with the NW to perform a testing process for the UE's artificial intelligence or machine learning (AI / ML) capabilities; receiving an AI / ML configuration from the NW for one or more AI / ML features, functions, or models, the one or more AI / ML features, functions, or models being selected by the NW from the UE's AI / ML capabilities based at least in part on the testing process; and performing inference using the one or more AI / ML features, functions, or models according to the AI / ML configuration.

[0005] In the above example implementation, the testing process includes: receiving a request message from the NW for testing the AI / ML features, functions, or models of the UE; performing inference using at least one AI / ML feature, function, or model to generate an inference result; and sending a response message to the NW, the response message including at least the inference result.

[0006] In any of the above example implementations, the request message includes at least one of the following: an indicator for indicating an AI / ML feature, function, or model for testing; and reference input data for the indicated applicable AI / ML feature, function, or model for testing.

[0007] In any of the above example implementations, performing the inference includes processing the reference input data using the indicated AI / ML features, functions, or models to generate the inference result.

[0008] In any of the above example implementations, the response message includes at least one of the following: an indicator for indicating the applicable AI / ML feature, function, or model used for testing; or an output data list including the inference result.

[0009] In any of the above example implementations, the response message also includes a set of UE additional conditions, which, along with the inference result, are used by the NW to evaluate the one or more AI / ML features, functions, or models.

[0010] In any of the above example implementations, the AI / ML configuration includes an RRC (Radio Resource Control) configuration structure.

[0011] In any of the above example implementations, the AI / ML configuration is included in the serving cell configuration structure.

[0012] In any of the above example implementations, the AI / ML configuration includes a list of configurations for the one or more AI / ML features, functions, or models.

[0013] In any of the example implementations described above, the AI / ML configuration includes a hierarchical list of configurations for the one or more AI / ML features, functions, or models.

[0014] In any of the above example implementations, each configuration in the configuration list for the AI / ML function includes at least one of the following: the index or ID of the AI / ML function; the inference-related configuration of the AI / ML function; the performance monitoring-related configuration of the AI / ML function; or the conventional configuration for non-AI functions corresponding to the AI / ML function.

[0015] In any of the above example embodiments, the inference-related configuration includes at least one of the following: a measurement configuration for obtaining input data for inference of the AI / ML function; an inference reporting configuration for performing inference of output data associated with the AI / ML function; a performance monitoring configuration for inference performance monitoring of the AI / ML function; or a performance reporting configuration for reporting inference performance monitoring of the AI / ML function.

[0016] In any of the example implementations described above, the AI / ML configuration is included in the Channel State Information (CSI) reporting configuration.

[0017] In any of the example implementations described above, the CSI report configuration includes a list of identifiers for the one or more AI / ML features, functions, or models.

[0018] In any of the example implementations described above, the CSI reporting configuration includes a reporting volume configuration, which includes a set of inference configurations for the one or more AI / ML features, functions, or models.

[0019] In any of the example implementations described above, the CSI reporting configuration includes a time configuration for AI / ML inference reporting.

[0020] In any of the above example implementations, the time configuration includes at least one of the following: periodic reporting; semi-persistent reporting on the Physical Uplink Shared Channel (PUSCH); semi-persistent reporting on the PUCCH (Physical Uplink Control Channel); or non-periodic reporting.

[0021] In any of the example implementations described above, the CSI report configuration includes a measurement resource configuration for inference of the one or more AI / ML features, functions, or models.

[0022] In any of the example implementations described above, the CSI reporting configuration includes a measurement resource configuration for monitoring the inference performance of the one or more AI / ML features, functions, or models.

[0023] In any of the above example implementations, utilizing the one or more AI / ML features, functions, or models includes: activating or deactivating the one or more AI / ML features, functions, or models.

[0024] In any of the above example implementations, activation or deactivation of the one or more AI / ML features, functions, or models is implicitly triggered by activating or deactivating semi-persistent CSI reports.

[0025] In any of the above example implementations, utilizing the one or more AI / ML features, functions, or models includes: activating or deactivating at least one of the one or more AI / ML features, functions, or models.

[0026] In any of the above example implementations, activation or deactivation of at least one of the one or more AI / ML features is triggered by the UE receiving signaling from the NW.

[0027] In any of the above example implementations, the signaling includes a DL MAC CE (Downlink Medium Access Control Element) structure.

[0028] In any of the above example embodiments, the DL MAC CE structure includes: a first octet having multiple bits, indicating whether an associated AI / ML function associated with the corresponding serving cell is activated or deactivated; a second octet having multiple bits, indicating the number of subsequent octets in the DL MAC CE structure for the corresponding serving cell; and the subsequent octets indicating AI / ML features and functions to be activated or deactivated for the corresponding serving cell.

[0029] In any of the above example implementations, the method further includes: instructing a lower layer on information regarding AI / ML activation or deactivation.

[0030] In any of the above example implementations, the signaling includes semi-persistent CSI report activation / deactivation signaling.

[0031] In any of the above example implementations, the method further includes: instructing a lower layer on information regarding AI / ML activation or deactivation.

[0032] In any of the above example implementations, the method further includes: reporting the inference to the NW.

[0033] In any of the above example embodiments, the inference includes at least one of the following: the top N beams having corresponding N Reference Signal Received Power (RSRP) values, where N≥1; the top N beams having RSRP values ​​of M beams among the top N beams, where N>M≥1; the top N beams having RSRP values ​​greater than a predefined RSRP threshold; or a flag indicating that if an RSRP value exists for the current beam ID, the RSRP value is inferred to be measured.

[0034] In another example implementation, a method performed by a network (NW) communicating with a user equipment (UE) in a wireless communication network is disclosed. The method may include: collaborating with the UE to perform a testing process for the UE's artificial intelligence or machine learning (AI / ML) capabilities; sending the UE an AI / ML configuration for one or more AI / ML features, functions, or models, which are selected by the NW from the UE's AI / ML capabilities based at least in part on the testing process; and receiving inference and / or measurement reports from the UE, which are generated by the UE based on the AI / ML configuration.

[0035] In the above example implementation, the testing process includes: sending a request message to the UE to test the AI / ML features, functions, or models of the UE; and receiving a response message from the UE, the response message including at least the inference result generated by the UE using at least one AI / ML feature, function, or model.

[0036] In any of the above example implementations, the request message includes at least one of the following: an indicator for indicating an AI / ML feature, function, or model for testing; and reference input data for the applicable AI / ML feature, function, or model for the testing indication.

[0037] In any of the above example implementations, the response message includes at least one of the following: an indicator for indicating the applicable AI / ML feature, function, or model used for testing; or an output data list including the inference result.

[0038] In any of the above example implementations, the response message also includes a set of UE additional conditions, which, along with the inference result, are used by the NW to evaluate the one or more AI / ML features, functions, or models.

[0039] In any of the above example implementations, the AI / ML configuration includes an RRC configuration structure.

[0040] In any of the above example implementations, the AI / ML configuration is included in the serving cell configuration structure.

[0041] In any of the above example implementations, the AI / ML configuration includes a list of configurations for the one or more AI / ML features, functions, or models.

[0042] In any of the example implementations described above, the AI / ML configuration includes a hierarchical list of configurations for the one or more AI / ML features, functions, or models.

[0043] In any of the above example implementations, each configuration in the configuration list for the AI / ML function includes at least one of the following: an index or ID of the AI / ML function; inference-related configurations for the AI / ML function; performance monitoring-related configurations for the AI / ML function; or conventional configurations for non-AI functions corresponding to the AI / ML function.

[0044] In any of the above example embodiments, the inference-related configuration includes at least one of the following: a measurement configuration for obtaining input data for inference of the AI / ML function; an inference reporting configuration for performing inference of output data associated with the AI / ML function; a performance monitoring configuration for inference performance monitoring of the AI / ML function; or a performance reporting configuration for reporting inference performance monitoring of the AI / ML function.

[0045] In any of the example implementations described above, the AI / ML configuration is included in the Channel State Information (CSI) reporting configuration.

[0046] In any of the example implementations described above, the CSI report configuration includes a list of identifiers for the one or more AI / ML features, functions, or models.

[0047] In any of the example implementations described above, the CSI reporting configuration includes a reporting volume configuration, which includes a set of inference configurations for the one or more AI / ML features, functions, or models.

[0048] In any of the example implementations described above, the CSI reporting configuration includes a time configuration for AI / ML inference reporting.

[0049] In any of the above example implementations, the timing configuration includes at least one of the following: periodic reporting; semi-persistent reporting on the Physical Uplink Shared Channel (PUSCH); semi-persistent reporting on the PUCCH; or non-periodic reporting.

[0050] In any of the example implementations described above, the CSI report configuration includes a measurement resource configuration for inference of the one or more AI / ML features, functions, or models.

[0051] In any of the example implementations described above, the CSI reporting configuration includes a measurement resource configuration for monitoring the inference performance of the one or more AI / ML features, functions, or models.

[0052] In any of the above example implementations, the method further includes: sending signaling to the UE to activate or deactivate at least one of the one or more AI / ML features, functions, or models.

[0053] In any of the above example implementations, the signaling includes a DL MAC CE structure.

[0054] In any of the above example embodiments, the DL MAC CE structure includes: a first octet having multiple bits indicating whether an associated AI / ML function associated with the corresponding serving cell is activated or deactivated; a second octet having multiple bits indicating the number of subsequent octets in the DL MAC CE structure for the corresponding serving cell; and the subsequent octets indicating AI / ML features and functions to be activated or deactivated for the corresponding serving cell.

[0055] In any of the above example implementations, the signaling includes semi-persistent CSI report activation / deactivation signaling.

[0056] In any of the above example embodiments, the inference includes at least one of the following: the top N beams having corresponding N reference signal received power (RSRP) values, where N≥1; the top N beams having RSRP values ​​of M beams among the top N beams, where N>M≥1; the top N beams having RSRP values ​​greater than a predefined RSRP threshold; or a flag indicating that if an RSRP value exists for the current beam ID, the RSRP value is inferred to have been measured.

[0057] A UE or NW that discloses any one of the above-described methods is disclosed. The UE or NW may include a processor and a memory, wherein the processor is configured to read computer code from the memory to cause the UE or NW to perform any one of the above-described methods.

[0058] A non-transitory computer-readable program medium is also disclosed, on which computer code is stored. The computer code, when executed by the processor of the UE or NW according to any of the above methods, is configured to cause the processor to implement any of the above methods.

[0059] The above embodiments and other aspects and alternatives to their implementation are described in more detail in the following drawings, description and claims. Attached Figure Description

[0060] Figure 1 An example wireless communication network including a wireless access network, a core network, and a data network is shown.

[0061] Figure 2 An example radio access network is shown, comprising multiple mobile stations / terminals or user equipment (UEs) and radio access network nodes, which communicate with each other via an over-the-air wireless communication interface.

[0062] Figure 3 An example radio access network (RAN) architecture is shown.

[0063] Figure 4 An example communication protocol stack including various network layers is shown in a wireless access network node or wireless terminal device.

[0064] Figure 5 This demonstrates a typical process for AI / ML lifecycle management of the UE-side AI model.

[0065] Figure 6 This demonstrates an example process for a UE-triggered AI / ML capability report.

[0066] Figure 7 An example uplink MAC CE for reporting available UE AI / ML functions is shown.

[0067] Figure 8 An example process for testing applicable AI / ML functions is shown.

[0068] Figures 9A to 9E An example RRC configuration structure for AI / ML functions is shown.

[0069] Figure 10 An example DL MAC CE for signaling activation / deactivation of AI / ML functions is shown.

[0070] Figure 11 The example process of attaching a change report to the UE is shown. Detailed Implementation

[0071] This disclosure will now be described in detail below with reference to the accompanying drawings, which form a part of this disclosure and illustrate specific examples of embodiments by way of illustration. This disclosure may be embodied in various different forms, and therefore the subject matter covered or claimed is intended to be construed as not being limited to any of the embodiments set forth below.

[0072] Throughout the specification and claims, terms may have nuanced meanings suggested or implied in the context beyond their explicitly stated meanings. Similarly, the phrases “in one embodiment” or “in some embodiments” as used herein do not necessarily refer to the same embodiment, and the phrases “in another embodiment” or “in other embodiments” as used herein do not necessarily refer to different embodiments. For example, the claimed subject matter is intended to include, in whole or in part, exemplary embodiments or combinations of embodiments.

[0073] For example, the claimed subject matter is intended to include, in whole or in part, combinations of exemplary embodiments or implementations. For example, terms used herein, such as “and,” “or,” or “and / or,” can include a variety of meanings, which may depend at least in part on the context in which these terms are used. Generally, “or,” when used with an associative list, such as A, B, or C, is intended to mean A, B, and C (used herein in an inclusive sense) and A, B, or C (used herein in an exclusive sense). Furthermore, depending at least in part on the context, the terms “one or more” or “at least one” as used herein can be used to describe any feature, structure, or characteristic in a singular sense, or can be used to describe a combination of features, structures, or characteristics in a plural sense. Similarly, terms such as “a,” “an,” or “the” can also be understood to convey either a singular or a plural usage, depending at least in part on the context. Furthermore, also depending at least in part on the context, the terms “based on” or “determined by” can be understood as not necessarily intended to convey an exclusive set of factors, but rather to allow for the existence of additional factors that are not necessarily explicitly described.

[0074] This disclosure generally relates to wireless communication networks, and more particularly to the configuration and provisioning of artificial intelligence (AI) and / or machine learning (ML) functions and models in both terminal devices and network nodes within wireless communication networks. For example, such AI / ML functions and models may reside on the wireless terminal side or the wireless network side. The wireless terminal and the network may perform a collaborative process to determine, configure, activate, or deactivate a set of AI / ML functions and models via a series of triggering messages for adaptive prediction and inference provisioned by the network.

[0075] Wireless Network Overview

[0076] Figure 1The example wireless communication network shown as 100 may include wireless terminal devices or user equipment (UEs) 110, 111, and 112, a carrier network 102, various service applications 140, and other data networks 150. The wireless terminal devices or UEs may alternatively be referred to as wireless terminals. The operator network 102 may, for example, include access network nodes 120 and 121, and a core network 130. The operator network 110 may be configured to transmit voice, data, and other information (collectively referred to as data services) between UEs 110, 111, and 112, between UEs and service applications 140, or between UEs and other data networks 150. Access network nodes 120 and 121 may be configured as various wireless access network nodes (WANNs, alternatively referred to as wireless base stations) to interact with the UE on one side of a communication session and the core network 130 on the other side. The term "access network" can be more broadly used to refer to the combination of wireless terminal devices 110, 111, and 112 with access network nodes 120 and 121. The radio access network can also be referred to as the radio access network (RAN). The core network 130 may include various network nodes configured to control communication sessions and perform network access management and service routing. Service applications 140 may be hosted by various application servers deployed outside the core network 130 but connected to it. Similarly, other data networks 150 may also be connected to the core network 130.

[0077] exist Figure 1 In the example wireless communication network 100, each UE can communicate with each other via a wireless access network. For example, UEs 110 and 112 can connect to the same access network node 120 and communicate via that access network 120. UEs can communicate with each other via both the access network and the core network. For example, UE 110 can connect to access network node 120, while UE 111 can connect to access network node 121, and thus, UEs 110 and UE 111 can communicate with each other via access network nodes 120 and 121 and the core network 130. UEs can also communicate with the serving application 140 and the data network 150 via the core network 130. Furthermore, as shown in 113, each UE can communicate directly with each other via sidelink communication.

[0078] Figure 2An example system diagram of a radio access network 120 including a WANN 202 serving UEs 110 and 112 via air interface 204 is also shown. Radio transmission resources for air interface 204 include a combination of frequency resources, time resources, and / or spatial resources. Each of UEs 110 and 112 can be a mobile or fixed terminal device equipped with a mobile access unit (such as a SIM / USIM (Subscriber Identity Module / Universal Subscriber Identity Module) module) for accessing the wireless communication network 100. Each of UEs 110 and 112 can be implemented as a terminal device including, but not limited to, mobile phones, smartphones, tablets, laptops, in-vehicle communication devices, roadside communication devices, sensor devices, smart appliances (such as televisions, refrigerators, and ovens), or other devices capable of wireless communication over a network. Figure 2 As shown, each of the plurality of UEs (such as UE 112) may include transceiver circuitry 206 coupled to one or more antennas 208 to enable wireless communication with WANN 120 or with another UE (such as UE 110). Transceiver circuitry 206 may also be coupled to processor 210, which may also be coupled to memory 212 or other storage devices. Memory 212 may be transient or non-transient and may store computer instructions or code therein that, when read and executed by processor 210, cause processor 210 to implement the various methods described herein.

[0079] Similarly, WANN 120 may include a wireless base station or other wireless network access point capable of wirelessly communicating with one or more UEs via air interface 204 and with core network 130. For example, WANN 120 may be implemented as, but is not limited to, a 2G base station, a 3G node B, an LTE eNB, a 4G LTE base station, a 5G gNB, a 5G NR base station, a 5G centralized unit base station, a 5G distributed unit base station, or a 6G base station. Each type of WANN can be configured to perform a corresponding set of wireless network functions. WANN 202 may include transceiver circuitry 214 coupled to one or more antennas 216 (which may include various forms of antenna towers 218) to enable wireless communication with UEs 110 and 112. Transceiver circuitry 214 may also be coupled to one or more processors 220, which may also be coupled to memory 222 or other storage devices. The memory 222 may be transient or non-transient and may store instructions or code therein that, when read and executed by one or more processors 220, cause one or more processors 220 to perform the various functions of the WANN 120 described herein.

[0080] Such as Figure 2 In the example radio access network described, data packets can be transmitted as protocol data units (PDUs). The data can be packaged into PDUs using nested and / or layered protocol headers at various network layers. Once a connection is established between the transmitting and receiving ends (e.g., a radio link control (RRC) connection), PDUs can be transmitted between the transmitting device or receiving end (these terms are used interchangeably) and the receiving device or receiving end (also the terms are used interchangeably). Either the transmitting or receiving device can be a wireless terminal device, such as... Figure 2 Devices 110 and 120; or wireless access network nodes, such as Figure 2 Node 202. Each device can be both a transmitting and receiving device for bidirectional communication.

[0081] Figure 1 The core network 130 may include various network nodes or functions that are geographically distributed and interconnected to provide network coverage for the service area of ​​the carrier network 102. These network nodes may be implemented as dedicated hardware network nodes. Alternatively, these network nodes may be virtualized and implemented as virtual machines or software entities. Each of these network nodes may be configured with one or more types of network functions that collectively provide configuration and routing capabilities for the core network 130.

[0082] Returning to the Radio Access Network (RAN). Figure 3 An example RAN 340 communicating with core network 310 and radio terminals UE1 through UE7 is shown. RAN 340 may include one or more types of radio base stations or WANNs 320 and 321, which may include, but are not limited to, gNBs, eNodeBs, NodeBs, or other types of base stations. RAN 340 may transmit back to core network 310. WANN 320 may also include, for example, multiple independent access network nodes in the form of a central unit (CU) 322 and one or more distributed units (DUs) 324 and 326. CU 322 is connected to DU1 324 and DU2 326 via various interfaces (e.g., F1 interfaces). The F1 interface may also include, for example, an F1-C interface and an F1-U interface, which may be used to carry control plane information and user plane data, respectively. In some embodiments, CU may be a gNB central unit (gNB-CU), and DU may be a gNB distributed unit (gNB-DU). Although the various implementations described below are provided in the context of 5G cellular wireless networks, the basic principles described herein are applicable to other types of wireless access networks, including but not limited to current or future generations of cellular networks (such as 4G LTE and 6G networks) as well as Wireless Fidelity (Wi-Fi), Bluetooth, ZigBee, and Worldwide Interoperability for Microwave Access (WiMax) networks.

[0083] The UE can connect to the network via the air interface through WANN 320. The UE can be served by at least one cell. Each cell is associated with a coverage area. These cells may be alternatively referred to as serving cells. The coverage areas between cells may partially overlap. Each UE can potentially connect to or be connectable to more than one cell while actively communicating with at least one cell. Figure 1 In the example, UE 1, UE 2, and UE 3 can be served by cell 1 330 of DU1, while UE 4 and UE 5 can be served by cell 2 332 of DU1, and UE 6 and UE 7 can be served by cell 3 associated with DU 2. In some implementations, a UE can be served by two or more cells simultaneously. Each UE can be mobile, and the signal strength and quality from each cell at the UE can depend on the UE's location and mobility.

[0084] In some example implementations, Figure 3 The cells shown can be alternatively referred to as serving cells. Serving cells can be grouped into serving cell groups (CGs). A serving cell group can be a master CG (MCG) or a secondary CG (SCG). Within each type of cell group, there can be one master cell and one or more secondary cells. For example, the master cell in an MSG can be called a PCell, while the master cell in an SCG can be called a PSCell. Secondary cells in either an MCG or an SCG can be called SCells. Master cells including both PCells and PSCells can be collectively referred to as spCells (specific cells). All of these cells can be referred to as serving cells or cells. Unless otherwise specified, the terms "cell" and "serving cell" can be used interchangeably in a general manner. The term "serving cell" can refer to a cell that is currently serving, will serve, or may serve a UE. In other words, a "serving cell" may not currently be serving a UE. While the various embodiments described below may sometimes refer to one type of serving cell among the above-described serving cell types, the basic principles apply to all types of serving cells in both types of serving cell groups.

[0085] Figure 4 Further shown Figures 1 to 3 A simplified view of the various network layers involved in transmitting user plane PDUs from transmitting device 402 to receiving device 404 in an example wireless access network. Figure 4 It is not intended to include all basic equipment components or network layers used to process the transmission of PDUs. Figure 4 This illustrates that data packaged by the network layer 420 at the transmitting device 402 can be transmitted via the Packet Data Convergence Protocol (PDCP) layer of the transmitting device (i.e., the PDCP layer). Figure 4 The data is transmitted to the corresponding upper layer 430 (e.g., radio resource control or RRC layer) at the receiving device 304, including the Radio Link Control (RLC) layer 422, the Physical (PHY) layer of the transmitting and receiving devices, the radio interface (as shown in 406), and the Media Access Control (MAC) layer 434 and RLC layer 432 of the receiving device. Various network entities in each of these layers can be configured to handle the transmission and retransmission of PDUs.

[0086] exist Figure 4 In the middle, the upper layer 420 can be called layer 3 or L3, while layers such as RLC layer and / or MAC layer and / or PDCP layer ( Figure 4Intermediate layers such as (not shown in the diagram) can be collectively referred to as Layer 2 or L2, while the term "Layer 1" is used to refer to layers such as the Physical Layer and the Radio Interface Association Layer. In some cases, the term "low layer" can be used to refer to the set of L1 and L2, while the term "high layer" can be used to refer to Layer 3. In some cases, the term "lower layer" can be used to refer to layers among L1, L2, and L3 that are below the current reference layer. Control signaling can be initiated and triggered within each of L1 to L3 and within each of the network layers therein. These signaling messages can be encapsulated and concatenated into lower-layer packets and transmitted via allocated control or data air radio resources and the air interface. The term "layer" typically includes its various corresponding entities. For example, the MAC layer includes corresponding MAC entities that can be created. For example, Layer 1 includes PHY entities. As another example, Layer 2 includes MAC layer / entities, RLC layer / entities, Service Data Adaptation Protocol (SDAP) layer, and / or PDCP layer / entities.

[0087] AI / ML-assisted wireless network provisioning and configuration

[0088] AI and ML (often referred to as AL) can facilitate more efficient provisioning and provisioning in wireless networks. At the heart of a general AI framework are various AI models. AI models typically contain a large number of model parameters, which are determined through a training process where correlations in the training dataset are learned and embedded into the trained model parameters. Therefore, the trained model parameters can be used to generate inferences or predictions from a set of input datasets. AI models are particularly suitable for situations where there is little traceable determinism or analytical derivation path between the input data and the output, but correlations in the data can be identified from historical data and embedded into the AI ​​model through the training process.

[0089] In wireless communication systems such as those described above, determining adaptive network configurations can rely on empirical characteristics and may require lengthy measurement processes and / or significant computational power. Such configuration types may include, but are not limited to, air interface beam management, channel state information (CSI) feedback compression and decompression, and wireless terminal location. The correlation between various network conditions and these adaptive configurations can be learned via AI technology. Therefore, using AI models to assist network configuration can help reduce the amount of measurement and computational requirements, thereby providing more agile and efficient network configurations. Thus, it may be desirable to provide a mechanism for providing AI models based on the AI ​​capabilities of the terminal device and network conditions to assist in the adaptive determination of these network configurations.

[0090] For example, AI technology can be applied to beam management in air communication interfaces. In current implementations, beam management typically relies on exhaustive search beam scanning and measurements. In other words, the network (NW) performs a full scan of the beam by sending a sufficient number of reference signals. The UE can be configured to monitor and measure each reference signal and then report the measurements to the NW so that the NW can determine the optimal beam the UE should switch to. However, this process is resource-intensive, power-intensive, and time-consuming. Utilizing a trained AI model that embeds learned correlations between various network condition parameters, it may be possible to infer the optimal beam with reasonable accuracy using fewer measurements (or fewer reference signals), thus achieving so-called AI / ML-based spatial beam management. In some implementations, the AI ​​model can help identify the inference of the optimal candidate beam in future time instances using other network conditions, and then only scan and measure the candidate beams to select the beam to use in the future time instance, thus achieving so-called AI / ML-based temporal beam management. Furthermore, since beam configuration is closely related to the UE's location, AI technology can be further used to infer or predict the UE's trajectory or location, thereby indirectly helping to select the optimal beam.

[0091] For example, AI technology can be applied to Channel State Information (CSI) feedback. Traditionally, CSI feedback can be achieved using codebooks known to both the UE and NW. The UE measures the CSI and obtains the measurement result, then maps the measurement result to the nearest vector in the codebook and sends the index of that vector to the NW to save air interface resources. However, since the codebook is not infinite or can not change dynamically over time, there is always some degree of mismatch, leading to uncontrollable CSI feedback errors as the wireless environment changes. Therefore, AI can be applied, for example, to compression-decompression of CSI feedback. Specifically, the CSI report can be compressed by a UE-side AI model and decompressed by a corresponding NW-side AI model. Such AI models can be initially trained and continuously developed over time and with the accumulation of network conditions. In some example implementations, the AI ​​model can help identify inferences about future CSI feedback values ​​using other network conditions, and then only measure the current reference signaling used for CSI, thereby achieving so-called AI / ML-based CSI prediction.

[0092] For example, AI technology can be applied to UE positioning. Traditional UE positioning methods rely on Positioning Reference Signal (PRS) or Sounding Reference Signal (SRS). Examples include the DL positioning reference signal and the uplink sounding reference signal. Regardless of the alternative method, the line-of-sight (LOS) beam is a key beam that needs to be identified in order to generate the most accurate position estimate through triangulation on the NW side. However, in most cases, it is difficult to identify the LOS beam from other non-line-of-sight (NLOS) beams, resulting in inaccurate UE positioning. On the other hand, trained AI models can identify various patterns and correlations in the PRS and SRS to extract LOS information and provide more accurate UE positioning.

[0093] In some example implementations, the AI ​​model may reside on the wireless terminal side. In other implementations, the AI ​​model may reside on the network (NW) side. The AI ​​model may be provided as a service. The AI ​​model may be retrained and updated as needed. The UE or NW may determine which models to retrieve or retain, and how to update and configure them to assist wireless communication, including but not limited to beam management, CSI feedback, UE positioning, etc. The selection, retrieval, storage, and execution of AI models may depend on the capabilities of the terminal device and / or NW. The terminal and NW may be configured to transmit such capabilities and network conditions to provide the selection, configuration, and use of AI models over time; this is often referred to as AI / ML Life Cycle Management (LCM).

[0094] AI / ML lifecycle management (LCM) for UE-side AI features / functions / models

[0095] AI / ML LCM can be performed at various levels or granularities. In one example implementation, AI / ML LCM can be performed at the level of AI / ML-based feature groups, such as AI / ML-based beam management groups (e.g., feature groups), including AI / ML-based spatial beam management aspects (e.g., features) and AI / ML-based temporal beam management aspects (e.g., features), and AI / ML-based CSI feedback groups, including AI / ML-based CSI feedback enhancement aspects (e.g., features) and AI / ML-based CSI prediction aspects (e.g., features). In another example implementation, AI / ML LCM can be performed at the level of AI / ML-based features. In another example implementation, AI / ML LCM can be performed at the level of AI / ML functions. In yet another example implementation, AI / ML LCM can be performed at the level of AI / ML models. The terms AI / ML feature groups, features, functions, and models are used only to indicate the various levels at which AI / ML can be configured and applied. They can be hierarchically related. In some cases, they can overlap. They can be divided in any suitable manner to facilitate the configuration and management of AI / ML use. Each of these levels can itself be hierarchical. For example, an AI / ML model may include a lower-level AI / ML model as an internal component. AI / ML features may include lower-level sub-features, and similarly, AI / ML functions may include lower-level sub-functions.

[0096] For example, an AI / ML model can refer to a specific training algorithm that processes one or more inputs to generate a prediction as output. AI / MI models may include components such as various types of neural networks, regression algorithms, support vector machine algorithms, K-nearest neighbors algorithms, random forests, K-means clustering, principal component analysis, Bayesian networks, etc.

[0097] For example, AI / ML models can be trained to support specific AI features or AI functions. AI / ML features or functions can be implemented by different AI models, which may vary in internal architecture, hyperparameters, trained model parameters, inputs, computational resource requirements, complexity, prediction accuracy, etc. For instance, an AI / ML function may comprise a single AI / ML model or a set of AI / ML models.

[0098] For example, an AI / ML-based feature may include one or more AI / MI functions. In some example implementations, an AI / ML feature may be synonymous with an AI / MI function (e.g., an AI / ML function may represent a corresponding AI / ML-based feature). As another example, an AI / ML-based feature group may include one or more AI / ML-based features to form a category of features.

[0099] As an example only, AI / ML features can refer to the classification of AI / ML-based spatial beam prediction, AI / ML-based temporal beam prediction, AI / ML-based CSI feedback compression and decompression, AI / ML-based CSI prediction, AI / ML-based temporal / spatial cell measurement result prediction for mobility, and AI / ML-based temporal / spatial beam prediction for mobility.

[0100] In some example implementations, AI / ML LCM may include at least one of the following non-limiting aspects:

[0101] ●AI / ML features (feature groups) / functions / model recognition.

[0102] ● Applicable AI / ML features / functions / model reports between wireless terminals and NW.

[0103] ●AI / ML feature / function / model control (including selection, activation, deactivation, and rollback).

[0104] ● AI / ML feature / function / model performance monitoring.

[0105] ●AI / MI features / functions / models can be located on the UE side or the NW side. The LCM of AI / MI features / functions / models on the UE side can depend on the UE capabilities.

[0106] UE-side AI / ML LCM - Overview

[0107] Figure 5 The document illustrates an example implementation of the aforementioned LCM aspect of the wireless terminal-side (UE-side) AI / ML model involving UE 502 and NW 504, including the following general steps:

[0108] ●Step 1: A UE capability reporting process for one or more AI / ML-based features / functions / models can be performed between UE 502 and NW 504.

[0109] ●Step 2: NW 504 and UE 502 may additionally perform the applicable AI / ML feature / function / model reporting process.

[0110] ●Step 3: NW 504 and UE 502 can perform a preparation phase to test AI / ML features / functions / models.

[0111] ●Step 4: NW 504 can configure the RRC configuration of AI / ML features / functions / models to UE 502 based on UE capabilities and / or applicable features / functions / model reports.

[0112] ●Step 5: NW 504 can send a message to UE 502 to activate / deactivate AI / ML features / functions / models.

[0113] ●Step 6: UE 502 can activate or deactivate AI / ML features / functions / models accordingly and perform predictions and inferences.

[0114] ●Step 7: UE 502 may then perform a prediction / reference or actual measurement and send the prediction and inference or actual measurement to NW 504 via one or more inference reports and / or measurement reports.

[0115] Therefore, in the general example process described above, the UE may first report a set of supported AI / ML-based features / functions / models to the NW based on the UE's capabilities. Then, the UE and the NW may collaboratively determine: (1) among the supported AI / ML-based features / functions / models, the applicable or suitable features / functions / models to be activated by the UE, based on capabilities, network conditions, and various other factors, and / or (2) among the supported features / functions / models, the features / functions / models that are no longer applicable or suitable and that the UE should deactivate.

[0116] UE-side AI / ML LCM – UE AI / ML Capability Report

[0117] In the above Figure 5 In step 1, the capability reporting process for one or more AI / ML-based features / functions / models can be triggered by the UE or NW, such as... Figure 6 As shown in the example. Step 608 shows the AI / ML capability report triggered by the NW, while step 606 is used for the AI / ML capability report triggered by the UE, which is before step 608 of the AI / ML capability report triggered by the NW.

[0118] Specifically, in one example implementation, UE 602 can proactively trigger an AI / ML capability report by sending a trigger message to the NW (e.g., a base station or core network node) in 606. The NW can then accept the request and initiate the UE AI / ML capability report process, as shown in 608. In other example implementations, the UE AI / ML capability report can be triggered by the NW and then executed between the UE and the NW as shown in 608, without the UE triggering step 606.

[0119] In some example implementations, the UE may support a superset of AI / ML features / functions / models (supported features / functions / models). In some cases, the UE may download / store only a subset of the superset models locally based on its location and other factors (available features / functions / models). While the supported features / functions / models of the superset may remain stable over time, they may still change. Even if the features / functions / models of the superset remain stable, the available features / functions / models may be more fluid and change more frequently over time. Therefore, the UE may determine to trigger the UE AI / ML capability report in 606 under at least one of the following conditions:

[0120] ●The AI / ML features / functions / models supported by the superset at the UE have changed, which may include, but are not limited to: Since the last UE capability report, at least one supported AI / ML feature / function / model has been updated. Since the last UE capability report, at least one new AI / ML feature / function / model has been supported. Since the last UE capability report, at least one AI / ML feature / function / model has been removed from the superset. Since the last UE capability report, the overall computing resources for AI / ML features / functions / models have been changed from hyper-centralized.

[0121] ●The actual AI / ML features / functions / models stored at the UE have been changed, which may include, but are not limited to: Since the last UE capability report, at least one AI / ML feature / function / model stored at the UE has been updated. Since the last UE capability report, the UE has acquired at least one new AI / ML feature / function / model. Since the last UE capability report, at least one stored feature / function / model has been removed from the UE. Since the last UE capability report, the overall computing resources available for AI / ML features / functions / models have been changed.

[0122] ● At least one AI / ML feature / function / model has been configured by NW for UE.

[0123] In some example implementations, the UE trigger message 606 may be sent using one of the following formats: uplink (UL) MAC control element (MAC CE); or UL RRC signaling (e.g., UE Assistance Information (UAI) message); or protocol signaling terminating between the AI / ML logical layer and the NW logical entity / unit.

[0124] In some example implementations, such trigger message 606 of the UE may include or indicate at least one of the following information or information items:

[0125] ● One or more AI / ML feature / function / model indicators: used to indicate that an AI / ML feature / function / model has been changed.

[0126] ● Change type indication, where the change type may include, for example: 1) adding AI / ML features / functions / models; 2) updating stored AI / ML features / functions / models; 3) releasing the AI / ML feature / function / model from a supported AI / ML feature / function / model.

[0127] The information in UE trigger message 606 can be used by NW to determine whether and when... Figure 6 The actual request for UEAI / ML capability report in error 608.

[0128] exist Figure 6 In step 608, the NW can send an inquiry message to the UE for UE AI / ML capability reporting (either as a trigger message for the NW to initiate UE AI / ML capability reporting or as a response to a 606 trigger message from the UE when the NW determines that a report is requested). This inquiry message can be referred to as... UECapacityEnquiring ,like Figure 6 As instructed.

[0129] In response, the UE can then send a UE AI / ML capability report to the NW. The message containing this report can be called... UECapability ,like Figure 6 As indicated. In some example implementations, UECapability The message may include or indicate at least one of the following information or information items:

[0130] ● Supported AI / ML-based features or feature groups (list): A list indicating the AI / ML-based features supported by the UE (these features may be supported but may or may not be available at the UE).

[0131] ● A list of one or more supported AI / ML-based functions: This list indicates the AI / ML-based functions supported by the UE for AI / ML-based features (these functions may be supported, but may be available or unavailable at the UE).

[0132] ● One or more supported AI / ML models: This indicates the list of AI / ML models supported by the UE for AI / ML-based features or functions.

[0133] ●Frequency band or frequency band list indicator: Used to indicate the frequency bands supported for each AI / ML-based feature / function / model.

[0134] ● One or more indicators for the computational resource consumption of each AI / ML feature / function / model: quantified values ​​indicating the computational resource consumption of AI / ML-based features / functions / models if activated.

[0135] ● Overall computing resource indicator for AI / ML: Used to indicate the maximum quantized value of computing resources that the UE can support for AI / ML.

[0136] ● One or more indicators for the preparation time of each AI / ML feature / function / model: used to indicate the maximum or minimum preparation time from when the UE receives the activation signaling for the AI / ML feature / function / model to when the AI / ML feature / function / model can be inferred to actually be executed.

[0137] ● Training configuration (e.g., UE settings, gNB settings), scenario, and condition indications: These indicate the configuration, scenario, and conditions on which the AI / ML-based models within the AI / ML-based features / functions / models supported by the UE are trained, including but not limited to:

[0138] o For AI / ML-based spatial beam prediction and AI / ML-based temporal beam prediction:

[0139] - Information items related to the scene: • Inter-Site Distance (ISD) indicates the cell range associated with AI / ML training. Supported ISD values ​​include, for example: UMa, UMi, 200 meters, and 500 meters. • The antenna height of the base station in the cell associated with AI / ML training. Supported base station antenna height values ​​can be 1 meter, 2 meters, 5 meters, 10 meters, etc. • NLOS probability of NLOS wireless propagation associated with AI / ML training. • Indoor / Outdoor indicator, used to indicate whether the scene is indoor or outdoor and / or the indoor / outdoor ratio for AI / ML training.

[0140] - Information items for gNB settings: • DL Tx beamcodebook indicator, used to indicate the DL TX beamcodebook used for AI / ML model training. • An indication of the gNB antenna array dimension associated with AI / ML model training. • Downtilt angle of the gNB antenna associated with AI / ML model training.

[0141] - Information items for training beam conditions and beam sets (set A may refer to the beam group containing the best K beams predicted by the AI / ML model / feature, set B may refer to the beams that are measured and whose corresponding measurements and / or associated beam IDs can be used as inputs to the AI / ML model / feature, while the full beam set may refer to all beams, e.g., all 64 beams in a specific configuration): • Beam pattern indicator for set B: Used to indicate the beam pattern of set B compared to set A. For example, set B used for training could be a 1 / 4 subset of set A and uniformly distributed. Alternatively, set B might not be a subset of set A. Yet another example is that set B could be an SSB and set A could be a CSI-RS, or vice versa. • Beam pattern of set A: Used to indicate the beam pattern of set A compared to the full beam set (e.g., 64 beams). For example, beam set A can be a 1 / 4 subset of the full beam set and be uniformly distributed, or beam set A can be the full beam set, etc.

[0142] - Information items related to UE settings associated with model training, such as: •UE Speed: Used to indicate UE speed information related to AI / ML feature / function / model training. • UE Orientation: Used to indicate the UE orientation associated with AI / ML feature / function / model training. In one implementation, it indicates the maximum value of the UE orientation change of the associated AI / ML feature / function / model, such as 45 degrees, 90 degrees, 120 degrees, 180 degrees, etc. •UE RX codebook. •UE antenna array dimension.

[0143] - Explicitly include the parameters (set) of one or more of the above information items with IDs.

[0144] - Implicitly indicates one or more of the above-mentioned parameter sets with IDs (e.g., cell identifiers (e.g., CGI, PCI, etc.), which may include information items about NW-specific configurations / settings, cell scenarios, etc.).

[0145] o For AI / ML-based CSI feedback compression / decompression (e.g., two-sided models):

[0146] - Model ID or list of model IDs: Model IDs used to indicate the UE portion supported by the UE in the two-sided AI / ML model list.

[0147] - Training dataset ID or list of training data IDs: Dataset IDs used to indicate a list of datasets used to train the UE portion of the two-sided AI model.

[0148] - UE setting indication: used to indicate the transceiver unit (TxRU) mapping pattern, such as [2, 8, 2], [4, 4, 2], etc.

[0149] - Scene indication: A training scene used to indicate the UE portion of the dual-sided functions / models supported by the UE, which may include at least one of the following: • Inter-site distance (ISD) is used to indicate the cell range associated with AI / ML training. Supported ISD values ​​can be, for example, UMa, UMi, 200 meters, and 500 meters. Outdoor / Indoor indicator, used to indicate indoor, outdoor, and indoor / outdoor ratios for training purposes. • Targeting the frequency range or band for training.

[0150] o For AI / ML-based temporal CSI prediction:

[0151] - Scenario-related information items similar to the list above for AI / ML-based temporal beam management.

[0152] - Similar gNB settings information items to the list above for AI / ML-based temporal beam management.

[0153] - Similar UE settings information items to the list above for AI / ML-based temporal beam management.

[0154] In some example implementations, the signaling structure for the UE capability report of AI / ML-based features / functions / models (608) can be carried in... UE-NR-Capability (UE-NR-capability) or RF Parameters In supportedBandListNR (List of Supported Bands NR)The RRC information element. In some alternative example implementations, UE capability reports for features, functions, and models can be separate or independent. As just one example, information related to AI / ML-based features or functions in the report can be carried within... UE-NR-Capability In the RRC information cells, the AI / ML functions corresponding to AI / ML-based features or the AI / ML models corresponding to AI / ML functions can be found in... RF-parameters In supportedBandListNR It is carried and transmitted in the RRC cell.

[0155] In some example implementations, the signaling structure for the UE capability report of 608 based on AI / ML features / functions / models can be carried in protocol signaling terminating between the AI / ML logical layer and the NW logical entity / unit.

[0156] In one example implementation of coordinating the computing power used for AI / ML between the MN and SN in a DC (e.g., dual-connectivity) scenario, inter-node information regarding the computing power used for AI / ML can be introduced between the MN and SN. The SN / MN can send information to the MN / SN about the computing power consumed by currently activated AI / ML features / functions / models to avoid exceeding the maximum computing power reported in the UE capabilities. In one example implementation, such information from the MN to the SN can be a parameter describing a quantified value of the computing power consumed by the MN. In another example implementation, such information from the SN to the MN can be a parameter describing a quantified value of the computing power consumed by the MN. In yet another example implementation, such information from the MN to the SN can be a parameter describing a maximum quantified value of the computing power that the SN can consume for AI / ML.

[0157] In another example implementation of coordinating the computational capabilities of MN and SN for AI / ML in a DC scenario, a UL MAC CE may be introduced. This UL MAC CE can be triggered by the MAC entity when at least one of the following conditions is met:

[0158] ● AI / ML features / functions / models have been activated or deactivated for this MAC entity.

[0159] ● The total computing power quantization value of the activated AI / ML features / functions / models in this MAC entity is greater than or equal to the pre-configured / predefined maximum value.

[0160] In one example implementation of generating this UL MAC CE, at least one of the following conditions may need to be met:

[0161] ●UL authorization has been received for MAC entities other than the MAC entity that triggered the UL MAC CE.

[0162] ● This UL authorization has the capability to accommodate the UL MAC CE according to the LCP process.

[0163] ●UL authorization has been received for any MAC entity.

[0164] In one example implementation, the UL MAC CE may include at least one of the following:

[0165] ● The quantized value of the computing power already used for AI / ML;

[0166] ● The maximum quantization value of computing power used for AI / ML;

[0167] ● Remaining quantized value of computing power used for AI / ML.

[0168] In one example implementation of coordinating AI / ML computing capabilities between a first NW and a second NW in a multi-SIM scenario, UAI (e.g., UE assistance information) messages from the UE to the NW can be utilized.

[0169] In one example implementation where the NW configures the UAI to the UE, the UAI configuration may include at least one of the following:

[0170] ●Minimum quantization value for AI / ML computing power: The minimum quantization value used to indicate the remaining computing power.

[0171] ● Maximum quantization value for AI / ML computing power: The maximum quantization value used to indicate the remaining computing power.

[0172] ●Disable Timer: The duration of the timer used to disable UAI triggering during its operation.

[0173] In one example implementation of triggering UAI, at least one of the following conditions may need to be met:

[0174] ● The remaining quantization value for computing power used in AI / ML is less than or equal to the pre-configured minimum quantization value.

[0175] ● No UAI has been sent since the remaining quantization value of computing power changed to be less than the pre-configured threshold.

[0176] ● The remaining quantization value for AI / ML computing power is greater than or equal to the pre-configured maximum quantization value.

[0177] ● No UAI has been sent since the remaining quantization value of computing power changed to be greater than or equal to the pre-configured threshold.

[0178] ●Prevent timers from running.

[0179] ● In one example implementation, the content of UAI may include the remaining quantized value of the computing power of AI / ML on the UE side.

[0180] In one example implementation of sending UAI, a timer can be started / restarted to disable it.

[0181] UE-side AI / ML LCM – Identification and Reporting of Applicable AI / ML Features / Functions / Models

[0182] Back Figure 5 Step 2 can be performed to determine the applicable UE AI / ML features / functions / models among the supported UE AI / ML features / functions / models. Such a process can be implemented so that the NW can further determine the applicable UE AI / ML features / functions / models based on network conditions and / or the available AI / ML features / functions / models actually stored at the UE (rather than all supported UE AI / ML features / functions / models). This process typically involves the NW sending a request to the UE to provide a set of additional conditions (e.g., network conditions) and / or a report of applicable UE AI / ML features / functions / models, the UE determining a set of applicable UE AI / ML features / functions / models based on the additional conditions and / or the available AI / ML features / functions / models at the UE, and reporting the determined set of applicable UE AI / ML features / functions / models to the NW.

[0183] Figure 5 Step 2 can be implemented using two example alternative solutions for messaging, which are in Figure 5 The instructions are for Solution 1 and Solution 2.

[0184] In the first solution, for example, messages(s) used to process interactions between the UE and the NW to determine and report applicable UE AI / ML features / functions / models can be based on signaling or message formats, including but not limited to: (1) RRC messages for NW transmission conditions, and / or (2) message response RRC messages for the UE to send reports applicable UE AI / ML features / functions / models in response to conditions or requests sent by the NW, such as... Figure 5 As shown in Solutions 1-1 and 1-2. In the second alternative solution, for example, the messages(s) used to process the interaction between the UE and NW to determine and report applicable UE AI / ML features / functions / models(s) can be based on RRC reconfiguration messages and corresponding RRC reconfiguration response messages, such as... Figure 5 The solutions shown are 2-1 and 2-2.

[0185] In some example implementations of the conditional transmission in the first solution, the conditions may be sent by the NW to the UE via a message format, which includes, but is not limited to: (1) an RRC system information message for transmitting the conditions and / or an application feature / function / model report requested by the NW, and / or (2) a message called Request Applicability Reporting tell) The message is used to send conditions from NW to UE.

[0186] In some example implementations, the aforementioned RRC system information message, when providing additional conditions for the UE to determine the applicable UE AI / M features / functions / models, may include or indicate at least one of the following information or information items:

[0187] ● AI / ML-based Feature Group Indicator: Used to indicate whether NW additional conditions are provided for the AI / ML-based feature group or to request applicable feature / feature / model reports for the AI / ML-based feature group. In one example implementation, the AI / ML-based feature group can be hard-coded. For example, AI / ML-based features can be represented by some ID. By way of example only, an AI / ML-based feature with Id=0 can indicate an AI / ML-based beam management feature group. An AI / ML-based feature with Id=1 can indicate an AI / ML-based CSI feature group, etc.

[0188] ● In one implementation, AI / ML-based features can be indicated by an enumeration type parameter with example values ​​of {beam management, CSI, CSI prediction, spare 1, ...}.

[0189] ●AI / ML-based feature indication: Used to indicate AI / ML-based features for which NW additional conditions are provided or for which applicable features / model reports are requested: In one implementation, AI / ML-based features can be hard-coded. For example, AI / ML-based features can be represented by some IDs. As an example only, an AI / ML-based feature with Id=0 can indicate AI / ML-based spatial beam management, and an AI / ML-based feature with Id=1 can indicate AI / ML-based temporal beam management, etc. In one implementation, AI / ML-based features can be indicated by an enumeration type parameter with example values ​​of {spatial beam management, temporal beam management, CSI feedback compression / decompression, CSI prediction, spare 1, ...}.

[0190] ● An indication of NW additional conditions, or an indication of NW additional conditions associated with the indicated AI / ML-based feature or feature group, used to indicate NW additional conditions regarding the AI / ML-based feature / feature group. At least one of the following information items may be indicated as NW additional conditions in the system message: o AI / ML-based feature indication: Used to indicate the AI / ML-based feature targeted by the message or NW appender. o ISD is used to indicate cell range, for example, UMa, UMi, 200 meters, 500 meters. o Antenna height, used to indicate the antenna height of the cell's base station. o NLOS probability: Used to indicate the probability of NLOS wireless propagation. o Indoor / Outdoor indicator, used to indicate whether the current scene is indoors or outdoors, or the indoor / outdoor ratio. o DL Tx beamcodebook indicator: Used to indicate the DL TX beamcodebook used for AI / ML model training. o gNB antenna array dimension indication. The downtilt angle of the o gNB antenna. o Beam pattern indicator for set B, used to indicate the beam pattern of set B compared to set A (see above). For example, set B could be a 1 / 4 subset of set A, etc. o Beam pattern for set A of AI / ML-based functions, used to indicate the beam pattern of set A compared to the full beam set (see above). For example, set A could be a 1 / 4 subset of the full set, etc.

[0191] Similarly, in utilizing RequestApplicationReporting In some example implementations of the first solution described above for message sending conditions, an example message can be sent from the NW to the UE. RequestApplicationReporting The message indicates / includes at least one of the following information or information items:

[0192] ● AI / ML Support Indicator: Used to indicate AI / ML-based features / FG on the UE side supported by NW.

[0193] ●AI / ML Feature Group Indication: Used to indicate an AI / ML feature group (e.g., AI / ML-based beam management, AI / ML-based CSI, AI / ML-based positioning), which can provide NW additional conditions for the AI / ML-based feature group or request that the AI / ML feature belong to the AI / ML-based feature group.

[0194] ● AI / ML-based feature indication: Used to indicate whether to provide NW additional conditions for AI / ML-based features or to request applicable features / model reports for AI / ML-based features. In one example implementation, AI / ML-based features / feature groups can be indicated via AI / ML-based feature indexes / identifiers that are already indicated in the UE capabilities. In one example implementation, AI / ML-based features / feature groups can be indicated via AI / ML-based feature indexes / identifiers according to the order of entries in the supported AI / ML feature list within the UE capability. For example, AI / ML-based feature / FG index = 0 indicates the first entry in the supported AI / ML-based feature / FG list. AI / ML-based feature / FG index = 1 indicates the second entry in the supported AI / ML-based feature / FG list, and so on. In one example implementation, AI / ML-based features / feature groups can be indicated by an enumeration type parameter with example values ​​of {spatial beam management, temporal beam management, CSI feedback compression / decompression, CSI prediction, spare 1, ...}. In one example implementation, AI / ML-based features / feature groups can be indicated via a bit string type parameter according to the order of entries in the supported AI / ML feature list in the UE capability. For example, the leftmost or rightmost bit in the bit string represents the first entry in the supported AI / ML-based feature / FG list, the second leftmost or rightmost bit in the bit string represents the second entry in the supported AI / ML-based feature / FG list, and so on. If the corresponding bit is set to 1, an applicable report corresponding to the AI / ML-based feature / FG is requested.

[0195] ● The NW additional conditions associated with the indicated AI / ML-based feature or feature group are used to indicate NW additional conditions regarding that AI / ML-based feature / feature group. (This can be found in...) RequestApplicationReporting The message indicates at least one of the following information as an additional condition for NW: o AI / ML-based feature indication: Used to indicate the AI / ML-based features targeted by the message or NW appended conditions. o ISD is used to indicate cell range, for example, UMa, UMi, 200 meters, 500 meters. o Antenna height, used to indicate the antenna height of the cell's base station. o NLOS probability, used to indicate the probability of NLOS wireless propagation. o Indoor / Outdoor indicator, used to indicate whether the current scene is indoors or outdoors, or the indoor / outdoor ratio. o DL Tx beamcodebook indicator, used to indicate the DL TX beamcodebook used for AI / ML model training. o gNB antenna array dimension indication. The downtilt angle of the o gNB antenna. o Beam pattern indicator for set B, used to indicate the beam pattern of set B compared to set A (see above). For example, set B could be a 1 / 4 subset of set A, etc. o A beam pattern for a set A of AI / ML-based functions, used to indicate the beam pattern of set A compared to the full beam set. For example, set A could be a 1 / 4 subset of the full set, etc.

[0196] In some example implementations, in Figure 5 In Solutions 1 and 2, the UE reports applicable UE features / functions / models to the NW via at least one of the following UL RRC message formats or MAC protocol signaling:

[0197] ●Option 1: UE Assistant Information (UAI) message.

[0198] ●Option 2: Dedicated RRC message, referred to as UE Applicable Functionality Reporting Report) .

[0199] ●Option 3: UL MAC CE.

[0200] For option 1 above, for example, a UAI message used to report applicable functions / models based on AI / ML features / feature groups or to report applicable features, it can be triggered by meeting some predefined conditions, which include, but are not limited to, at least one or more of the following example triggering conditions:

[0201] ● Triggering condition 1: RRC configuration related to the applicable function / model report for AI / ML-based features / feature groups has been configured, and / or the applicable function / model for this AI / ML-based feature / feature group has never been sent before.

[0202] ● Triggering condition 2: At least one AI / ML function / model for this AI / ML-based feature / feature group has been activated;

[0203] ● Triggering Condition 3: Since the last UE applicable function report, the AI / ML-based features / functions / models stored in the UE for the configured AI / ML-based features / feature groups have changed. In one example implementation of changes to the AI / ML-based features / functions / models stored in the UE, this may include adding, removing, or modifying AI / ML-based features / functions / models.

[0204] ● Triggering condition 4: An SIB has been received, indicating that an applicable feature / function / model associated with an AI / ML-based feature group is supported or has been requested by an NW, and no UAI related to an applicable AI / ML feature / function / model report has been sent previously.

[0205] Furthermore, regarding option 1, in some example implementations, the UAI message triggered and sent to the NW may contain various information about applicable functions / models for AI / ML-based features, and / or may contain various information about applicable AI / ML-based features. In some example implementations, the applicable functions / models for AI / ML-based features or the applicable AI / ML features for a group of AI / ML-based features may be a subset or the entire set of supported applicable features / functions / models reported in the UE capabilities. In some example implementations, the applicable functions / models for AI / ML-based features or the applicable AI / ML features for a group of AI / ML-based features may be additional content beyond the supported applicable features / functions / models reported in the UE capabilities.

[0206] When a UAI message is triggered by a modification or update of an AI / ML-based feature / function / model, the UAI message may contain at least one of the following pieces of information regarding the NW additional conditions of the modified / updated AI / ML-based feature / function / model:

[0207] ● AI / ML-based feature indication: Used to indicate the AI / ML-based features targeted by the message or NW appended conditions.

[0208] ●ISD is used to indicate cell range, for example, UMa, UMi, 200 meters, 500 meters.

[0209] ● Antenna height, used to indicate the antenna height of the cell's base station.

[0210] ●NLOS probability, used to indicate the probability of NLOS wireless propagation.

[0211] ●Indoor / Outdoor indicator, used to indicate whether the current scene is indoors or outdoors, or the indoor / outdoor ratio.

[0212] ●DL Tx beamcodebook indicator, used to indicate the DL TX beamcodebook used for AI / ML model training.

[0213] ●gNB antenna array dimension indication.

[0214] ● Downtilt angle of the gNB antenna.

[0215] ● Beam pattern indicator for set B, used to indicate the beam pattern of set B compared to set A (see above). For example, set B could be a 1 / 4 subset of set A, etc.

[0216] ● Beam pattern for set A of AI / ML-based functions, used to indicate the beam pattern of set A compared to the full beam set. For example, set A could be a 1 / 4 subset of the full set, etc.

[0217] When a UAI message is triggered by the addition of an AI / ML-based feature / function / model, the UAI message may contain at least one of the following pieces of information regarding the NW additional conditions of the newly added AI / ML-based feature / function / model:

[0218] ● AI / ML-based feature indication: Used to indicate the AI / ML-based features targeted by the message or NW appended conditions.

[0219] ●ISD is used to indicate cell range, for example, UMa, UMi, 200 meters, 500 meters.

[0220] ● Antenna height, used to indicate the antenna height of the cell's base station.

[0221] ●NLOS probability, used to indicate the probability of NLOS wireless propagation.

[0222] ●Indoor / Outdoor indicator, used to indicate whether the current scene is indoors or outdoors, or the indoor / outdoor ratio.

[0223] ●DL Tx beamcodebook indicator, used to indicate the DL TX beamcodebook used for AI / ML model training.

[0224] ●gNB antenna array dimension indication.

[0225] ● Downtilt angle of the gNB antenna.

[0226] ● Beam pattern indicator for set B, used to indicate the beam pattern of set B compared to set A (see above). For example, set B could be a 1 / 4 subset of set A, etc.

[0227] ● Beam pattern for set A of AI / ML-based functions, used to indicate the beam pattern of set A compared to the full beam set. For example, set A could be a 1 / 4 subset of the full set, etc.

[0228] For option 2 (dedicated RRC message) above, example UEApplicableFunctionalityReporting(UE (Applicable Function Report) The message can be related to the above. RequestApplicableFunctionalityThe response message contains NW additional conditions for at least one AI / ML-based feature / feature group and / or requests for an applicable AI / ML-based feature / function / model report for at least one AI / ML-based feature / feature group. Example UEApplicableFunctio nalityReporting The message may, for example, indicate / include applicable AI / ML features / models for AI / ML-based features or applicable AI / ML feature groups. In some example implementations, the applicable features / models for AI / ML-based features or applicable AI / ML feature groups are a subset or the entire set of supported applicable features / features / models reported in the UE capabilities. In some example implementations, the applicable features / models for AI / ML-based features or applicable AI / ML feature groups are additional content beyond the supported applicable features / features / models reported in the UE capabilities.

[0229] exist UEApplicableFunctionalityReporting In cases where AI / ML features / functions / models have been updated or added compared to those supported in the (latest) reported UE capabilities, UEApplicabl eFunctionalityReporting It may include at least one of the following information regarding the NW additional conditions for the modified / updated / added AI / ML features / functions / models:

[0230] o AI / ML-based feature indication: Used to indicate the AI / ML-based features targeted by the message or NW appended conditions.

[0231] o ISD is used to indicate cell range, for example, UMa, UMi, 200 meters, 500 meters.

[0232] o Antenna height, used to indicate the antenna height of the cell's base station.

[0233] o NLOS probability, used to indicate the probability of NLOS wireless propagation.

[0234] o Indoor / Outdoor indicator, used to indicate whether the current scene is indoors or outdoors, or the indoor / outdoor ratio.

[0235] o DL Tx beamcodebook indicator, used to indicate the DL TX beamcodebook used for AI / ML model training.

[0236] o gNB antenna array dimension indication.

[0237] The downtilt angle of the o gNB antenna.

[0238] o Beam pattern indicator for set B, used to indicate the beam pattern of set B compared to set A (see above). For example, set B could be a 1 / 4 subset of set A, uniformly distributed, etc.

[0239] o Beam pattern for set A of AI / ML-based functions, used to indicate the beam pattern of set A compared to the full beam set (see above). For example, set A could be a 1 / 4 subset of the full set, uniformly distributed, etc.

[0240] In some example implementations, for both Options 1 and 2 above, the applicable AI / ML function / model for AI / ML-based features / feature groups can be reported / indicated via a BIT STRING type parameter (or bitmap type parameter), or the applicable AI / ML feature can be reported / indicated via a BIT STRING type parameter (or bitmap type parameter). As an example syntax, the bit string type parameter indicating the applicable AI / ML function for space beam management features is shown below: AI-ML-SpatialBeamManagement BIT STRING (SIZE (1..maxNrofFunctionalities / Models)) In such example bit strings or bitmaps, the first bit from the left (or right) can be mapped to the first entry in the list of functions reported by the UE capabilities for AI / ML-based features / feature groups (e.g., AI / ML-based spatial beam management), the second bit from the left (or right) can be mapped to the second entry in the list of functions reported by the UE capabilities for AI / ML-based features / FGs (e.g., AI / ML-based spatial beam management), and so on. If a specific bit in the report is set to 1, the AI / ML function mapped to that specific bit can be considered applicable. Otherwise, the AI / ML function mapped to and associated with that specific bit may not be considered applicable.

[0241] As another example syntax for applicable AI / ML-based features, the bit string type parameter indicating the applicable AI / ML features for beam management is as follows: AI-ML-BeamManagement BIT STRING (SIZE (2)) In such examples, a value of AI-ML-BeamManagement = 01 indicates that the features of AI / ML spatial beam management and AI / ML temporal beam management are applicable. A value of AI-ML-BeamManagement = 10 indicates that the features of AI / ML temporal beam management and AI / ML spatial beam management are applicable. A value of AI-ML-BeamManagement = 00 indicates that neither AI / ML temporal beam management nor AI / ML spatial beam management is applicable. A value of AI-ML-BeamManagement = 11 indicates that both AI / ML temporal beam management and AI / ML spatial beam management are applicable.

[0242] In some example implementations, for both Option 1 and Option 2 above, the applicable AI / ML functions / models for AI / ML-based features / feature groups can be reported / indicated via a list of AI / ML function indexes / identifiers or AI / ML model indexes / identifiers. As an example syntax, the list of integer-type parameters indicating the applicability of AI / ML-based spatial beam management is shown below: ApplicableFuntionalities SEQUENCE(SIZE(0..maxNrofAIMLfunctionalities) OF FunctionalityId Or ApplicableModels SEQUENCE(SIZE (0..maxNrofAIMLModels) OF ModelId In such example AI / ML feature / model indexes or identifiers, the feature or model indexes may be numbered according to the order of entries in the AI / ML feature list or model list for an AI / ML-based feature or function within the UE capabilities. For example, the first entry in the list is numbered with index=0, the second entry with index=1, and so on.

[0243] In such example AI / ML function / model indexes or identifiers, the function or model index / identifier can be aligned with the function or model index / identifier reported in the UE capabilities.

[0244] In some example implementations, for both Option 1 and Option 2 above, the applicable AI / ML-based features can be reported / indicated via an AI / ML-based feature list. As an example syntax, the list parameters indicating the applicability of AI / ML-based beam management are shown below: ApplicableFeaturesForBeamManagement SEQUENCE(SIZE (0..maxNrofAI / MLfeaturesForBM) OF FeatureId In such example AI / ML feature indexes or identifiers, the feature indexes or identifiers may be numbered sequentially according to the order of entries in the AI / ML feature list for an AI / ML-based feature group within the UE capability. For example, the first entry in the list is numbered with index=0, the second entry with index=1, and so on. In such example AI / ML feature indexes or identifiers, the features may be aligned with the AI / ML-based feature indexes / identifiers reported in the UE capability for the AI / ML-based feature group.

[0245] For option 3 above, a UL MAC CE (e.g., an Applicable AI / ML Reporting MAC CE) can be triggered for reporting applicable UE AI / ML features / functions / models. In some example implementations, such an Applicable Function Reporting MAC CE can be triggered by any one or more of the following conditions, similar to options 1 and 2:

[0246] ● Triggering condition 1: RRC configuration has been configured for applicable feature / function / model reports for AI / ML-based feature groups / features / functions, and applicable feature / function / model reports have never been sent before.

[0247] ● Triggering condition 2: At least one AI / ML feature / function / model has been activated for an AI / ML-based feature group / feature / function;

[0248] ● Triggering Condition 3: Since the last UE applicable feature / function / model report was sent, the AI / ML-based feature / function / model stored in the UE for the feature group / feature / function has been changed. In one example implementation of the change of the AI / ML-based feature / function / model stored in the UE, this may include adding, removing, or modifying the AI / ML-based feature / function / model.

[0249] In some example implementations, the aforementioned UL MAC CE may trigger a scheduling request (SR) if no available UL authorization for sending such a UL MAC CE has been received. In some example implementations, only one SR configuration may be applied in a cell group for applicable function reporting.

[0250] In some example implementations, the UL MAC CE described above may include at least one of the following information or information items:

[0251] ●Serving Cell Indicator: Used to indicate the serving cell of an application (one or more) based on AI / ML features.

[0252] ● AI / ML-based feature group indication: One or more AI / ML-based feature groups used to indicate the applicable AI / ML features / functions / models of interest.

[0253] ● AI / ML-based feature indication: One or more AI / ML-based features used to indicate the applicable AI / ML features / functions / models of interest.

[0254] ●AI / ML Function Indicator: Used to indicate one or more AI / ML-based functions of the applicable AI / ML model of interest.

[0255] ● AI / ML Applicable Feature / Function / Model Indication: Used to indicate the currently applicable AI / ML feature / function / model at the UE for (one or more) indicated AI / ML-based features.

[0256] In some example implementations, the above UL MAC CE may comply with Figure 7 The example structure / format is shown below. Figure 7 In the first octet of UL MAC CE, the "F" i "Indicates whether (or applicable) there exists a Fu containing the i-th AI / ML-based feature." i,j An octet. If Fi = 1, then it contains Fu. i,j The octet exists; otherwise, it contains Fu. i,j The octet does not exist. Therefore, the number n of the octets of Fu will be equal to the number of F octets in the first octet. i The number of 1s in the middle.

[0257] Furthermore, in Figure 7 In the example, “Fu i,j "The octet indicates the applicable functions at the UE for AI / ML-based features, which are related to "F" i (If indicated to exist) corresponds to "Fu" i,j Each bit in “” corresponds to and indicates the UE capability (e.g., Figure 5 In the list of supported functions reported in Step 1 (UE capabilities), is the j-th AI / ML function of this AI / ML feature applicable? If "Fu i,jIf set to 1, the j-th AI / ML feature in the AI / ML-based features is applicable. Otherwise, the AI / ML feature is not applicable.

[0258] Back Figure 5 Step 2 utilizes the RRC reconfiguration mechanism to determine and report a second alternative solution (Solution 2) for the AI / ML features / functions / models available to the UE. The RRC reconfiguration process may include sending data from the NW to the UE. RRCReconfiguration (RRC reconfiguration) message ( Figure 5 Solution 2-1) provides the above conditions, and the UE sends a response message RRCReconfigurationComplete (RRC reconfiguration complete) to the NW. Figure 5 Solution 2-2) reports applicable AI / ML features / functions / models.

[0259] Specifically, NW can be accessed via RRCReconfiguration The message requests the UE to report applicable AI / ML functions for AI / ML-based features / feature groups, and the UE can accordingly... RRCReconfigurationComplete The message responds to NW for applicable AI / ML functions based on AI / ML features / feature groups.

[0260] In requesting applicable UE AI / ML functions / models for AI / ML-based features. RRCReconfiguration In some example implementations of the message, parameters may be introduced or included, such as an ENUMERATED type parameter for each AI / ML-based feature / feature group, as exemplified below: RequestForAI-ML-SpatialBeamManagement ENUMERATED{TRUE} OPTIONAL --Need N RequestForAI-ML-TemporalBeamManagement ENUMERATED{TRUE} OPTIONAL --Need N RequestForAI-ML-CSI-RS-Compression ENUMERATED{TRUE} OPTIONAL, -- NeedN RequestForAI-ML-CSI-RS-Prediction ENUMERATED{TRUE} OPTIONAL, -- NeedN In requesting applicable UE AI / ML features / functions / models based on AI / ML feature groups RRCReconfigurationIn some example implementations of the message, parameters may be introduced or included, such as an ENUMERATED type parameter for each AI / ML-based feature / feature group, as exemplified below: RequestForAI-ML-BeamManagement ENUMERATED{TRUE} OPTIONAL -- Need N RequestForAI-ML-CSI ENUMERATED{TRUE} OPTIONAL, -- Need N In one example implementation of the above parameters, these parameters can be set per UE, per cell group, per frequency band, or per cell.

[0261] In the request for applicable features / functions / models RRCReconfiguration In an example implementation of the message, one or more information elements may be introduced for requesting a report of applicable AI / ML features / functions / models, which may include at least one of the following information items regarding additional NW conditions to be considered by the UE when reporting AI / ML features / functions / models:

[0262] ● AI / ML-based feature indication: Used to indicate the AI / ML-based features targeted by the message or NW appended conditions.

[0263] ●ISD is used to indicate cell range, for example, UMa, UMi, 200 meters, 500 meters.

[0264] ● Antenna height, used to indicate the antenna height of the cell's base station.

[0265] ●NLOS Probability: Used to indicate the probability of NLOS wireless propagation.

[0266] ●Indoor / Outdoor indicator, used to indicate whether the current scene is indoors or outdoors, or the indoor / outdoor ratio.

[0267] ●DL Tx beamcodebook indicator: Used to indicate the DL TX beamcodebook used for AI / ML model training.

[0268] ●gNB antenna array dimension indication.

[0269] ● Downtilt angle of the gNB antenna.

[0270] ● Beam pattern indicator for set B, used to indicate the beam pattern of set B compared to set A (see above). For example, set B could be a 1 / 4 subset of set A, etc.

[0271] ● Beam pattern for set A of AI / ML-based functions, used to indicate the beam pattern of set A compared to the full beam set (see above). For example, set A could be a 1 / 4 subset of the full set, etc.

[0272] In the UE response to the above RRCReconfiguration The request in RRCReconfigurationComplete In some example implementations of the included parameters(s), the applicable AI / ML-based features / functions / models for AI / ML-based feature groups / features can be indicated via example BIT STRING or bitmap type parameters. In such example bit strings or bitmaps, the first bit from the left (or right) can be mapped to the first entry in the list of features / functions / models (e.g., AI / ML-based beam management or AI / ML-based spatial beam management) reported by the UE capability, the second bit from the left (or right) can be mapped to the second entry in the list of features / functions / models (e.g., AI / ML-based beam management or AI / ML-based spatial beam management) reported by the UE capability, and so on. If a specific bit is set to 1 in the report, the AI / ML feature / function / model mapped to that specific bit can be considered applicable. Otherwise, the AI / ML feature / function / model mapped to and associated with that specific bit can not be considered applicable. Example bit strings are shown below: AI-ML-SpatialBeamManagement BIT STRING (SIZE(1..maxNrofFunctionalities)) OPTIONAL In one example implementation of the above applicable AI / ML features / functions / model parameters, these parameters can be set per UE, per cell group, per frequency band, or per cell.

[0273] UE-side AI / ML LCM – Testing applicable to UE AI / ML features / functions / models

[0274] Go to Figure 5 Step 3 involves preparation for the activation of AI / ML features / functions / models that may occur after the UE reports the applicable AI / ML functions in Step 2. Testing of the applicable AI / ML features / functions / models may be performed before further consideration for adoption and activation. For example, the conditions provided by the NW to the UE in Step 2 may be restricted / constrained to minimize the exposure of sensitive NW-related conditions (e.g., conditions that expose sensitive user information should not be disseminated due to potential risks of analysis). The determination of the available AI / ML functions reported by the UE in Step 2 may be performed under the restricted / constrained conditions provided by the NW. Therefore, in Figure 5 In step 3, it may be desirable for the UE and NW to test the applicable AI / ML functions reported by the UE to determine their predictive performance / accuracy, thereby further determining whether they can be adopted / activated for prediction, for example, gNB setup, deployment scenarios, etc.

[0275] Figure 8 The document illustrates an example of such a test procedure for UE-side AI / ML functions by UE 802 and NW 804, which includes the following example steps:

[0276] ●Step 3-1: NW 804 can send message A to test one or more applicable AI / ML features / functions / models.

[0277] ●Step 3-2: UE 802 can input the list of reference data received from message A into one or more applicable AI / ML features / functions / models to obtain the corresponding inference output or value.

[0278] ●Step 3-3: UE 802 can send the inferred output or value to NW 804 via message B, so that NW can compare the inferred output or value with the true value known to NW 804.

[0279] In some example implementations of step 3-1 above, message A can be implemented as a DL RRC message, DL MACCE, or PDCCH signaling. In another example implementation, message A can be protocol signaling terminating between the AI / ML logical layer and the NW logical entity / unit.

[0280] Furthermore, in some example implementations of step 3-1 above, message A may include / indicate at least one of the following information or information items:

[0281] ● Applicable AI / ML Features / Functions / Models Indicator: Used to indicate the applicable AI / ML features / functions / models that need to be tested. In some example implementations, the applicable AI / ML features / functions / models may be derived from the above. Figure 5 The applicable AI / ML report in step 2.

[0282] ●Supported AI / ML Features / Functions / Models Indicator: This indicates the supported AI / ML features / functions / models that need to be tested. In some example implementations, the supported AI / ML features / functions / models may be from the above. Figure 5 The (latest) reported UE capabilities in step 1.

[0283] ● Reference Data List (Input Data List for AI / ML Functions to be Tested): A data list used as input data for the AI / ML features / functions / models that need to be tested.

[0284] ● Reference Data List (Benchmark List): A list of data used as benchmarks to evaluate the output data of AI / ML features / functions / models that need to be tested.

[0285] In some example implementations of steps 3-2 above, the UE may use one or more applicable AI / ML features / functions / models to process the input reference data list and / or benchmark reference data list received from message A to obtain corresponding inferred output values ​​or performance metrics. These inferred output values ​​may be used by the NW to check the performance and effectiveness of these AI / ML features / functions / models; or these performance metrics may be used by the UE to check the performance and effectiveness of these AI / ML features / functions / models.

[0286] In some example implementations of steps 3-3 above, the protocol format of message B can be implemented as a UL RRC message (e.g., UAI), UL MAC CE, or via PUCCH signaling.

[0287] Furthermore, in some example implementations of steps 3-3 above, message B may contain / indicate at least one of the following information or information items:

[0288] ● Applicable AI / ML Features / Functions / Model Indicators: Used to indicate the applicable AI / ML features / functions / models that have been tested.

[0289] ● Output Data List: Used to include output data obtained from inferences from each applicable AI / ML feature / function / model.

[0290] ●UE Additional Conditions: Used to include additional UE conditions to assist NW in determining the validity or performance of the tested AI / ML features / functions / models.

[0291] ●Available AI / ML Features / Functions / Model Indicators: Used to indicate valid AI / ML features / functions / models that have been tested.

[0292] ● KPI (Key Performance Indicator) value indication: Used to indicate the KPI value for each AI / ML feature / function / model. In some example implementations, the KPI value is used to indicate the KPI value for each AI / ML feature / function / model considered available.

[0293] In some example implementations of an alternative approach to UE AI / ML feature / function / model testing, the following steps may be performed:

[0294] ●Step 1: The NW can configure RRC signaling for the UE to apply AI / ML features / functions / models on the UE side.

[0295] ●Step 2: The UE can send a message to report the available AI / ML features / functions / models to the NW.

[0296] In some example implementations of the RRC signaling in step 1, the RRC signaling may include an RRC configuration, which may be one or more CSI-ResourceConfigs for one or more applicable AI / ML features / functions / models reported in step 2 (e.g., applicable AI / ML feature / model reports). The RRC configuration may also be one or more CSI-ResourceConfigs for one or more supported AI / ML features / functions / models reported in step 1 (e.g., UE capabilities). In some example implementations, the CSI-ResourceConfig may be a configuration of a reference signal used by the UE to perform measurements to obtain inputs to the AI / ML function / model and / or to obtain a benchmark value (e.g., measurement result) for calculating performance metrics by comparing actual measurement results with inferred values ​​(e.g., the output of the tested AI / ML function / model).

[0297] In some example implementations of the RRC signaling in step 1, the RRC signaling may include RRC configurations, which may be one or more RRC configurations used to indicate the AI / ML features / functions / models to be tested. In some example implementations, the AI / ML features / functions / models to be tested may be those displayed by the UE in... Figure 5 Step 2 reports one of the multiple applicable AI / ML features / features / models. In some example implementations, the AI / ML feature / feature / model to be tested may be... Figure 5 Step 1 is one of the supported AI / ML features / functions / models in the UE capabilities.

[0298] In some example implementations, the above message may be a UL RRC signaling, a UL MAC CE, or a UL RRC message / UL MAC CE. It may include / indicate at least one of the following information:

[0299] ● Available AI / ML-based feature indicators.

[0300] ● Available AI / ML-based feature indicators.

[0301] ● Available AI / ML-based model indicators.

[0302] ●Performance KPI values ​​for each available AI / ML-based feature / function / model.

[0303] UE-side AI / ML LCM – UE AI / ML configuration

[0304] Now go to Figure 5 In step 4, the NW can now determine the configuration for the AI / ML functions based on the available UE AI / ML functions, test output, and / or the additional UE conditions received from the UE in step 3, and further transmit such configuration to the UE. For example, the NW can determine which of the applicable AI / ML models under test provide acceptable prediction accuracy or performance and their configurations.

[0305] In one example implementation of step 4 above, the configuration for the UE AI / ML function determined by the NW can be sent to the UE via various alternative RRC configuration structures.

[0306] In the first alternative or alternative example, it is possible to construct as follows: Figures 9A to 9E The RRC configuration structure shown is used for UE AI / ML related configurations and is sent from the NW to the UE. For example... Figures 9A to 9E As shown in the example:

[0307] ●1: AL / ML (AI / ML) related configurations (“ALMLConfig”) can be configured in the serving cell configuration (“ServingCellConfig”), such as Figure 9A As shown. In another example implementation, AI / ML related configurations (“AI / MLConfig”) can be configured in CellGroupConfig, such as... Figure 9B As shown, or in PhysicalCellGroupConfig (physical cell group configuration), such as Figure 9C As shown, or in MAC-CellGroupConfig (MAC-cell group configuration), such as Figure 9D As shown. In another implementation, the AI / ML (AI / ML) related configuration (“AIMLConfig”) can be configured per UE, that is, the AIMLConfig and IE CellGroupConfig are at the same level, such as... Figure 9E As shown.

[0308] ●2: AI / ML related configurations (such as AIMLConfig) may contain a list of configurations based on AI / ML features (e.g., “AIMLBasedFeatureConfig#1 (AIML-based feature configuration #1)” to “AIMLBasedFeatureConfig#4 (AIML-based feature configuration #4)”).

[0309] ●3: Each AI / ML-based feature configuration may include a list of AI / ML function configurations (e.g., “AIMLFunctionalityAddModList (AIML function add or modify list)”, including “AIMLFunctionalityConfig#1 (AIML function configuration #1)” to “AIMLFunctionalityConfig#3 (AIML function configuration #3)”) and / or an index / identifier of the AI / ML-based feature configuration, and / or an indication of the AI / ML-based feature for such configuration.

[0310] ●4: AI / ML feature configuration may include at least one of the following information items: o Index / identifier of AI / ML function or AI / ML function ID (e.g., "AIMLFunctionalityID (AIML Function ID)"). o The list of AI / ML models for this AI / ML function (e.g., "AIMLModelToAddModList (AIML Models to be Added or Modified)"). Each AI / ML model in this list may include a model ID to represent that AI / ML model, and / or a quantified value of the computing power consumed when the corresponding AI / ML model is activated. o Inference-related configurations (list) (e.g., "InferenceConfig (Inference Configuration)" or "InferenceConfigToAddModList (Inference Configurations to be Added or Modified)"). o Performance monitoring related configurations (list) (e.g., "MonitoringConfig" or "MonitoringToAddModList"). o Traditional configuration used to enable or disable corresponding non-AI-based features. o The quantified value of computing power consumption when the corresponding AI / ML feature / function / model is activated.

[0311] Furthermore, in Figures 9A to 9EIn this example RRC configuration structure, all AI / ML functions / features / models can be configured together to facilitate future AI development. Adding new AI / ML functions / features / models and their configurations is easily implemented. In this example, AI / ML-related functions are configured per cell, per cell group, or per UE.

[0312] In some example implementations, AI / ML functions can be indexed to align the UE and NW's understanding of correctly identifying AI / ML functions / models. For example:

[0313] ● In some example implementations, the AI / ML function index / identifier can be based on UE capabilities (e.g., Figure 5 The location of the AI / ML function for the AI / ML-based feature in the supported AI / ML function list (as reported in step 1) is used to construct the configuration. For example, an AI / ML function index for the AI / ML-based feature in the RRC configuration of 0 indicates the first entry in the list of supported AI / ML functions for that AI / ML-based feature in the UE capability, an AI / ML function index for the AI / ML-based feature in the RRC configuration of 1 indicates the second entry in the list of supported AI / ML functions for that AI / ML-based feature in the UE capability, and so on.

[0314] ● In some example implementations, the AI / ML function represented by the AI / ML function index / identifier in the above RRC configuration can be associated with UE capabilities (such as...) Figure 5 It is built together with the same AI / ML feature index / identifier reported in step 1.

[0315] ● In some example implementations, the AI / ML function index / identifier for AI / ML functions in the RRC configuration can be based on the applicable function report (such as...). Figure 5 The AI / ML features are located in the list of applicable features reported in step 2 and built accordingly.

[0316] ●In some other implementations, the AI / ML function represented by the AI / ML function index / identifier in the RRC configuration can be associated with the applicable function report (such as...). Figure 5 The AI / ML feature index / identifier is the same as that reported in step 2. For example, AI / ML feature index = 0 for AI / ML-based features in the RRC configuration can indicate the first entry in the list of supporting AI / ML features for that AI / ML-based feature in the applicable AI / ML feature report, AI / ML feature index = 1 for AI / ML-based features in the RRC configuration can indicate the second entry in the list of supporting AI / ML features for that AI / ML-based feature in the AI / ML feature report, and so on.

[0317] In some example implementations, the above Figures 9A to 9E of Inference Configurations It may include at least one of the following information or information items:

[0318] ● Measurement configurations (list) used to obtain inference input data, such as MeasurementForInferenceConfig (Measurement for inference configuration) or MeasurementForInferenceCo nfigToAddmodList (Measurement used to infer the list of configurations to be added or modified) .

[0319] ● Output data, inference results, report configuration (list) (e.g.) Inference Reporting or inferenceReportingToAddModList (List of inference reports to be added or modified) For example, in one implementation, Inference Reporting may include instructions about the content to be reported. For example, in one implementation of reporting content information, for beam management, it may be at least one of the following information about the content:

[0320] o Based on the inference results, the first N DL beams, N>0;

[0321] o Based on the inferred results, the first N DL beams with corresponding RSRP values, where N>0;

[0322] o Based on the inference results, the top N DL beams with the best M DL beam RSRP values ​​are N > M >= 0.

[0323] ●In one implementation method Inference Reporting Instructions regarding report types may be included:

[0324] o Semi-persistent cycle

[0325] o cycle

[0326] o event triggered - Based on the inference, the TOP 1 beam has been changed. - Based on the inference results, the current DL TX beam is no longer among the first N beams. - Based on the inference results, the current DL TX beam has an RSRP value below the threshold. - Based on the inference results, the current DL TX beam has an RSRP value below the threshold, and at least one beam has an RSRP value above the threshold.

[0327] In some example implementations of performance monitoring-related configurations, it may include at least one of the following:

[0328] ● Measurement Configuration: Measurement configuration used for performance monitoring.

[0329] ● Measurement report configuration: Report configuration for performance measurement or performance result reporting.

[0330] ● In one implementation of a conventional related configuration, it may include at least one of the following:

[0331] ●CSI-ReportingConfigId: Indicates the CSI reporting configuration based on non-AI beam management. In one implementation, the reporting type configured in the CSI-ReportingConfig indicated by such CSI-ReportingConfigId is semi-persistent on PUSCH or semi-persistent on PUCCH, or non-periodic.

[0332] In the implementation of Option 1, it is assumed that there is an AI / ML model (list), each entry of InferenceToAddModList is associated with each entry of the AI / ML model (list) in ascending order, and / or each entry of MonitoringToAddModList is associated with each entry of the AI / ML model (list) in ascending order.

[0333] In a second optional or alternative example of the RRC configuration structure for AI / ML features / features / models, AI / ML-based features and / or AI / ML-based features / models can be associated with CSI-ReportConfig. The following shows an example CSI-ReportConfig structure that includes configurations for AI / ML-based features / features / models: CSI-ReportConfig { CSI-ReportConfigId; AIMLInformation: { AIML-FunctionalityId; AIML-FeatureId; AIML-ModelId; } MeasurementConfigForAIInference; MeasurementConfigForAIMonitoring; ReportQuantity Choice { None CRI-RI-PMI-CQI ... Inference-SSB-Index-RSRP Inference-Cri-RSRP Monitoring-SSB-Index-RSRP Monitoring-CRI-Index-RSRP } ReportConfigType { Periodic SemiPersistentOnPUSCH SemiPersistentOnPUCCH Aperiodic EventTriggerredForAIML } } Using the example RRC structure described above, AI / ML functions / features / models can be implicitly activated / deactivated through the activation / deactivation of semi-persistent CSI reports. In some example implementations of configuring AI / ML features / functions / models for the UE, AI / ML related information can be configured in the RRC configuration used for CSI reporting (e.g., CSI-ReportConfig), as shown above.

[0334] In some example implementations, the RRC configuration for AI / ML features / functions / models may include and / or indicate at least one of the following information or information items:

[0335] ●AI / ML Feature / Function / Model Indicator: Used to indicate the AI / ML feature / function / model corresponding to this CSI report configuration.

[0336] ● Measurement resource configuration for inference indication: This indicates the measurement resource configuration for AI / ML function / model inference inputs associated with this CSI report configuration. This indication can be via... CSI-ResourceConfigId (CSI-ResourceConfigId) (Set ID) of.

[0337] ● Measurement resource configuration for monitoring indications: This indicates the measurement resource configuration for monitoring AI / ML functions / model performance in relation to this CSI report configuration. The indication can be... CSI-ResourceConfigId In one example implementation, the measurement resource configuration used for monitoring indications may be mutually exclusive with the measurement resource configuration used for inferring indications.

[0338] ●AI / ML Report Indicator: Used to indicate reporting information for relevant AI / ML features / functions / models. In one example implementation, it may be included in the ReportQuantity element of the CSI reporting configuration.

[0339] ●AI / ML Report Type Indicator: Used to indicate the report type for the relevant AI / ML feature / function / model. In one example implementation, event-triggered type reports are introduced, details of which will be described below.

[0340] In some example implementations, for configuring CSI report settings for AI / ML-based features / features / models, the CSI report type can be configured as semi-persistent on PUCCH or semi-persistent on PUSCH, or if the CSI report configuration is related to the measurement configuration used for inference, the CSI report type is configured as an event-triggered CSI report. The CSI report type can also be configured as a semi-persistent CSI report on PUCCH or semi-persistent CSI report on PUSCH, or a non-periodic CSI report, or if the CSI report configuration is related to the measurement configuration used for monitoring, the CSI report type is configured as an event-triggered CSI report.

[0341] Now go to Figure 5 In step 5, the UE may perform, for example, activation and / or deactivation of one or more AI / ML features / functions / models indicated to the UE by the NW.

[0342] In one example implementation of step 5, when using Figures 9A to 9E In the example RRC configuration structure, AI / ML functionality can be activated via separate signaling from the NW to the UE. This signaling may contain at least one of the following information or information items:

[0343] ●Serving Cell Indicator: Used to indicate the serving cell for which AI / ML features / functions / models are activated / deactivated.

[0344] ● AI / ML-based feature indication: Used to indicate the AI / ML-based feature to which the AI / ML function belongs.

[0345] ●AI / ML Function Indicator: Used to indicate which AI / ML functions need to be activated / deactivated.

[0346] ●AI / ML Function Indicators: Used to indicate which AI / ML models need to be activated / deactivated.

[0347] ●Activation / Deactivation Indicator: Used to indicate the activation / deactivation of the indicated AI / ML feature / function / model.

[0348] Such signaling can be implemented, for example, as DL MAC CE. Figure 10 An example DL MAC CE for signaling activation / deactivation of AI / ML functions is shown.

[0349] exist Figure 10 In the example DL MAC CE, the "C" in the first octet... i "Indicates a serving cell with servingCellId = i, where (when Ci = 1) there exists an octet that the AI / ML function should be activated / deactivated. Ci can be 0 or 1. The 'F' in the second octet..." i This indicates how many octets will exist after the second octet. For example, if C i =1,F i =1 will mean targeting C i The serving cell represented by Ci will have two octets (e.g., for AI / ML-based spatial beam management and AI / ML-based temporal beam management), and Ci=1, Fi=0 will mean that there will be one octet for the serving cell represented by Ci. “Feat IDi,j(featureIDi,j)” indicates the AI / ML-based feature ID to which the j-th function to be activated / deactivated belongs for the serving cell with servingCellId=i. “Functionality IDi,j(functionIDi,j)” indicates the AI / ML function to be activated / deactivated for the serving cell with servingCellId=i. Figure 10 The R bit indicated in the text represents a reserved bit.

[0350] exist Figure 5 In another example implementation of step 5, when utilizing the example CSI-ReportConfig RRC structure, the signaling for activation / deactivation of AI / ML features / functions / models can be accomplished via activation / deactivation of a related semi-persistent CSI report, the configuration of which can be inferred (e.g., ReportQuantity set to...). Inference-SSB-Index-RSRP For example, DL MAC CE can be used with at least one of the following information or information items:

[0351] ●1: Serving Cell ID: Used to indicate the serving cell in the CSI report.

[0352] ●2: BWP ID: Used to indicate the UL BWP for reporting CSI reports.

[0353] ●3: CSI Report Configuration Instructions: Used to indicate the activation / deactivation of CSI report configurations for AI / ML features / functions / models.

[0354] UE-side AI / ML LCM – UE AI / ML activation / deactivation

[0355] Go to Figure 5 In step 6, the UE can perform AI / ML feature / function / model activation and / or deactivation based on the indication / signaling from the NW in step 5 above. In some example implementations, the UE can perform one or more of the following AI / ML feature / function / model activation operations when it receives DL MAC CE signaling, or in another example implementation, when it receives the protocol signaling in step 5, which terminates between the AI / ML logical layer at the UE and the AI / ML logical entity / unit at the NW:

[0356] ● Instruct lower layers about information regarding AI / ML function activation / deactivation of MAC CE.

[0357] ● Instruct lower layers on the use of traditional beam management associated with activated AI / ML functions. CSI- ReportConfig Deactivate.

[0358] In some example implementations, the UE may perform one or more of the following AI / ML functions to deactivate one or more of the operations:

[0359] ● Instruct lower layers about information regarding AI / ML function activation / deactivation of MAC CE.

[0360] ● Instruct lower layers on the use of traditional beam management related to the deactivated AI / ML functions. CSI- ReportConfig Activate.

[0361] UE-side AI / ML LCM – UE AI / ML Inference Report

[0362] Now go to Figure 5 Step 7: After the UE performs activation and / or deactivation of AI / ML functions and / or actual measurements, an inference and / or measurement report can be provided from the UE to the NW. Such inference / measurement reports may contain predictions or inferences and / or actual measurements by the UE based on the activated AI / ML functions and models, which the NW can use for network configuration and provision.

[0363] In some example implementations of step 7, when the associated AI / ML function / model is activated, the inference / measurement report can be implemented as PUCCH signaling or PUSCH signaling.

[0364] For example, for AI / ML-based space beam management, the inference / measurement report may include at least one of the following information or information items:

[0365] ●1: The first N beams Id with corresponding N reference signal received power (RSRP) values, where N>=1;

[0366] ●2: The first N beams Id of the M beams with RSRP values ​​out of N beams, where N>M>=0;

[0367] ●3: The first N beams with RSRP values ​​greater than the predefined RSRP threshold.

[0368] ●4: Flag indicating the RSRP value for a beam. If this flag is present, its value indicates whether the RSRP value of the corresponding beam with the beam ID is inferred via AI / ML functions / models or actually measured.

[0369] ● In addition, such example inference / measurement reports can be triggered by at least one of the following events:

[0370] ● When the UE determines, based on the inference results, that the current DL TX beam is not among the first N beams, and therefore needs to switch beams.

[0371] ● When the UE determines, based on the inference results, that the RSRP value of the current DL TX beam is lower than a predefined RSRP threshold, and there exists at least one DL TX beam with an RSRP value higher than a predefined RSRP threshold. These two predefined thresholds may be the same or different.

[0372] For example, in AI / ML-based temporal beam management, the inference / measurement report in step 7 above can be reported as a measurement result or inference result at each configured reporting time. Such reports may include and / or indicate one or more of the following information or information items:

[0373] ●N beam Ids with the best N RSRP values ​​in each inference report, where N>=1.

[0374] ●N beam Id with the best M RSRP values ​​in each inference report, where N>M>=1.

[0375] ● Indication: Used to indicate whether the report contains inferred results or actual measurement results.

[0376] NW-side AI / ML LCM – Overview

[0377] In some example implementations, AI / ML functions may reside on the NW side rather than the UE side. Regarding AI / ML LCM for NW-side AI / ML functions, while the NW may be able to control the selection, activation, and deactivation of AI / ML functions and models, it may still require UE assistance in providing parameters and information so that the NW can apply appropriate AI / ML functions. Such UE assistance information may be provided to the NW as one or more UE-additional condition reports. Figure 11 The document illustrates a general LCM procedure for NW-side AI / ML functions performed by UE 1102 and NW 1104, which includes the following example steps:

[0378] ●Step 1: NW 1104 can send a request message to UE 1102 for UE additional condition reporting.

[0379] ●Step 2: UE 1102 may determine whether to trigger a UE Additional Conditions Report based on one or more criteria. If one or more criteria are met, UE 1102 continues with step 3. Otherwise, the process ends.

[0380] ●Step 3: UE 1102 can then send a UE additional conditions report to NW 1104.

[0381] In some example implementations, the request message in step 1 of the UE Additional Conditions report (e.g., RequestUEAdditionalConditions) may be a DL RRC message (e.g., RRCReconfiguration). In another example implementation, the protocol signaling terminating between the AI / ML logical layer and the NW logical entity / unit may include and / or indicate at least one of the following information or information items:

[0382] ●UE speed. For example, UE speed can be provided as an enumeration type parameter with values ​​{15, 30, 60, 90, 120, etc.}. Alternatively, UE speed can be provided as an enumeration type parameter with values ​​{stationary, low speed, medium speed, high speed, etc.}, where each value represents a speed range. For example, the value "stationary" can represent a speed range of [0, 5 km / h), the value "low speed" can represent a speed range of [5 km / h, 15 km / h), the value "medium speed" can represent a speed range of [15 km / h, 60 km / h), and the value "high speed" can represent a speed range of [60 km / h, 120 km / h].

[0383] ●UE rotation, for example, the UE rotation can be an enumeration type parameter with a value of one of {30, 60, 90, 120, 180}.

[0384] In some example implementations, a disable timer (with a predefined initial timer value) may be introduced to control the reporting of additional conditions by the UE. For example, if the disable timer is still running, such reports can be disabled and thus cannot be triggered.

[0385] In some example implementations of step 2 above, at least one of the following conditions may need to be met to trigger a UE additional condition report:

[0386] ●The aforementioned timer that is disabled for UE additional condition reporting (if set or configured) is not running.

[0387] ●The UE's additional conditions have changed since the last UE additional conditions report. For example: Since the last report, the UE speed has been changed, for example, from stationary to low speed, and from low speed to high speed. Since the last report, the UE speed has changed beyond the configured speed range. For example, the UE speed has reached a value exceeding 15 km / h, or the UE speed has exceeded 30 km / h, etc. The UE rotation has reached a value greater than the threshold. For example, the UE rotation has exceeded 30 degrees or 60 degrees since the last report.

[0388] In some example implementations, the first UE additional condition reported to the NW may depend on the condition since the last time the above was included. RequestUEAdditionalConditions of RRCReconfiguration Since the message was sent, no additional conditions have been reported by the UE.

[0389] In some example implementations of step 3 above, the report may be included RRCReconfigurationComplete The message may include protocol signaling terminating between the AI / ML logical layer and the NW logical entity / unit, which may contain at least one of the following information or information items regarding additional conditions for the UE:

[0390] ●1: UE speed information: Used to indicate the current speed information of the UE when the RRCReconfigurationComplete message is generated.

[0391] ●2: UE codebook information: Codebook information used to indicate the UE regarding the RX beam.

[0392] ●3: UE antenna array distribution information: used to indicate the antenna array distribution of the UE.

[0393] In some other example implementations of step 3 above, the report may be included UE Assistance InformationThe message may be included in protocol signaling terminating between the AI / ML logical layer and the NW logical entity / unit. This protocol signaling may also contain at least one of the following information or information items regarding additional conditions for the UE:

[0394] ●1: UE speed information: Used to indicate the current speed information of the UE when generating the UE supplementary report.

[0395] ●2: UE rotation information: used to indicate and receive... RequestUEAdditionalConditions add conditions) Compared to the UE orientation / direction information at that time, the UE's current rotation information is used.

[0396] The above description and accompanying drawings provide specific example embodiments and implementations. However, the described subject matter can be implemented in a variety of different forms, and therefore, the covered or claimed subject matter is intended to be construed as not being limited to any of the example embodiments set forth herein. A reasonably broad scope is intended for the claimed or covered subject matter. In particular, for example, the subject matter can be implemented as a method, apparatus, component, system, or non-transitory computer-readable medium for storing computer code. Thus, embodiments can take the form of, for example, hardware, software, firmware, storage medium, or any combination thereof. For example, the above-described method embodiments can be implemented by a component, apparatus, or system including a memory and a processor by executing computer code stored in the memory.

[0397] Throughout the specification and claims, terms may have nuanced meanings suggested or implied in the context beyond their explicitly stated meanings. Similarly, the phrase "in one embodiment / implementation" as used herein does not necessarily refer to the same embodiment, and the phrase "in another embodiment / implementation" as used herein does not necessarily refer to different embodiments. For example, the claimed subject matter is intended to include, in whole or in part, combinations of exemplary embodiments.

[0398] References to features, advantages, or similar language throughout this specification do not imply that all features and advantages achievable using this solution should be, or be included in, any single implementation thereof. Rather, references to such features and advantages are to be understood as meaning that a particular feature, advantage, or characteristic described in connection with an embodiment is included in at least one embodiment of this solution. Therefore, throughout this specification, discussions of features and advantages and similar language may, but do not necessarily, refer to the same embodiment.

[0399] Furthermore, the features, advantages, or characteristics described in this solution can be combined in one or more embodiments in any suitable manner. Based on the description herein, those skilled in the art will recognize that this solution can be practiced without one or more of the specific features or advantages of a particular embodiment. In other instances, additional features and advantages may be recognized in certain embodiments that may not be present in all embodiments of this solution.

Claims

1. A method performed by a user equipment (UE) communicating with a network (NW) in a wireless communication network, comprising: Collaborate with the NW to perform a testing process for the UE's artificial intelligence or machine learning (AI / ML) capabilities; Receives AI / ML configuration from the NW for one or more AI / ML features, functions, or models, said one or more AI / ML features, functions, or models selected by the NW from the AI / ML capabilities of the UE based at least in part on the testing process; and Based on the AI / ML configuration, inference is performed using one or more AI / ML features, functions, or models.

2. The method according to claim 1, wherein, The testing process includes: Receive a request message from the NW for testing the AI / ML features, functions, or models of the UE; Perform inference using at least one AI / ML feature, function, or model to generate an inference result; and A response message is sent to the NW, the response message including at least the inference result.

3. The method according to claim 2, wherein, The request message includes at least one of the following: Indicators used to indicate AI / ML features, functions, or models used for testing; and Reference input data for applicable AI / ML features, functions, or models used for testing.

4. The method according to claim 3, wherein, Performing the inference includes processing the reference input data using the indicated AI / ML features, functions, or models to generate the inference result.

5. The method according to claim 4, wherein, The response message includes at least one of the following: The indicator used to indicate the applicable AI / ML feature, function, or model used for testing; or The output data list includes the inference results.

6. The method according to claim 5, wherein, The response message also includes a set of additional UE conditions, which, along with the inference result, are used by the NW to evaluate the one or more AI / ML features, functions, or models.

7. The method according to claim 1, wherein, The AI / ML configuration includes an RRC configuration structure.

8. The method according to claim 7, wherein, The AI / ML configuration is included in the serving cell configuration structure.

9. The method according to claim 8, wherein, The AI / ML configuration includes a list of configurations for one or more AI / ML features, functions, or models.

10. The method according to claim 9, wherein, The AI / ML configuration includes a hierarchical list of configurations for one or more AI / ML features, functions, or models.

11. The method according to claim 9, wherein, Each configuration in the list of configurations for AI / ML features includes at least one of the following: The index or ID of the AI / ML function; Inference-related configurations for the aforementioned AI / ML functions; Configuration related to performance monitoring of the aforementioned AI / ML functions; or The traditional configuration for non-AI functions, corresponding to the AI / ML functions.

12. The method according to claim 11, wherein, The inference-related configuration includes at least one of the following: Measurement configuration for obtaining input data for inference of the AI / ML functions; Inference report configuration for performing inferences on the output data associated with the AI / ML functions; Performance monitoring configuration for inference performance monitoring of the AI / ML functions; or Performance reporting configuration for reporting the inference performance monitoring of the AI / ML functions.

13. The method according to claim 7, wherein, The AI / ML configuration is included in the Channel State Information (CSI) reporting configuration.

14. The method according to claim 13, wherein, The CSI report configuration includes a list of identifiers for the one or more AI / ML features, functions, or models.

15. The method according to claim 13, wherein, The CSI report configuration includes a report volume configuration, which includes a set of inference configurations for the one or more AI / ML features, functions, or models.

16. The method according to claim 13, wherein, The CSI report configuration includes a time configuration for AI / ML inference reports.

17. The method according to claim 16, wherein, The time configuration includes at least one of the following: Periodic reports; Semi-persistent reporting on the Physical Uplink Shared Channel (PUSCH); Semi-continuous reporting on PUCCH; or Non-periodic.

18. The method according to claim 13, wherein, The CSI report configuration includes a measurement resource configuration for inference of the one or more AI / ML features, functions, or models.

19. The method according to claim 13, wherein, The CSI report configuration includes a set of measurement resources for monitoring the inference performance of one or more AI / ML features, functions, or models.

20. The method according to claim 13, wherein, Utilizing the one or more AI / ML features, functions, or models includes: activating or deactivating the one or more AI / ML features, functions, or models.

21. The method according to claim 18, wherein, Activating or deactivating one or more AI / ML features, functions, or models is implicitly triggered by activating or deactivating semi-persistent CSI reports.

22. The method according to claim 1, wherein, Utilizing the one or more AI / ML features, functions, or models includes: activating or deactivating at least one of the one or more AI / ML features, functions, or models.

23. The method according to claim 22, wherein, Activation or deactivation of at least one of the one or more AI / ML features is triggered by the UE receiving signaling from the NW.

24. The method of claim 23, wherein, The signaling includes a DL MAC CE structure.

25. The method of claim 24, wherein, The DL MAC CE structure includes: The first octet, consisting of multiple bits, indicates whether the associated AI / ML function associated with the corresponding serving cell is activated or deactivated; A second octet, comprising multiple bits, indicates the number of subsequent octets in the DL MAC CE structure for the corresponding serving cell; and The subsequent eight bits indicate the AI / ML features and functions to be activated or deactivated for the corresponding serving cell.

26. The method according to claim 24, wherein, The method also includes instructing lower layers about AI / ML activation or deactivation.

27. The method according to claim 23, wherein, The signaling includes semi-persistent CSI report activation / deactivation signaling.

28. The method according to claim 27, wherein, The method also includes instructing lower layers about AI / ML activation or deactivation.

29. The method according to claim 1, further comprising: Report the inference to the NW.

30. The method according to claim 29, wherein, The inference includes at least one of the following: The first N beams with corresponding N reference signal received power (RSRP) values, where N ≥ 1; The first N beams having RSRP values ​​of M of the first N beams, where N>M≥1; The first N beams with RSRP values ​​greater than a predefined RSRP threshold; or This is used to indicate that if an RSRP value exists for the current beam ID, the RSRP value is inferred to be a measured flag.

31. A method performed by a network (NW) communicating with a user equipment (UE) in a wireless communication network, comprising: Collaborate with the UE to perform a testing process targeting the UE's artificial intelligence or machine learning (AI / ML) capabilities; Sending AI / ML configurations for one or more AI / ML features, functions, or models to the UE, wherein the one or more AI / ML features, functions, or models are selected by the NW from the UE's AI / ML capabilities based at least in part on the testing process; and Receive inference and / or measurement reports from the UE, which are generated by the UE based on the AI / ML configuration.

32. The method according to claim 31, wherein, The testing process includes: Send a request message to the UE to test the UE's AI / ML features, functions, or models; and Receive a response message from the UE, the response message including at least the inference result generated by the UE using at least one AI / ML feature, function or model.

33. The method according to claim 32, wherein, The request message includes at least one of the following: Indicators used to indicate AI / ML features, functions, or models used for testing; and Reference input data for applicable AI / ML features, functions, or models used for testing.

34. The method of claim 33, wherein, The response message includes at least one of the following: The indicator used to indicate the applicable AI / ML feature, function, or model used for testing; or The output data list includes the inference results.

35. The method according to claim 34, wherein, The response message also includes a set of additional UE conditions, which, along with the inference result, are used by the NW to evaluate the one or more AI / ML features, functions, or models.

36. The method according to claim 1, wherein, The AI / ML configuration includes an RRC configuration structure.

37. The method of claim 36, wherein, The AI / ML configuration is included in the serving cell configuration structure.

38. The method according to claim 37, wherein, The AI / ML configuration includes a list of configurations for one or more AI / ML features, functions, or models.

39. The method according to claim 38, wherein, The AI / ML configuration includes a hierarchical list of configurations for one or more AI / ML features, functions, or models.

40. The method of claim 38, wherein, Each configuration in the list of configurations for AI / ML features includes at least one of the following: The index or ID of the AI / ML function; Inference-related configurations for the aforementioned AI / ML functions; Configuration related to performance monitoring of the aforementioned AI / ML functions; or The traditional configuration for non-AI functions, corresponding to the AI / ML functions.

41. The method according to claim 40, wherein, The inference-related configuration includes at least one of the following: Measurement configuration for obtaining input data for inference of the AI / ML functions; Inference report configuration for performing inferences on the output data associated with the AI / ML functions; Performance monitoring configuration for inference performance monitoring of the AI / ML functions; or Performance reporting configuration for reporting the inference performance monitoring of the AI / ML functions.

42. The method according to claim 36, wherein, The AI / ML configuration is included in the Channel State Information (CSI) reporting configuration.

43. The method according to claim 42, wherein, The CSI report configuration includes a list of identifiers for the one or more AI / ML features, functions, or models.

44. The method according to claim 42, wherein, The CSI report configuration includes a report volume configuration, which includes a set of inference configurations for the one or more AI / ML features, functions, or models.

45. The method according to claim 42, wherein, The CSI report configuration includes a time configuration for AI / ML inference reports.

46. ​​The method according to claim 45, wherein, The time configuration includes at least one of the following: Periodic reports; Semi-persistent reporting on the Physical Uplink Shared Channel (PUSCH); Semi-continuous reporting on PUCCH; or Non-periodic.

47. The method according to claim 42, wherein, The CSI report configuration includes a measurement resource configuration for inference of the one or more AI / ML features, functions, or models.

48. The method according to claim 42, wherein, The CSI report configuration includes a set of measurement resources for monitoring the inference performance of one or more AI / ML features, functions, or models.

49. The method of claim 31, further comprising: Sending signaling to the UE to activate or deactivate at least one of the one or more AI / ML features, functions, or models.

50. The method of claim 49, wherein, The signaling includes a DL MAC CE structure.

51. The method of claim 50, wherein, The DL MAC CE structure includes: The first octet, consisting of multiple bits, indicates whether the associated AI / ML function associated with the corresponding serving cell is activated or deactivated; A second octet, comprising multiple bits, indicates the number of subsequent octets in the DL MAC CE structure for the corresponding serving cell; and The subsequent eight bits indicate the AI / ML features and functions to be activated or deactivated for the corresponding serving cell.

52. The method according to claim 49, wherein, The signaling includes semi-persistent CSI report activation / deactivation signaling.

53. The method according to claim 31, wherein, The inference includes at least one of the following: The first N beams with corresponding N reference signal received power (RSRP) values, where N ≥ 1; The first N beams having RSRP values ​​of M of the first N beams, where N>M≥1; The first N beams with RSRP values ​​greater than a predefined RSRP threshold; or This is used to indicate that if an RSRP value exists for the current beam ID, the RSRP value is inferred to be a measured flag.

54. The UE or network node according to any one of claims 1 to 53, wherein the UE or network node comprises a processor and a memory, wherein, The processor is configured to read computer code from the memory to cause the UE or the network node to perform the method according to any one of claims 1 to 53.

55. A computer program product comprising a non-transitory computer-readable program medium, wherein computer code is stored on the non-transitory computer-readable program medium, the computer code, when executed by a processor of a UE or a network node according to any one of claims 1 to 53, causes the processor to implement the method according to any one of claims 1 to 53.

Images