Communication method and user equipment

Abstract

A communication method according to one aspect of the present invention is carried out in a mobile communication system. The communication method comprises a step in which user equipment receives, from a network device, a first message that includes training data transmission configuration information indicating transmission of training data used for training an artificial intelligence (AI) / machine learning (ML) model, in response to the occurrence of a prescribed event. The communication method also comprises a step in which the user equipment transmits the training data to the network device in accordance with the training data transmission configuration information. The prescribed event is at least one of: the training data for which the importance or priority is equal to or higher than a first threshold value being detected; the quality of the AI / ML model being equal to or lower than a second threshold value; and a life cycle management (LCM) event having occurred for the AI / ML model.

Description

Communication method and user equipment ,

[0005]

[0001] The present disclosure relates to a communication method and a user equipment.

[0002] In recent years, in the 3GPP (Third Generation Partnership Project) (registered trademark, the same hereinafter), which is a standardization project of a mobile communication system, there has been a study on applying artificial intelligence (AI: Artificial Intelligence) technology, particularly, machine learning (ML: Machine Learning) technology to the wireless communication (air interface) of the mobile communication system.

[0003] 3GPP TR 38.843 V18.0.0 (2023-12)

[0004] The communication method according to the first aspect is a communication method in a mobile communication system. The communication method includes a step of receiving, from a network device, a first message including learning data transmission setting information indicating that a user equipment transmits learning data used for learning an AI / ML model in response to a predetermined event occurring. Further, the communication method includes a step of the user equipment transmitting the learning data to the network device according to the learning data transmission setting information. Here, the predetermined event is at least any one of that learning data having an importance or priority equal to or higher than a first threshold is detected, that the quality of the AI / ML model becomes equal to or lower than a second threshold, and that a LCM (Life Cycle Management) event for the AI / ML model occurs.

[0005] The user device according to the second embodiment is a user device in a mobile communication system. The user device has a receiving unit that receives a first message from a network device, which includes training data transmission setting information indicating that the user device will transmit training data used for training an AI / ML model in response to the occurrence of a predetermined event. The user device also has a transmitting unit that transmits training data to the network device in accordance with the training data transmission setting information. Here, the predetermined event is at least one of the following: detection of training data with importance or priority equal to or greater than a first threshold; the quality of the AI / ML model falling below a second threshold; and the occurrence of an LCM event for the AI / ML model.

[0006] Figure 1 is a diagram showing an example configuration of a mobile communication system according to the first embodiment. Figure 2 is a diagram showing an example configuration of a UE (User Equipment) according to the first embodiment. Figure 3 is a diagram showing an example configuration of a network node (base station) according to the first embodiment. Figure 4 is a diagram showing an example configuration of a protocol stack according to the first embodiment. Figure 5 is a diagram showing an example configuration of a protocol stack according to the first embodiment. Figure 6 is a diagram showing an example configuration of a functional block of AI / ML technology according to the first embodiment. Figure 7(A) shows an example configuration of a functional block of a mobile communication system according to the first embodiment, and Figure 7(B) shows an example configuration of a functional block of a UE according to the first embodiment. Figure 8 is a diagram showing an example configuration of a functional block of a mobile communication system according to the first embodiment. Figures 9(A) and 9(B) are diagrams showing an example configuration of a functional block of a mobile communication system according to the first embodiment. Figures 10(A) and 10(B) are diagrams showing an example configuration of a functional block of a mobile communication system according to the first embodiment. Figure 11 is a diagram showing an example of operation according to the first embodiment.

[0007] This disclosure aims to provide a communication method and user device that enable the user device to transmit learning data at an appropriate time.

[0008] The mobile communication system according to the first embodiment will be described with reference to the drawings. In the drawings, identical or similar parts are denoted by the same or similar reference numerals.

[0009] [First Embodiment] The configuration of the mobile communication system according to the first embodiment will now be described. Figure 1 is a diagram showing an example of the configuration of the mobile communication system 1 according to the first embodiment. The mobile communication system 1 conforms to the 5th generation system (5GS) of the 3GPP standard. In the following description, 5GS will be used as an example, but the mobile communication system may also have an LTE (Long Term Evolution) system applied to it at least partially. The mobile communication system may also have a 6th generation (6G) system or later system applied to it at least partially.

[0010] The mobile communication system 1 comprises a network (NW) 10 and a user device (UE) 100. The UE 100 is a mobile communication device that performs wireless communication with the NW 10. The UE 100 may be any device used by a user, such as a mobile phone terminal (including a smartphone) and / or a tablet terminal, a notebook PC (Personal Computer), a communication module (including a communication card or chipset), a sensor or a device installed on a sensor, a vehicle or a device installed on a vehicle (Vehicle UE), or an aircraft or a device installed on an aircraft (Aerial UE).

[0011] NW10 includes a radio access network (RAN) 20 and a core network (CN) 30. When the mobile communication system is a fifth-generation system (5GS), RAN20 is referred to as NG-RAN (Next Generation Radio Access Network) and CN30 is referred to as 5GC (5G Core Network).

[0012] RAN20 includes multiple network nodes 200 (network nodes 200a to 200c in the example in Figure 1). The network nodes 200 are interconnected via inter-network node interfaces. In RAN20, network nodes 200 are sometimes referred to as base stations. When a network node 200 is a base station, it consists of a CU (Central Unit) and a DU (Distributed Unit) (i.e., functionally divided), and the two units may be connected by a front-haul interface. When the mobile communication system 1 is 5GS, the network nodes 200 are referred to as gNBs, the inter-network node interfaces as Xn interfaces, and the front-haul interfaces as F1 interfaces.

[0013] Furthermore, if at least a part of the mobile communication system 1 is an LTE system, the network node 200 may be an eNB (evolved Node B) which is an LTE base station. Also, if the mobile communication system 1 is a sixth-generation system or later, the network node 200 has the function of a base station and may be a device equivalent to a gNB or eNB.

[0014] Each network node 200 manages one or more cells. Each network node 200 performs wireless communication with the UE 100 that has established a connection with its own cell. Each network node 200 has functions such as wireless resource management (RRM), routing of user data (also simply referred to as "data"), and measurement and control functions for mobility control and scheduling. The term "cell" is used to indicate the smallest unit of a wireless communication area. The term "cell" is also used to indicate a function or resource that performs wireless communication with the UE 100. One cell belongs to one carrier frequency. One cell may be associated with one downlink component carrier and one uplink component carrier. The bandwidth corresponding to one cell (system bandwidth) may be divided into multiple bandwidth parts (BWP: Bandwidth Part). In the following explanation, the gNB may be used as an example of a network node 200.

[0015] CN30 includes a CN (Core Network) device 300. The CN device 300 may include a C-plane device corresponding to the control plane (C-plane) and a U-plane device corresponding to the user plane (U-plane). The C-plane device performs various mobility controls and paging for the UE100. The C-plane device communicates with the UE100 using NAS (Non-Access Stratum) signaling. The U-plane device controls data transfer. When the mobile communication system is 5GS, the C-plane device is called AMF (Access and Mobility Management Function), the U-plane device is called UPF (User Plane Function), and the interface between the network node 200 and the CN device 300 is called the NG interface.

[0016] In the following, the network node 200 and the CN device 300 may be referred to as the network device. The network device may also be the network node 200. The network device may also be the CN device 300.

[0017] Figure 2 shows an example configuration of UE100 (user device) according to the first embodiment. UE100 comprises a receiving unit 110, a transmitting unit 120, and a control unit 130. The receiving unit 110 and the transmitting unit 120 constitute a communication unit 140 that performs wireless communication with the network node 200. UE100 is an example of a communication device.

[0018] The receiving unit 110 performs various types of reception under the control of the control unit 130. The receiving unit 110 includes an antenna and a receiver. The receiver converts the radio signal received by the antenna into a baseband signal (received signal) and outputs it to the control unit 130.

[0019] The transmitting unit 120 performs various types of transmissions under the control of the control unit 130. The transmitting unit 120 includes an antenna and a transmitter. The transmitter converts the baseband signal (transmission signal) output by the control unit 130 into a wireless signal and transmits it from the antenna.

[0020] The control unit 130 performs various control and processing operations in the UE 100. Such processing includes processing in each layer described later. The control unit 130 includes at least one processor and at least one memory. The memory stores programs executed by the processor and information used in processing by the processor. The processor may include a baseband processor and a CPU (Central Processing Unit). The baseband processor performs modulation, demodulation, encoding, and decoding of baseband signals. The CPU executes programs stored in memory and performs various processing operations. Note that processing or operations performed in the UE 100 may also be performed in the control unit 130.

[0021] Figure 3 shows an example configuration of a network node 200 according to the first embodiment. The network node 200 includes a transmitting unit 210, a receiving unit 220, a control unit 230, and an NW communication unit 240. The transmitting unit 210 and the receiving unit 220 constitute a communication unit 250 that performs wireless communication with the UE 100. The NW communication unit 240 constitutes a backhaul communication unit that communicates with the CN 30. The network node 200 is an example of a communication device.

[0022] The transmitting unit 210 performs various types of transmissions under the control of the control unit 230. The transmitting unit 210 includes an antenna and a transmitter. The transmitter converts the baseband signal (transmission signal) output by the control unit 230 into a wireless signal and transmits it from the antenna.

[0023] The receiving unit 220 performs various types of reception under the control of the control unit 230. The receiving unit 220 includes an antenna and a receiver. The receiver converts the radio signal received by the antenna into a baseband signal (received signal) and outputs it to the control unit 230.

[0024] The control unit 230 performs various control and processing operations on the network node 200. Such processing includes processing at each layer described later. The control unit 230 includes at least one processor and at least one memory. The memory stores programs executed by the processor and information used for processing by the processor. The processor may include a baseband processor and a CPU. The baseband processor performs modulation, demodulation, encoding, and decoding of baseband signals. The CPU executes programs stored in memory and performs various processing operations. Processing or operations performed at the network node 200 may also be performed by the control unit 230.

[0025] The NW communication unit 240 is connected to an adjacent network node via the Xn interface, which is an inter-network node interface. The NW communication unit 240 is connected to the CN device 300 via the NG interface, which is an inter-network node-core network interface. The network node 200 may consist of a central unit (CU) and a distributed unit (DU) (i.e., functionally divided), and the two units may be connected by the F1 interface, which is a front-haul interface.

[0026] Figure 4 shows an example of the protocol stack configuration for a user plane wireless interface that handles data.

[0027] The user plane radio interface protocol comprises a physical (PHY) layer, a medium access control (MAC) layer, a radio link control (RLC) layer, a packet data convergence protocol (PDCP) layer, and a service data adaptation protocol (SDAP) layer.

[0028] The PHY layer performs encoding / decoding, modulation / demodulation, antenna mapping / demapping, and resource mapping / demapping. Data and control information are transmitted between the PHY layer of UE100 and the PHY layer of network node 200 via a physical channel. The PHY layer of UE100 receives downlink control information (DCI) transmitted from network node 200 on the physical downlink control channel (PDCCH). Specifically, UE100 performs blind decoding of the PDCCH using a Radio Network Temporary Identifier (RNTI) and acquires the successfully decoded DCI as the DCI addressed to its own UE. The DCI transmitted from network node 200 has a CRC (Cyclic Redundancy Code) parity bit, which is scrambled by the RNTI, added to it.

[0029] In NR, UE100 can use a bandwidth narrower than the system bandwidth (i.e., the cell bandwidth). Network node 200 configures UE100 with a bandwidth portion (BWP) consisting of consecutive PRBs (Physical Resource Blocks). UE100 sends and receives data and control signals in the active BWP. For example, up to four BWPs may be configured for UE100. Each BWP may have a different subcarrier spacing. The frequencies of these BWPs may overlap. If multiple BWPs are configured for UE100, network node 200 can specify which BWP to apply by controlling the downlink. This allows network node 200 to dynamically adjust the UE bandwidth according to the amount of data traffic on UE100, thereby reducing UE power consumption.

[0030] The network node 200 can, for example, configure up to three control resource sets (CORESETs) for each of up to four BWPs on a serving cell. A CORESET is a radio resource for control information that the UE 100 should receive. The UE 100 may have up to twelve or more CORESETs configured on a serving cell. Each CORESET may have an index from 0 to 11 or more. A CORESET may consist of six resource blocks (PRBs) and one, two, or three consecutive OFDM (Orthogonal Frequency Division Multiplex) symbols in the time domain.

[0031] The MAC layer performs data priority control, retransmission processing using Hybrid ARQ (HARQ: Hybrid Automatic Repeat reQuest), and random access procedures. Data and control information are transmitted between the MAC layer of UE100 and the MAC layer of network node 200 via the transport channel. The MAC layer of network node 200 includes a scheduler. The scheduler determines the transport format for the up and down links (transport block size, modulation and coding scheme (MCS)) and the resource blocks to be allocated to UE100.

[0032] The RLC layer transmits data to the receiving RLC layer using the functions of the MAC layer and PHY layer. Data and control information are transmitted between the RLC layer of UE100 and the RLC layer of network node 200 via a logical channel.

[0033] The PDCP layer performs header compression / decompression, encryption / decryption, etc.

[0034] The SDAP layer maps IP flows, which are the units under which the core network performs QoS (Quality of Service) control, to wireless bearers, which are the units under which the access layer (AS: Access Stratum) performs QoS control. Note that if the RAN is connected to the EPC, the SDAP is not required.

[0035] Figure 5 shows the configuration of the protocol stack of the wireless interface of the control plane that handles signaling (control signals).

[0036] The protocol stack of the control plane's radio interface includes a Radio Resource Control (RRC) layer and a Non-Access Stratum (NAS) layer, instead of the SDAP layer shown in Figure 4.

[0037] RRC signaling for various settings is transmitted between the RRC layer of UE100 and the RRC layer of network node 200. The RRC layer controls the logical channel, transport channel, and physical channel in response to the establishment, re-establishment, and release of the wireless bearer. If there is a connection (RRC connection) between the RRC of UE100 and the RRC of network node 200, UE100 is in the RRC connected state. If there is no connection (RRC connection) between the RRC of UE100 and the RRC of network node 200, UE100 is in the RRC idle state. If the connection between the RRC of UE100 and the RRC of network node 200 is suspended, UE100 is in the RRC inactive state.

[0038] The NAS, located above the RRC layer, handles session management and mobility management, among other things. NAS signaling is transmitted between the UE100's NAS and the AMF's NAS. In addition to the wireless interface protocol, the UE100 also has an application layer, etc. Furthermore, the layer below the NAS is called the AS (Access Stratum).

[0039] (AI / ML Technology) Next, the AI / ML (Artificial Intelligence / Machine Learning) technology according to the embodiment will be described. Figure 6 is a diagram showing an example of the configuration of the functional block of the AI / ML technology in the mobile communication system 1 according to the first embodiment.

[0040] The block configuration example of the functions shown in FIG. 6 includes a data collection unit (Data Collection) A1, a model training unit (Model Training) A2, a model inference unit (Inference) A3, a management unit (Management) A5, and a model recording unit (Model Storage) A6.

[0041] The block configuration example of the functions shown in FIG. 6 represents a functional framework of general AI / ML technologies. Therefore, depending on virtual use cases, some parts of the block configuration example (such as the model recording unit A6, etc.) may not be included in the block configuration example. Also, the block configuration example shown in FIG. 6 may be distributed and arranged between the UE 100 and the network device. Alternatively, for some functions in the block configuration example (such as the model training unit A2 or the model inference unit A3, etc.), they may be arranged in both the UE 100 and the network device.

[0042] The data collection unit A1 provides input data to the model training unit A2, the model inference unit A3, and the management unit A5. The input data includes training data (Training Data) for the model training unit A2, inference data (Inference Data) for the model inference unit A3, and monitoring data (Monitoring Data) for the management unit A5.

[0043] The training data serves as the data required for input when the AI / ML model performs learning. The inference data serves as the data required for input when the AI / ML model performs inference. Furthermore, the monitoring data serves as the data required for input during the management of the AI / ML model.

[0044] Note that data collection (Data collection) may be, for example, the process of collecting data in a network node, a management entity, or the UE 100 for performing learning of an AI / ML model, management of an AI / ML model, and inference of an AI / ML model.

[0045] The model learning unit A2 performs AI / ML model training, AI / ML model validation, and AI / ML model testing.

[0046] AI / ML model training is a process of training an AI / ML model based on the relationship between inputs and outputs to obtain a trained AI / ML model for inference. For example, considering y = ax + b, the process of optimizing a (slope) and b (intercept) by providing inputs (x) and outputs (y) (i.e., providing training data) can be AI / ML model training.

[0047] Generally, machine learning includes supervised learning, unsupervised learning, and reinforcement learning. Supervised learning is a method that uses correct data for training data. Unsupervised learning is a method that does not use correct data for training data. For example, in unsupervised learning, feature points are memorized from a large amount of training data, and the correct judgment (estimation of the range) is made. Reinforcement learning is a method of learning a method that assigns a score to the output result and maximizes the score. Hereinafter, supervised learning will be described, but as machine learning, unsupervised learning or reinforcement learning may be applied.

[0048] The model learning unit A2 outputs the trained AI / ML model obtained by AI / ML model training to the model recording unit A6, and the model learning unit A2 also outputs the updated AI / ML model obtained by retraining the trained AI / ML model to the model recording unit A6.

[0049] Note that hereinafter, AI / ML model training may be referred to as "model training" or "training".

[0050] The model inference unit A3 performs AI / ML model inference. Specifically, the model inference unit A3 applies the inference data provided by the data collection unit A1 to a trained AI / ML model (or an updated AI / ML model) to obtain inference output data. For example, considering y = ax + b, x corresponds to the inference data and y corresponds to the inference output data. Note that "y = ax + b" is an AI / ML model. A model with optimized slope and intercept, for example "y = 5x + 3", is a trained AI / ML model. Here, there are various modeling methods (approaches), including linear regression analysis, neural networks, and decision tree analysis. The above "y = ax + b" can also be considered a type of linear regression analysis.

[0051] The model inference unit A3 outputs inference output data to the management unit A5. The model inference unit A3 also receives management instructions from the management unit A5. For example, management instructions include selecting an AI / ML model, activating (deactivating) an AI / ML model, switching between AI / ML models, and fallback (performing inference without using an AI / ML model). The model inference unit A3 performs model inference according to the management instructions.

[0052] AI / ML model inference refers to the process of obtaining a set of outputs from a set of inputs using, for example, a trained AI / ML model (or an updated AI / ML model). Alternatively, model inference may be the process of obtaining inference output data from inference data using a trained AI / ML model (or an updated AI / ML model). Hereafter, AI / ML model inference may be referred to as "model inference" or "inference."

[0053] In the following text, an AI / ML model that is being trained (or updated) may be referred to as a "Training AI / ML model" (or "Updating AI / ML Model"). Also, in the following text, if there is no distinction between an AI / ML model being trained (or updated) and an AI / ML model that has been trained (or updated), it may simply be referred to as an "AI / ML model."

[0054] The management unit A5 supervises operations on the AI / ML model (selection, activation, deactivation, switching, fallback, etc.). The management unit A5 also supervises monitoring of the AI / ML model. Based on monitoring data and inference output data, the management unit A5 can also perform actions to ensure appropriate inference operation. Therefore, the management unit A5 outputs a Model Transfer / Delivery Request to the Model Recording Unit A6, and causes the trained (or updated) AI / ML model recorded in the Model Recording Unit A6 to output to the Model Inference Unit A3. Furthermore, the management unit A5 outputs management instructions to the Model Inference Unit A3, supervising operations on the AI / ML model. Furthermore, the management unit A5 can output performance feedback and retraining requests to the model learning unit A2, which can then retrain the AI / ML model (i.e., update the trained AI / ML model).

[0055] (Use Cases) Next, we will explain use cases in which AI / ML technology is applied. For example, there are three use cases in which AI / ML technology is applied:

[0056] (X1.1) "CSI (Channel State Information) Feedback Improvement"

[0057] (X1.2) "Beam management"

[0058] (X1.3) “Positioning accuracy enhancement”

[0059] (X1.1) CSI Feedback Improvement "CSI feedback improvement" describes a use case where AI / ML technology is applied to the CSI that is fed back from UE100 to network node 200. CSI is information about the channel status in the downlink between UE100 and network node 200. CSI includes at least one of the following: Channel Quality Indicator (CQI), Precoding Matrix Indicator (PMI), and Rank Indicator (RI). Network node 200 performs, for example, downlink scheduling based on the CSI feedback from UE100.

[0060] In the "CSI Feedback Improvement" use case, there are two sub-use cases: CSI compression in the frequency domain and CSI prediction in the time domain.

[0061] (X1.1.1) Sub-use case: CSI compression In CSI compression, the CSI inferred using the trained AI / ML model in UE100 is compressed in UE100. The compressed CSI is transmitted from UE100 to network node 200.

[0062] Figure 7(A) is a diagram showing an example of the configuration of a functional block in the mobile communication system 1 when CSI compression is used. As shown in Figure 7(A), the UE 100 has a CSI generation unit 101, and the network node 200 has a CSI reconstruction unit 201.

[0063] The CSI generation unit 101 includes a CSI generation inference unit 1010 and a quantization unit 1011. The CSI generation inference unit 1010 infers CSI (inference output data) from the input using a trained AI / ML model. The input to the CSI generation inference unit 1010 may be, for example, a partial (or punctured) CSI. The partial CSI may be a CSI measured using a CSI reference signal (CSI-RS (Channel State Information-Reference Signal)) (or demodulation reference signal (DMRS)) transmitted using a certain amount of resources or less. Alternatively, the input to the CSI generation inference unit 1010 may be a CSI reference signal. The output of the CSI generation inference unit 1010 is a CSI with a larger number of CSIs than the partial CSI input, if a partial CSI is input. Also, the output of the CSI generation inference unit 1010 is a CSI if a CSI reference signal is input. Hereinafter, the output of the CSI generation inference unit 1010 will be referred to as the full CSI. The quantization unit 1011 quantizes the full CSI. Compression is performed through quantization. The quantization unit 1011 transmits the quantized full CSI to the network node 200 as CSI feedback.

[0064] The CSI reconstruction unit 201 includes an inverse quantization unit 2010 and a CSI reconstruction inference unit 2011. The inverse quantization unit 2010 receives CSI feedback, inverse quantizes the quantized full CSI, and outputs the full CSI. The CSI reconstruction inference unit 2011 uses a trained AI / ML model to infer the reconstructed full CSI from the output of the inverse quantization unit 2010 (full CSI). The reconstructed full CSI is output from the CSI reconstruction unit 201.

[0065] Furthermore, the quantization unit 1011 may be merged with the CSI generation inference unit 1010, and the inverse quantization unit 2010 may also be merged with the CSI reconstruction inference unit 2011. In addition, a pre-processing unit may be provided before the CSI generation inference unit 1010, and a post-processing unit may be provided after the CSI reconstruction inference unit 2011.

[0066] As shown in Figure 7(A), the sub-use case of CSI compression is based on a two-sided model in which inference is performed using a trained AI / ML model on both the UE100 side and the network node 200 side.

[0067] (X1.1.2) Sub-use case: CSI prediction In CSI prediction, a trained AI / ML model is used to infer (predict) future CSIs from the history of past CSIs.

[0068] Figure 7(B) shows an example of the configuration of a functional block in UE100 when CSI prediction is used. CSI prediction is based on the UE-side model in UE100, where inference is performed using a pre-trained AI / ML model.

[0069] As shown in Figure 7(B), UE100 has a CSI prediction model 102. The CSI prediction model 102 has a trained AI / ML model that takes past CSI history as input and infers future CSI (predicted CSI). The CSI prediction model 102 may also be a CSI prediction inference unit. UE100 (CSI prediction model 102) transmits the predicted CSI as CSI feedback to the network node 200.

[0070] Furthermore, the CSI prediction model 102 may have a pre-processing unit before it, or a post-processing unit after it.

[0071] (X1.2) Beam Management In beam management use cases, there are two sub-use cases: spatial-domain downlink beam prediction, which performs beam prediction in the spatial direction, and temporal downlink beam prediction, which performs beam prediction in the temporal direction. Spatial-domain downlink beam prediction is called "BM Case 1" (BM-Case 1), and temporal downlink beam prediction is called "BM Case 2" (BM-Case 2).

[0072] Figure 8 shows an example of the configuration of a functional block in the mobile communication system 1 when BM case 1 is used. In the case of BM case 1, either a UE side model in which inference is performed at UE 100 or an NW side model in which inference is performed on the network side may be applied. Therefore, as shown in Figure 8, the AI / ML model 103 that performs inference may reside at UE 100 or at NW 10 (including the network node 200 or CN device 300).

[0073] In BM Case 1, the input to the AI / ML model 103 is the measured value for each beam included in beamset B. On the other hand, the output from the AI / ML model 103 (inference output data) is the probability that each (downstream) beam included in (predicted) beamset A will be the top beam. Beamset A and beamset B may be different. Alternatively, beamset B may be a subset of beamset A. The measured values ​​of beamset B, which are the input to the AI / ML model 103, may be represented by RSRP (Reference Signal Received Power).

[0074] Figure 8 also shows an example configuration of the mobile communication system 1 in the case of BM case 2. In the case of BM case 2, either the UE side model or the NW side model may be applied. In the case of BM case 2 as well, the AI / ML model 103 is located in either UE 100 or NW 10.

[0075] In BM Case 2, the input to the AI / ML model 103 is the history of measurements for each beam included in beamset B. The previously measured measurements for each beam are input to the AI / ML model 103. On the other hand, the output from the AI / ML model 103 (inference output data) is the probability that each (downstream) beam included in (predicted) beamset A will be the top beam, similar to BM Case 1. Beamset A and beamset B may be different, and beamset B may be a subset of beamset A. Also, in BM Case 2, beamset A and beamset B may be the same.

[0076] In both BM Case 1 and BM Case 2, the UE side model allows UE100 to report prediction results to NW10. Also, in both BM Case 1 and BM Case 2, the NW side model can predict the top beam based on the measured values ​​for each beam included in beamset B reported by UE100.

[0077] (X1.3) "Positioning Accuracy Enhancement" In the use case for improving positioning accuracy, there are two sub-use cases: Direct AI / ML positioning, which directly infers the position of UE100 using a trained AI / ML model, and AI / ML assisted positioning, which infers intermediate position measurements. In the latter, AI / ML assisted positioning, the position of UE100 is measured or inferred in the LMF (Location Management Function) using intermediate position measurements. Intermediate position measurements can serve as assisting information for measuring or inferring the position of UE100 in the LMF. The LMF may have a pre-trained AI / ML model, and uses this pre-trained AI / ML model to infer the position of UE100.

[0078] (X1.3.1) Sub-use case: Direct AI / ML positioning Figure 9(A) is a diagram showing an example of the configuration of a functional block in the mobile communication system 1 when direct AI / ML positioning is used. In the case of direct AI / ML positioning, the UE side model and the network side model are applied. Therefore, the AI / ML model 104 used for inference may reside in the UE 100 or in the NW 10.

[0079] In the case of direct AI / ML positioning, the input to the AI / ML model 104 is the measured value at each measurement point (TRP: Transmission and / or Reception Point). The measured value can be, for example, a Channel Impulse Response (CIR), a Power Delay Profile (PDP), or a fingerprint. For example, both CIR and PDP represent the delay time for a signal at a specific frequency, but CIR represents the instantaneous delay time, while PDP represents the statistical delay time. On the other hand, the output from the AI / ML model 104 (inference output data) is the position information of the UE 100. The position information may also be represented by a fingerprint. The fingerprint represents, for example, the measurement information for the cell of the UE 100.

[0080] (X1.3.2) Sub-use case: AI / ML assisted positioning Figures 9(B) to 10(B) are diagrams showing examples of the configuration of functional blocks in the mobile communication system 1 when AI / ML assisted positioning is used. In the case of AI / ML assisted positioning, the UE side model and the network side model are also applied. Therefore, the AI / ML model 105 used for inference may reside in the UE 100 or in the NW 10. In the case of AI / ML assisted positioning, there are cases in which one AI / ML model 105 is used for multiple inputs (Figure 9(B)), cases in which the same AI / ML model is used for each of the multiple inputs (Figure 10(A)), and cases in which different AI / ML models are used for each of the multiple inputs (Figure 10(B)).

[0081] In either case, the input to the AI / ML model 105 is the channel measurement value at each measurement point (TRP). The channel measurement value may be CIR, PDP, or fingerprint, similar to direct AI / ML positioning. On the other hand, the output from the AI / ML model 105 (inference output data) is, in either case, an intermediate position measurement value used for position measurement. The intermediate position measurement value may be LOS or NLOS identification, measurement timing and / or measurement angle, or likelihood of measurement.

[0082] (LCM) Currently, 3GPP is discussing the Life Cycle Management (LCM) of AI / ML models.

[0083] Specifically, the LCM of an AI / ML model may include at least one of the following actions:

[0084] (L1) Data collection

[0085] (L2) Model training

[0086] (L3) Identification

[0087] (L4) Model delivery or transfer

[0088] (L5) Model inference operation

[0089] (L6) Selection, Activation, Deactivation, Switching, and Fallback operations

[0090] (L7) Monitoring

[0091] (L8) Model update

[0092] (L9) UE capability

[0093] On the network side, for example, by controlling each operation of the LCM, it becomes possible to properly manage everything from the generation of AI / ML models to their deletion (or disposal). Fallback refers to switching from an AI / ML model to a model that does not use an AI / ML model. A model that does not use an AI / ML model is sometimes called a "legacy model."

[0094] 3GPP specifies two types of LCM: function-based LCM and model ID-based LCM.

[0095] A function-based LCM may be an operation performed on a function of an AI / ML model. Specifically, a function-based LCM may perform one of the following operations on a function of an AI / ML model: activation, deactivation, switching, or fallback. The network side can, for example, use 3GPP signaling (RRC messages, MAC CE, or DCI) to instruct the LCM operation on a function of an AI / ML model. In a function-based LCM, UE100 may have one AI / ML model for one function. UE100 may have multiple AI / ML models for a single function.

[0096] On the other hand, a model ID-based LCM may be an operation performed on individual AI / ML models using the model ID. The model ID is identification information used to distinguish one AI / ML model from other AI / ML models. Specifically, a model ID-based LCM may use the model ID to perform one of the following operations on each AI / ML model: activation, deactivation, switching, or selection.

[0097] (Communication method according to the first embodiment) One of the LCM operations is data collection. As described above, the data collected during data collection may be used for training an AI / ML model. Here, we assume that training of the AI / ML model is performed on the network 10 side. In this case, the training data used for the training may be collected by UE 100 and transmitted from UE 100 to the network 10 side. On the network 10 side, training (or retraining) of the AI / ML model is performed using the training data collected by UE 100.

[0098] 3GPP currently has the following agreements regarding training data:

[0099] - UE100 stores training data in the AS layer. - UE100 stops measuring for data collection purposes when the buffer limit is reached. - However, whether or not to support AS buffer event-based reporting requires further discussion.

[0100] In the first embodiment, we focus on the transmission timing of training data in UE100. As mentioned above, reporting based on AS buffer events is subject to future discussion, but currently, the following proposals have been made by various companies: namely, UE100 transmits training data when the buffer limit is reached, a buffer threshold is sent from network 10 to UE100, transmission is set based on wireless conditions such as when the RSRP value measured by UE100 is greater than the threshold, when it is decided to hand over the serving cell of UE100 to another cell, and when UE100 returns from an RRC idle or RRC inactive state to an RRC connected state.

[0101] However, if training data with an importance or priority level above the first threshold (i.e., "high importance or priority") remains stored in memory and is not transmitted after a predetermined period of time, the network device may not perform training on the AI / ML model and may not be able to generate an AI / ML model with a certain level of accuracy. This is because high-importance or high-priority training data influences the training of the AI / ML model, and by training with such data, it is possible to generate an AI / ML model with a certain level of accuracy.

[0102] Therefore, the objective of the first embodiment is to enable UE100 to transmit training data at an appropriate time.

[0103] Therefore, in the first embodiment, firstly, the user device (e.g., UE100) receives a first message from the network device that includes training data transmission setting information indicating that it will transmit training data used for training an AI / ML model in response to a predetermined event. Secondly, the user device transmits training data to the network device in accordance with the training data transmission setting information. Here, the predetermined event may be the detection of training data with an importance or priority of a first threshold or higher. Alternatively, the predetermined event may be the quality of the AI / ML model falling below a second threshold. Alternatively, the predetermined event may be the occurrence of an LCM event for the AI / ML model.

[0104] Thus, in the first embodiment, the UE 100 transmits training data to the network device when it detects training data that is "high in importance or priority" and whose importance or priority is equal to or greater than a first threshold, triggered by such detection. Therefore, it is possible to avoid a situation where training data with high importance or priority is detected by the UE 100 but is not transmitted to the network device even after a predetermined time has elapsed. Consequently, the UE 100 can transmit the training data at an appropriate time.

[0105] In the first embodiment, the UE 100 determines whether the acquired training data is data of high importance or priority. Specifically, the UE 100 determines the importance or priority of the training data, and if the importance or priority is above a threshold, it determines that the training data is data of high importance or priority. This threshold may be included in the training data transmission setting information transmitted from the network device to the UE 100.

[0106] Alternatively, the training data transmission settings information may include determination information indicating whether or not the data is of high importance or priority. The UE 100 may use this determination information to determine the importance or priority of the collected training data.

[0107] In the following, training data may be referred to as "training data," and training data transmission settings may be referred to as "training data transmission settings."

[0108] (Importance of the First Embodiment) Here, we will explain the importance of the first embodiment.

[0109] Firstly, ground truth data may be "high-importance data." Ground truth data refers to real-world data provided through direct observation, rather than, for example, the inference output data of an AI / ML model. Ground truth data may also be the correct data used as a comparison when monitoring the inference output data. The training data transmission setting information includes the type of data that may be ground truth data (determination information), and the control unit 130 of UE 100 may determine that data acquired through data collection is ground truth data and therefore high-importance data if it falls under that type. On the other hand, if the data does not fall under that type, the control unit 130 of UE 100 may determine that it is not ground truth data but "low-importance data."

[0110] Secondly, training data that is ground truth data and has a certain level of reliability may also be considered "high-importance data." For example, assuming that GNSS (Global Navigation Satellite System) data acquired using a GNSS receiver includes information indicating the type of satellite, the control unit 130 of the UE100 may determine that the acquired GNSS data (training data) is "high-importance data" if that information indicates a specific satellite (e.g., "Michibiki"). On the other hand, if the type of satellite included in the GNSS data is not a specific satellite, the control unit 130 of the UE100 may determine that the GNSS data is "low-importance data." The information representing the type of a specific satellite becomes the "determination information."

[0111] Alternatively, the control unit 130 of the UE 100 may determine that the acquired GNSS data (training data) indicates the location of a specific building (for example, a network node 200) and that the GNSS data indicates a location other than the specific building and that the GNSS data indicates a location other than the specific building and that the GNSS data indicates a location other than the specific building and that the GNSS data indicates a location other than the specific building. In this case, the location information of the specific building becomes the "determination information".

[0112] Alternatively, the control unit 130 of the UE100 may determine the importance of the data based on the GNSS data. For example, if the received electric field strength of the GNSS data is within a certain range, it may be determined to be high-importance data, and if it is outside that range, it may be determined to be low-importance data. Alternatively, the control unit 130 of the UE100 may determine the importance of the GNSS data based on the DOP (Dilution of Precision) value. The DOP value is an index representing the positioning accuracy calculated based on the position of the satellite from which the GNSS data is acquired and the position of the receiver.

[0113] Thirdly, unique or rare data may also be considered "high-importance data." Unique or rare data, as training data, has a certain level of potential to influence the AI / ML model, and therefore, it may be desirable to train it as quickly as possible as "high-importance data." On the other hand, if non-unique or non-rare data is used as training data for an AI / ML model, overfitting may occur, causing the AI / ML model to become overfit to the training data and thus losing its versatility.

[0114] For example, assuming that the training data follows a normal distribution, UE100 determines that the training data is specific or rare, i.e., highly important, when the absolute value of the Z-value shown in equation (1) below is greater than or equal to a predetermined value (e.g., "2"). On the other hand, UE100 may determine that the training data is not specific or rare, i.e., not highly important, when the absolute value of the Z-value for the training data is less than the predetermined value. This predetermined value becomes the "threshold" for importance determination and is included in the training data transmission setting information.

[0115] Z = (xi - μ) / σ ... (1) In equation (1), xi represents the value of a specific data point, μ represents the mean of the entire dataset, and σ represents the standard deviation of the entire dataset. UE100 uses equation (1) to calculate the Z value of the training data, and then compares the calculated Z value with a predetermined value to determine its importance.

[0116] (Priority according to the first embodiment) Next, the priority according to the first embodiment will be explained.

[0117] Firstly, the aforementioned "data of high importance" may also be "data of high priority," and the aforementioned "data of low importance" may also be "data of low priority."

[0118] Secondly, training data with a certain level of real-time capability is considered "high-priority data," while training data with a level of real-time capability below a certain point may be considered "low-priority data." Training data with a certain level of real-time capability may affect the learning of the AI / ML model depending on the type of model, so it may be desirable to perform learning as quickly as possible. On the other hand, training data with a level of real-time capability below a certain point may be acceptable even if there is a transmission delay, depending on the type of AI / ML model.

[0119] For example, UE100 may determine that training data acquired during movement is data with a real-time quality above a certain level, and training data acquired while stationary may determine that training data has a real-time quality below a certain level. UE100 has a GNSS receiver, and if the temporal change in GNSS data acquired by the GNSS receiver is greater than or equal to a "threshold," it may determine that the training data acquired during that time is training data acquired during movement, and if the temporal change is less than the "threshold," it may determine that the training data acquired during that time is training data acquired while stationary. This "threshold" may be included in the training data transmission setting information.

[0120] Alternatively, training data acquired at a specific time may be determined to have a certain level of real-time capability, while training data acquired at times other than the specific time may be determined to have a lower level of real-time capability. The training data transmission setting information includes "determination information" indicating the specific time, and the UE100 can determine whether or not the training data was acquired at the specific time based on whether or not the acquisition time of the training data falls within that time.

[0121] Alternatively, UE100 may determine that training data with a real-time capability above a certain level is "high-priority data," and training data with a real-time capability below a certain level is "low-priority data."

[0122] Thirdly, UE100 may determine that data has a high priority if the KPI is above a certain level, and low priority if the KPI is below a certain level. For example, UE100 may further train an existing AI / ML model to create a trained AI / ML model, and then use this trained AI / ML model to determine the KPI.

[0123] (Example of operation according to the first embodiment) Next, an example of operation according to the first embodiment will be described.

[0124] Figure 11 is a diagram illustrating an example of operation according to the first embodiment. In the example shown in Figure 11, a network node 200 is used as the network device, and the diagram illustrates an example where learning for an AI / ML model is performed at the network node 200.

[0125] As shown in Figure 11, in step S10, the receiving unit 220 of the network node 200 receives the AI / ML model from the OTT server.

[0126] In step S11, the receiving unit of the network node 200 receives information regarding the importance or priority of the AI / ML model from the OTT server. The information regarding importance or priority may be a threshold or determination information used by the UE 100 when determining the importance or priority of the training data. Since the AI / ML model is generated in the OTT server, information regarding the importance or priority of the training data used to train the AI / ML model is also generated in the OTT server. Note that the information regarding importance or priority may also be generated in the network node 200.

[0127] In step S12, the transmission unit 210 of the network node 200 sends a message containing training data transmission setting information to the UE 100. The training data transmission setting information includes information indicating that training data used for training the AI / ML model should be transmitted in response to the occurrence of a predetermined event. The predetermined event is, as described above, at least one of the following: the detection of training data with importance or priority equal to or greater than the first threshold; the quality of the AI / ML model falling below the second threshold; and the occurrence of an LCM event for the AI / ML model. The training data transmission setting information may also include threshold or judgment information acquired in step S11. The message may be an RRC message or a message of a newly defined layer for AI / ML (e.g., an AI / ML message). The training data transmission setting information may also be transmitted using MAC CE or downlink control information (DCI). The receiving unit 110 of the UE 100 receives the training data transmission setting information.

[0128] In step S13, the control unit 130 of the UE 100 acquires training data and stores it in memory. This memory may be the memory of the AS layer. The memory of the AS layer may be used to store measurement results of Quality of Experience (QoE) in services such as VR (Virtual Reality) services or high-quality streaming services (DASH: Dynamic Adaptive Streaming over HTTP). The minimum memory capacity of the AS layer memory may be 64 KBytes, as described above.

[0129] In step S14, the control unit 130 of the UE 100 detects the occurrence of a predetermined event. The control unit 130 of the UE 100 may detect "important or high-priority data" when the training data acquired in step S13 is equal to or greater than a threshold (e.g., a first threshold) included in the training data transmission setting information acquired in step S12, thereby detecting the predetermined event. Alternatively, the control unit 130 of the UE 100 may detect the predetermined event when the training data acquired in step S13 matches the judgment information included in the training data transmission setting information acquired in step S12, thereby detecting "important or high-priority data". Alternatively, the control unit 130 of the UE 100 may detect the predetermined event when the receiving unit 110 of the UE 100 receives a message (second message) containing information indicating that the quality of the AI / ML model (e.g., KPI) is below a second threshold, for example, information indicating that the KPI has decreased. Alternatively, the control unit 130 of the UE 100 may detect a predetermined event in response to the receiving unit 110 of the UE 100 receiving a message (e.g., a third message) containing information indicating that an LCM event occurred on the network 10 side. The LCM event may be any of the following: AI / ML model deactivation, AI / ML model switching, or AI / ML model fallback.

[0130] In step S15, the transmitting unit 120 of UE 100, in response to detecting a predetermined event in step S14, transmits the training data stored in memory to the network node 200 according to the training data transmission setting information. The training data may be transmitted using user plane messages. Alternatively, the training data may be transmitted using MAC CE or uplink control information (UCI). The receiving unit 220 of the network node 200 receives the training data.

[0131] In step S16, the control unit 230 of the network node 200 performs training (i.e., model training) on ​​the AI / ML model using the training data received in step S15.

[0132] (Other operational examples related to the first embodiment) In the first embodiment, a network node 200 was used as an example of a network device, but a CN device 300 may also be used as an example of a network device. In this case, in the example shown in Figure 11, a CN device is used instead of a network node 200. If the CN device is an AMF, NAS messages may be used in steps S12 and S15.

[0133] Alternatively, an LMF may be used as an example of a network device. In this case, in the example shown in Figure 11, the LMF is used instead of the network node 200. In this case, steps S12 and S15 may use LPP messages via LPP (LTE Positioning Protocol).

[0134] Alternatively, in the example shown in Figure 11, an OTT server may be used instead of the network node 200. In this case, steps S12 and S15 may use IP messages via the IP (Internet Protocol) protocol.

[0135] Alternatively, UE100 may be connected via DC (Dual Connectivity) to two cells: a master cell belonging to MCG (Master Cell Group) and a secondary cell belonging to SCG (Secondary Cell Group). In this case, the network node 200 on the master cell side transmits training data transmission setting information to UE100, similar to the first embodiment (step S12). Meanwhile, the network node 200 on the secondary cell side also transmits training data transmission setting information to UE100. This training data transmission setting information, like the training data transmission setting information transmitted from the master cell side, includes information indicating that training data will be transmitted triggered by the detection of a predetermined event. As a result, for example, UE100 can transmit training data to the secondary cell side when it detects the occurrence of a predetermined event.

[0136] [Other Embodiments] In the first embodiment, an example was described in which the importance and priority of training data are determined once, but the determination may be performed multiple times. For example, in UE100, if it is detected multiple times that the threshold included in the training data transmission setting information obtained in step S12 is greater than or equal to the first threshold for the training data obtained in step S13, the data may be determined to be "data with high importance or priority".

[0137] Alternatively, UE100 may determine that the training data acquired in step S13 is "data of high importance or priority" if it detects for a certain period of time that the threshold included in the training data transmission setting information acquired in step S12 is equal to or greater than the first threshold.

[0138] The first embodiment described above primarily focuses on supervised learning, but is not limited to this. For example, the first embodiment may also apply to unsupervised learning or reinforcement learning.

[0139] The above-described operation flows can be performed not only independently, but also in combination of two or more operation flows. For example, some steps of one operation flow may be added to another operation flow, or some steps of one operation flow may be replaced with some steps of another operation flow. It is not necessary to execute all steps in each flow; only some steps may be executed. Furthermore, the order of steps in each flow may be changed as appropriate.

[0140] In the above embodiment, an example was described in which the base station is an NR base station (gNB), but the base station may also be an LTE base station (eNB) or a 6G base station. Furthermore, the base station may also be a relay node such as an IAB (Integrated Access and Backhaul) node. The base station may also be a DU of an IAB node. Additionally, UE100 may be an MT (Mobile Termination) of an IAB node. That is, UE100 may be a terminal function unit (a type of communication module) for the base station to control a relay device that performs signal relay. Such a terminal function unit is referred to as an MT. Examples of multi-transmission architectures (MTs) include IAB-MT, NCR (Network Controlled Repeater)-MT, and RIS (Reconfigurable Intelligent Surface)-MT.

[0141] Furthermore, the term "network node" primarily refers to a base station, but may also refer to a core network device or a part of a base station (CU, DU, or RU). Additionally, a network node may consist of a combination of at least a part of the core network device and at least a part of a base station.

[0142] A program is provided that causes a computer to execute each of the processes according to the above embodiment. The program may be recorded on a computer-readable medium. Using a computer-readable medium, it is possible to install the program on a computer. Here, the computer-readable medium on which the program is recorded may be a non-transient storage medium. The non-transient storage medium is not particularly limited, but for example, it may be a storage medium such as a CD-ROM and / or DVD-ROM. Furthermore, the circuits that execute each of the processes performed by the apparatus according to the above embodiment may be integrated, and at least a part of the apparatus may be configured as a semiconductor integrated circuit (chipset, SoC: System on a chip).

[0143] The functions realized by the apparatus according to the above-described embodiment may be implemented in a circuit or processing circuit, including a general-purpose processor, application-specific processor, integrated circuit, ASICs (Application Specific Integrated Circuits), CPU (a Central Processing Unit), conventional circuits, and / or combinations thereof, which are programmed to realize the described functions. The processor includes transistors and / or other circuits and is considered to be a circuit or processing circuit. The processor may be a programmed processor that executes a program stored in memory. In this disclosure, circuit, unit, and means are hardware programmed to realize or execute the described functions. The hardware may be any hardware disclosed herein, or any hardware known to be programmed to perform or execute the functions described herein. If the hardware is a processor that is considered to be of the type of circuit, the circuit, means, or unit is a combination of hardware and software used to constitute the hardware and / or processor.

[0144] The phrases “based on” and “depending on / in response to” as used in this disclosure do not mean “based solely on” or “in response solely” unless otherwise specified. “Based on” means both “based solely on” and “at least partially on.” Similarly, “depending” means both “at least partially on” and “at least partially on.” The terms “include,” “comprise,” and variations thereof do not mean that they include only the listed items, but may include only the listed items or may include additional items in addition to the listed items. Furthermore, the term “or” as used in this disclosure is not intended to mean exclusive OR. Additionally, any reference to elements using designations such as “first,” “second,” etc., as used in this disclosure does not generally limit the quantity or order of those elements. These designations may be used herein as a convenient way to distinguish between two or more elements. Therefore, references to the first and second elements do not imply that only two elements may be adopted therein, or that the first element must precede the second element in any way. In this disclosure, where articles are added by translation, such as a, an, and the in English, these articles shall be plural unless it is clearly indicated from the context that they are not.

[0145] Although the embodiments have been described in detail above with reference to the drawings, the specific configuration is not limited to those described above, and various design changes can be made without departing from the gist of the work. Furthermore, each embodiment, each operation example, or each process can be combined as appropriate, as long as they do not contradict each other.

[0146] This application claims priority to Japanese Patent Application No. 2024-171552 (filed September 30, 2024), the entirety of which is incorporated into the specification of this application.

[0147] (Note) The above can be summarized as stated in the note, but the note does not limit the embodiments.

[0148] (Note 1) A communication method in a mobile communication system, comprising the steps of: receiving a first message from a network device, which includes learning data transmission setting information indicating that a user device will transmit learning data used for training an AI / ML model in response to the occurrence of a predetermined event; and the user device transmitting the learning data to the network device in accordance with the learning data transmission setting information, wherein the predetermined event is at least one of the following: the detection of learning data with an importance or priority of a first threshold or higher; the quality of the AI / ML model falling below a second threshold; and the occurrence of an LCM (Life Cycle Management) event for the AI / ML model.

[0149] (Note 2) The communication method described in Note 1, wherein the learning data is data stored in the AS buffer of the user device.

[0150] (Note 3) The communication method according to Note 1 or Note 2, wherein the transmission step includes detecting the predetermined event in response to the user device receiving a second message containing information indicating that the quality of the AI / ML model is below a second threshold.

[0151] (Note 4) The communication method according to any one of Notes 1 to 3, wherein the transmission step includes a step of detecting the predetermined event in response to the user device receiving a third message containing information indicating that the LCM event has occurred.

[0152] (Note 5) The communication method according to any one of Notes 1 to 4, wherein the user device acquires the training data and the network device performs training of the AI / ML model using the training data.

[0153] (Note 6) A user device in a mobile communication system, comprising: a receiving unit that receives from a network device a first message including learning data transmission setting information indicating that learning data used for training an AI / ML model will be transmitted in response to the occurrence of a predetermined event; and a transmitting unit that transmits the learning data to the network device in accordance with the learning data transmission setting information, wherein the predetermined event is at least one of the following: detection of learning data with importance or priority equal to or greater than a first threshold; the quality of the AI / ML model falling below a second threshold; and the occurrence of an LCM event for the AI / ML model.

[0154] 1: Mobile communication system 10: NW 20: RAN 30: CN 100: UE 101: CSI generation unit 102: CSI prediction model 103, 104, 105: AI / ML model 110: Receiving unit 120: Transmitting unit 130: Control unit 140: Communication unit 200: Network node 201: CSI reconstruction unit 210: Transmitting unit 220: Receiving unit 230: Control unit 240: NW communication unit 250: Communication unit 300: CN device A1: Data acquisition unit A2: Model learning unit A3: Model inference unit A5: Management unit A6: Model recording unit 1010: Inference unit for CSI generation 1011: Quantization unit 2010: Inverse quantization unit 2011: Inference unit for CSI reconstruction

Claims

1. A communication method in a mobile communication system, comprising: a user device receiving a first message from a network device, which includes learning data transmission setting information indicating that the user device will transmit learning data used for training an AI (Artificial Intelligence) / ML (Machine Learning) model in response to the occurrence of a predetermined event; and the user device transmitting the learning data to the network device in accordance with the learning data transmission setting information, wherein the predetermined event is at least one of the following: the detection of learning data with an importance or priority of 1 threshold or higher; the quality of the AI / ML model falling below 2 thresholds; and the occurrence of an LCM (Life Cycle Management) event for the AI / ML model.

2. The communication method according to claim 1, wherein the learning data is data stored in the AS buffer of the user device.

3. The communication method according to claim 1, wherein the transmission includes detecting the predetermined event in response to the user device receiving a second message containing information indicating that the quality of the AI / ML model is below a second threshold.

4. The communication method according to claim 1, wherein the transmission includes detecting the predetermined event in response to the user device receiving a third message containing information indicating that the LCM event has occurred.

5. The communication method according to claim 1, wherein the user device acquires the training data, and the network device performs training on the AI / ML model using the training data.

6. A user device in a mobile communication system, comprising: a receiving unit that receives from a network device a first message including learning data transmission setting information indicating that learning data used for training an AI (Artificial Intelligence) / ML (Machine Learning) model will be transmitted in response to the occurrence of a predetermined event; and a transmitting unit that transmits the learning data to the network device in accordance with the learning data transmission setting information, wherein the predetermined event is at least one of the following: detection of learning data with importance or priority equal to or greater than a first threshold; the quality of the AI / ML model falling below a second threshold; and the occurrence of an LCM event for the AI / ML model.

Images