Base station, communication terminal, control method, communication method, and program

Abstract

The purpose of the present invention is to provide a base station capable of selecting a learning model through collaboration between the base station and a communication terminal. This base station comprises: a transmission unit that, when a learning model used for determining a beam to be used by a communication terminal for communication with the base station is switched to another learning model, transmits, to the communication terminal, a first RRC message for inquiring about information related to at least one candidate learning model serving as a switching destination candidate; and a reception unit that receives, from the communication terminal, a second RRC message including the information related to the at least one candidate learning model.

Description

Base station, communication terminal, control method, communication method and program

[0001] The present disclosure relates to a base station, a communication terminal, a control method, a communication method, and a program.

[0002] The 3GPP (3rd Generation Partnership Project) (registered trademark), which formulates communication standards for mobile networks, is studying functions to support AI (Artificial Intelligence) and ML (Machine Learning) in 5G. Non-Patent Document 1 shows Channel State Information (CSI) feedback enhancement, beam management, and positioning accuracy enhancement as use cases related to AI / ML.

[0003] The beam management use cases include downlink (DL) spatial-domain beam prediction and temporal beam prediction. Downlink spatial-domain beam prediction is called BM-Case 1. Downlink temporal beam prediction is called BM-Case 2.

[0004] Downlink (DL) beam prediction in the spatial domain predicts beam set A based on measurements from beam set B. Beam set B is a set of beams whose measurements are used as inputs to the AI / ML model. Beam set A is different from beam set B, or beam set B is a subset of beam set A.

[0005] Temporal downlink (DL) beam prediction is the prediction of beam set A based on historical measurement results of beam set B. Beam set B is a set of beams whose measurements are used as inputs to an AI / ML model. In temporal DL beam prediction, beam set A may be different from beam set B, beam set B may be a subset of beam set A, or beam sets A and B may be the same.

[0006] Network-side AI / ML is when the network performs AI / ML inference, and UE-side AI / ML is when the UE performs AI / ML inference.

[0007] Monitoring AI / ML inference is called performance monitoring. Non-Patent Document 1 discusses Type 1 performance monitoring and Type 2 performance monitoring.

[0008] Type 1 performance monitoring includes Option 1 NW-side performance monitoring and Option 2 UE-assistant performance monitoring. In NW-side performance monitoring, the UE sends reports to the network and calculates performance metrics on the network side. The UE reports to the network for performance metric calculation. In UE-assistant performance monitoring, the UE calculates performance metrics and reports them to the network, or reports events to the network based on the performance metrics. The UE performs life cycle management (LCM) operations according to instructions from the network.

[0009] Type 2 performance monitoring involves configuration / signaling from the gNB to the UE, and instructions, requests, or reports from the UE to the gNB. In the case of UE-side model monitoring, the UE makes the decision to select, activate, deactivate, switch, or perform fallback operations on the model.

[0010] For performance monitoring of BM-Case 1 and BM-Case 2, the performance metrics include the following four options: ・Key Performance Indicators (KPIs) related to beam prediction accuracy, e.g., Top-1 or Top-K beam prediction accuracy ・Link quality related KPIs, e.g., throughput, L1-RSRP (Layer 1 (L1) Reference Signal Received Power), L1 SINR (Signal-to-Interference-plus-Noise Ratio), virtual BLER (Block Error Rate) ・Performance indicators based on AI / ML input / output data distribution ・L1-RSRP difference, evaluated by comparing measured RSRP with predicted RSRP

[0011] 3rd Generation Partnership Project; Technical Specification Group Radio Access Network; Study on Artificial Intelligence (AI) / Machine Learning (ML) for NR air interface (Release 18), 3GPP TR 38.843 V18.0.0 (2023-12)

[0012] When a communication terminal (UE) supports multiple learning models, if the inference accuracy of a learning model used by the communication terminal among the multiple learning models deteriorates, a scenario is assumed in which the communication terminal switches the learning model to be used. In this case, it is desired that the communication terminal and a base station work together to switch the learning model used by the communication terminal, thereby appropriately selecting a learning model to improve the inference accuracy.

[0013] An object of the present disclosure is to provide a base station, a communication terminal, a control method, a communication method, and a program that contribute to the selection of a learning model in cooperation between a communication terminal and a base station.

[0014] A base station according to the present disclosure includes a transmitter that transmits a first RRC message to a communication terminal inquiring about information regarding at least one candidate learning model to be used as a switching target when the learning model used by the communication terminal to determine a beam to be used for communication between the base station and the communication terminal is switched to another learning model, and a receiver that receives a second RRC message from the communication terminal that includes information regarding the at least one candidate learning model.

[0015] A communication terminal according to the present disclosure includes a receiving unit that receives a first RRC message from a base station inquiring about information regarding at least one candidate learning model to be used as a switching target when switching a learning model used to determine a beam to be used for communication with a base station to another learning model, and a transmitting unit that transmits a second RRC message to the base station that includes information regarding the at least one candidate learning model.

[0016] A control method executed in a base station according to the present disclosure includes, when switching a learning model that a communication terminal uses to determine a beam to be used for communication with a base station to another learning model, sending a first RRC message to the communication terminal inquiring about information on at least one candidate learning model to switch to, and receiving a second RRC message from the communication terminal including information on the at least one candidate learning model.

[0017] A communication method executed in a communication terminal according to the present disclosure includes, when switching a learning model used to determine a beam to be used for communication with a base station to another learning model, receiving a first RRC message from the base station inquiring about information on at least one candidate learning model to switch to, and transmitting a second RRC message to the base station including information on the at least one candidate learning model.

[0018] The program disclosed herein causes a computer to execute the following steps: when switching a learning model that a communication terminal uses to determine a beam to be used for communication with a base station to another learning model, the program sends a first RRC message to the communication terminal inquiring about information about at least one candidate learning model to switch to, and receives a second RRC message from the communication terminal including information about the at least one candidate learning model.

[0019] The present disclosure makes it possible to provide a base station, a communication terminal, a control method, a communication method, and a program that contribute to the selection of a learning model through cooperation between a communication terminal and a base station.

[0020] FIG. 1 shows an example configuration of a base station. FIG. 2 shows an example configuration of a communication terminal. FIG. 3 shows the flow of control processing executed in a base station. FIG. 4 shows the flow of communication processing executed in a communication terminal. FIG. 5 shows a transmission beam used by a gNB for communication with a UE. FIG. 6 shows the flow of processing for detecting deterioration in the accuracy of inference results of a learning model. FIG. 7 shows the flow of transmission processing of UECapability. FIG. 8 shows an example configuration of UE-CapabilityRequestFilterNR. FIG. 9 shows example parameters set in UE-NR-Capability. FIG. 10 shows the flow of processing for switching learning models. FIG. 11 shows CSI-reportconfigIE. FIG. 12 shows CSI-reportconfigIE. FIG. 13 shows the flow of processing for switching learning models. FIG. 14 shows CSI-reportconfigIE transmitted from a gNB to a UE. FIG. 15 shows the flow of processing for switching learning models. FIG. 16 shows the flow of processing for switching learning models. FIG. 17 shows the flow of processing for switching learning models. FIG. 18 shows the flow of processing for switching learning models. FIG. 19 shows the flow of the learning model switching process. FIG. 20 shows the flow of the learning model switching process. FIG. 21 shows the flow of the learning model switching process. FIG. 22 shows the flow of the learning model switching process. FIG. 23 shows the flow of the process for detecting deterioration in the accuracy of the inference result of the learning model. FIG. 24 shows the flow of the learning model switching process. FIG. 25 shows an example of identification information for the learning model. FIG. 26 shows the flow of the learning model switching process. FIG. 27 shows the flow of the UEAssistanceInformation transmission process. FIG. 28 shows an example configuration of OtherConfig. FIG. 29 shows example parameters set in UEAssistanceInformation. FIG. 30 is a block diagram showing example configurations of a base station and a gNB. FIG. 31 is a block diagram showing example configurations of a communication terminal and a UE. FIG. 32 shows an example configuration of a base station.

[0021] (First Embodiment) FIG. 1 illustrates a configuration example of a base station 10. The base station 10 may be a computer device operated by a processor executing a program stored in a memory. The base station 10 may be a base station system including a device for performing wireless communication and a device for performing baseband processing. In other words, the components constituting the base station 10 may be distributed across multiple devices. The multiple devices may be connected via a network. The base station 10 may be, for example, a gNB (g Node B) supporting a wireless communication standard known as 5G (5th Generation) in 3GPP, or an eNB (evolved Node B) supporting a wireless communication standard known as 4G (4th Generation). Alternatively, the base station 10 may be, for example, a base station supporting a wireless communication standard known as 6G (6th Generation).

[0022] The base station 10 includes a transmitter 11 and a receiver 12. The transmitter 11 and receiver 12 may be software or modules that are executed by a processor executing a program stored in a memory, or may be hardware such as a circuit or a chip.

[0023] The transmitter 11 transmits a first RRC (Radio Resource Control) message to the communication terminal 20. The first RRC message is used to inquire about information on at least one candidate learning model that is a candidate for switching to when switching the learning model used by the communication terminal 20 to another learning model. The learning model and the candidate learning model are learning models used by the communication terminal 20 to determine a beam to be used for communication with the base station 10. Determining a beam may mean selecting an optimal beam from multiple beams that can be used when the communication terminal 20 communicates with the base station 10.

[0024] The RRC message is a message generated in accordance with the provisions of the RRC protocol. The learning model may be, for example, a learning model generated by performing supervised learning using training data in the learning phase. Alternatively, the learning model may be a learning model generated by performing unsupervised learning without using training data. Alternatively, the learning model may be a learning model generated by performing reinforcement learning to determine optimal control to be performed by the communication terminal 20.

[0025] The training data used when performing supervised learning may be, for example, data including information about multiple beams used between the base station 10 and the communication terminal 20 and information about an optimal beam. The information about the beam may include, for example, identification information for identifying the beam, quality information about the beam, etc. The quality information about the beam may be information about the received power of a signal transmitted via the beam, the throughput when communication is performed using the beam, etc. The quality information may be referred to as measurement information. The optimal beam may be the beam with the best quality among multiple beams. The training data may be data generated by simulating communication between the base station 10 and the communication terminal 20, or may be data collected when the base station 10 and the communication terminal 20 actually communicate.

[0026] The communication terminal 20 may have multiple learning models and may switch between them. Each learning model may be generated using a different learning method from the other learning models. The learning method may be, for example, supervised learning, unsupervised learning, or reinforcement learning. Alternatively, each learning model may be trained using data different from the data used in the other learning models. The data used in a learning model may be, for example, data when the number of communication terminals communicating with the base station 10 is greater than a predetermined number, or data when the number is less than a predetermined number. Alternatively, the data used in a learning model may be data obtained under different weather conditions when the data was acquired, or data simulated or measured under various other conditions.

[0027] In the inference phase, the learning model receives, for example, information about a plurality of beams used between the base station 10 and the communication terminal 20 as input and outputs information about an optimal beam. The information about the optimal beam may be, for example, information about a ranking of beams arranged in order of optimality, or information that numerically represents the optimality of each beam. Alternatively, the information about the optimal beam may be quality information estimated when each beam is used.

[0028] The switching of the learning model may be performed in the communication terminal 20 based on an instruction from the base station 10 to the communication terminal 20, or may be performed in the communication terminal 20 based on a trigger or event detected autonomously by the communication terminal 20.

[0029] The receiving unit 12 receives a second RRC message including information about at least one candidate learning model from the communication terminal 20. The information about the candidate learning models includes information for identifying the learning model. Furthermore, the information about the candidate learning models may include information about the processing time required to execute each learning model, information about the size of each learning model, information indicating a learning method for each learning model, etc. Without being limited to this, the information about the candidate learning models may include information indicating the characteristics of each learning model, etc.

[0030] Here, FIG. 32 shows an example configuration of the base station 100 including the control unit 13. The base station 100 has a configuration in which the control unit 13 is added to the transmission unit 11 and the reception unit 12 in the base station 10. The control unit 13 may determine a switching destination learning model from at least one candidate learning model. Here, the transmission unit 11 may transmit a message including information about the switching destination learning model determined by the control unit 13 to the communication terminal 20. The information about the switching destination learning model may be transmitted using an RRC message.

[0031] The control unit 13 may determine the learning model to switch to by judging the accuracy or validity of the inference results of each candidate learning model according to a predetermined standard.

[0032] 2 shows an example of the configuration of the communication terminal 20. The communication terminal 20 may be a computer device that operates when a processor executes a program stored in a memory. The communication terminal 20 may be a smartphone terminal, an IoT (Internet of Things) terminal, or the like. The communication terminal 20 may be a UE (User Equipment), which is used as a general term for communication terminals in 3GPP.

[0033] The communication terminal 20 has a receiving unit 21 and a transmitting unit 22. The receiving unit 21 and the transmitting unit 22 may be software or modules that are executed by a processor executing a program stored in a memory, or may be hardware such as a circuit or a chip.

[0034] The receiver 21 receives a first RRC message from the base station 10. The transmitter 22 transmits a second RRC message to the base station 10. The receiver 21 may also receive, from the communication terminal 20, information indicating a learning model selected from at least one candidate learning model.

[0035] 3 shows the flow of control processing executed in the base station 10. First, the transmitter 11 transmits a first RRC message to the communication terminal 20 (S11). The first RRC message is used to inquire about information on at least one candidate learning model that is a candidate for switching to when switching the learning model used by the communication terminal 20 to another learning model. The learning model and the candidate learning model are learning models that the communication terminal 20 uses to determine a beam to be used for communication with the base station 10.

[0036] Next, the receiver 12 receives a second RRC message including information about at least one candidate learning model from the communication terminal 20 (S12).

[0037] 4 shows the flow of communication processing executed in communication terminal 20. First, receiver 21 receives a first RRC message from base station 10 (S21). Next, transmitter 22 transmits a second RRC message to base station 10 (S22).

[0038] As described above, when switching the learning model used by the communication terminal 20 to another learning model, the base station 10 acquires information about candidate learning models that are candidates for the learning model to which the communication terminal 20 is to be switched from the communication terminal 20. Furthermore, the base station 10 selects a learning model to which the communication terminal 20 is to be switched from the candidate learning models. In this way, the base station 10 can determine the learning model to be used in the communication terminal 20 in cooperation with the communication terminal 20. Furthermore, the base station 10 can select a learning model that improves the accuracy of inference by judging the accuracy or validity of the inference result of each candidate learning model according to a predetermined criterion.

[0039] (Embodiment 2) Figure 5 shows transmission beams used by gNB 30 for communication with UE 40. gNB 30 corresponds to base station 10 in Figure 1. UE 40 corresponds to communication terminal 20 in Figure 2. A1 to A16 indicate beams corresponding to CSI-RS (Channel State Information-Reference Signal) (hereinafter sometimes referred to as "CSI-RS beams"). Also, B1 to B4 are shown as subsets of A2, A7, A9, and A16. That is, B1 to B4 also indicate beams corresponding to CSI-RS.

[0040] UE 40 compares the RSRP (Reference Signal Received Power) of different CSI-RS transmitted using different transmission beams, and selects, for example, the transmission beam with the highest RSRP. UE 40 notifies gNB 30 of the selected transmission beam. Also, although FIG. 5 omits the reception beam in UE 40, UE 40 may select a pair of transmission beam and reception beam with the highest RSRP.

[0041] Here, a learning model used by UE 40 to select an appropriate CSI-RS beam will be described. The learning model used to select an appropriate CSI-RS beam may be a learning model that learns, in a learning phase, quality information regarding the CSI-RS beam of Set A, quality information regarding the optimal CSI-RS beam or the optimal CSI-RS beam, quality information regarding the CSI-RS beam of Set B, and quality information regarding the optimal CSI-RS beam or the optimal CSI-RS beam as training data. Furthermore, the training data may include identification information of the CSI-RS beam. The quality information regarding the CSI-RS beam may be, for example, RSRP or SINR (Signal-to-Interference-plus-Noise Ratio). The training data used in the learning phase may be generated by performing a simulation, or may be measurement results in a communication environment that actually has a gNB and a UE.

[0042] Furthermore, the learning model used to select an appropriate CSI-RS beam may be a learning model that inputs quality information about the CSI-RS beams of Set B and outputs an optimal CSI-RS beam in Set A during the inference phase. The learning model used to select an appropriate CSI-RS beam may also be a learning model that outputs quality information about the optimal CSI-RS beam. The optimal CSI-RS beam may be, for example, the CSI-RS beam with the highest RSRP among multiple CSI-RS beams. Alternatively, the learning model used to select an appropriate CSI-RS beam may output a ranking of multiple CSI-RS beams arranged in order from highest quality to lowest quality.

[0043] The UE 40 has a plurality of learning models. The UE 40 may switch between the learning models. For example, when the UE 40 detects that the accuracy of the inference result of the learning model has deteriorated, the UE 40 may switch the learning model to another learning model. The deterioration in the accuracy of the inference result of the learning model may be detected by the UE 40 or by the gNB 30.

[0044] Figure 6 shows the flow of a process for detecting deterioration in the accuracy of the inference results of a learning model. In the explanation of Figure 6, the terms Set A and Set B are used. Set A indicates a set of CSI-RS beams that can be generated by gNB30. For example, Set A indicates the set of beams A1 to A16 shown in Figure 5. Set B indicates a set of CSI-RS beams that can be generated by gNB30. For example, Set B indicates the set of beams B1 to B4 shown in Figure 5.

[0045] Set A resource indicates resources allocated to CSI-RS in each of beams A1 to A16. Set B resource indicates resources allocated to CSI-RS in each of beams B1 to B4. Resources may be indicated using, for example, the frequency domain and the time domain. The frequency domain may be indicated using, for example, subcarriers. The time domain may be indicated using slots, symbols, etc.

[0046] First, the gNB 30 configures the UE 40 to use the Set A resources and to report measurement results using the Set A resources (S31). Specifically, the gNB 30 transmits an RRC message including the Set A resources and information instructing the UE 40 to report measurement results using the Set A resources to the UE 40. Next, the gNB 30 configures the UE 40 to use the Set B resources and to report inference results using the Set B resources to the UE 40 (S32). Specifically, the gNB 30 transmits an RRC message including the Set B resources and information instructing the UE 40 to report inference results using the Set B resources to the UE 40.

[0047] Next, the UE 40 measures the quality of the CSI-RS beams A1 to A16 using resources in Set A (S33). For example, the UE 40 measures the RSRP using the CSI-RS transmitted in each of the beams A1 to A16.

[0048] Next, UE 40 performs inference using Set B resources by inputting the measurement results for the CSI-RS beams B1 to B4 into the currently used learning model (S34). For example, UE 40 measures reception quality using CSI-RS transmitted in the CSI-RS beams B1 to B4. The reception quality may be, for example, RSRP. The currently used learning model may be referred to as an active learning model. The measurement results for the CSI-RS beams B1 to B4 may be information including identification information and quality information for each CSI-RS beam.

[0049] Here, the UE 40 may perform multiple inferences related to the active learning model, i.e., the UE may perform one measurement using the single transmitted Set A signal, and multiple inferences using the multiple transmitted Set B signals.

[0050] Next, UE 40 transmits the quality measurement results for the CSI-RS beams A1 to A16 to gNB 30 (S35). Next, UE 40 transmits the results of the inference performed in the active learning model to gNB 30 (S36). In steps S35 and S36, an RRC message including the quality measurement results or the inference results may be transmitted.

[0051] Next, the gNB 30 performs performance monitoring for the active learning model (S37). The performance monitoring may be performed using, for example, a performance metric.

[0052] The performance metrics may be, for example, beam prediction accuracy related key performance indicators (KPIs) such as Top-1 beam prediction accuracy, Top-K / 1 beam prediction accuracy, and Top-1 / K beam prediction accuracy.

[0053] The Top-1 beam prediction accuracy may be the rate at which the CSI-RS beam with the best quality measured in step S33 matches the optimal CSI-RS beam indicated in the inference performed in step S34. For example, a case will be described where inference using Set B resources is performed five times. In this case, if the CSI-RS beam with the best quality in the measurement results matches the optimal CSI-RS beam indicated in the inference five times, the Top-1 beam prediction accuracy may be indicated as 100%. If they match once, the Top-1 beam prediction accuracy may be indicated as 20%.

[0054] The Top-K / 1 beam prediction accuracy may be the rate at which the CSI-RS beam with the best quality in the measurement results is ranked within K (K is an integer equal to or greater than 1) of the rankings arranged in descending order of quality from the CSI-RS beam with the highest quality indicated in the inference. For example, a case will be described in which inference using Set B resources is performed five times. In this case, if the CSI-RS beam with the best quality in the measurement results is ranked within K five times of the rankings indicated in the inference, the Top-K / 1 beam prediction accuracy may be indicated as 100%.

[0055] The Top-1 / K beam prediction accuracy may be the percentage of the CSI-RS beam that is determined to be optimal in the inference that is ranked within K (K is an integer greater than or equal to 1) in a ranking of CSI-RS beams in order of highest quality in the measurement results.

[0056] Furthermore, the performance metrics may be link quality related KPIs. The link quality related KPIs may be throughput, RSRP, SINR (Signal-to-Interference plus Noise power Ratio), or BLER (Block Error Rate) related to data transmission between the gNB 30 and the UE 40.

[0057] The performance metric may also be a value determined based on the distribution of input and output data in AI / ML, i.e., machine learning.

[0058] The performance metric may also be the difference between the RSRP indicated in the measurement result in step S33 and the RSRP indicated in the inference result in step S34.

[0059] The gNB 30 calculates a performance metric in performance monitoring. Alternatively, the UE 40 may calculate the performance metric. In this case, the gNB 30 acquires the performance metric from the UE 40. Furthermore, the gNB 30 may detect that the accuracy of the inference result of the active learning model has deteriorated when the performance metric falls below a predetermined standard. The case where the performance metric falls below the predetermined standard may be when the performance metric falls below a predetermined quality. If an increase in the value of the performance metric indicates a decrease in quality, the gNB 30 may detect that the accuracy of the inference result of the active learning model has deteriorated when the performance metric exceeds the predetermined standard.

[0060] FIG. 7 shows the flow of the UE Capability transmission process. First, the gNB 30 transmits a UECapabilityEnquiry message, which is an RRC message, to the UE 40 (S41). The gNB 30 may transmit the UECapabilityEnquiry message to the UE 40 when detecting a deterioration in the accuracy of the inference result of the active learning model in the UE 40. Alternatively, the gNB 30 may transmit the UECapabilityEnquiry message to the UE 40 periodically or at any timing. The UECapabilityEnquiry message includes a UE-CapabilityRequestFilterNR. Furthermore, as shown in FIG. 8, the UE-CapabilityRequestFilterNR includes a parameter or field called AI / MLRequest-r19. r19 may be updated as appropriate to a different value or a different notation.

[0061] AI / MLRequest-r19 is used to request transmission of information related to a learning model managed in UE 40. Alternatively, AI / MLRequest-r19 may be rephrased as being used to request transmission of UE Capability related to the learning model of UE 40. Managing a learning model may be rephrased as maintaining, recording, saving, etc. the learning model.

[0062] Next, when the UE 40 receives a UECapabilityEnquiry message including an AI / MLRequest-r19, the UE 40 transmits a UECapabilityInformation message, which is an RRC message, to the gNB 30 (S42). The UECapabilityInformation message includes a UE-NR-Capability. Furthermore, as shown in FIG. 9, the UE-NR-Capability includes a parameter or field called AIML-Parameters-r19. r19 may be updated as appropriate to a different value or representation. The AIML-Parameters-r19 includes candidatemodebitmap-r19, monitoring-measurement-time-r19, and monitoring-metric-r19.

[0063] The candidatemodebitmap-r19 is used to indicate at least one learning model available to the UE 40. The at least one learning model available to the UE 40 is a learning model managed by the UE 40. Of the learning models available to the UE 40, learning models excluding the active learning model may be referred to as candidate learning models. A candidate learning model is a learning model that is a candidate for switching to the active learning model when switching the active learning model. The candidatemodebitmap-r19 may indicate only the candidate learning models, or may indicate the active learning model and the candidate learning models. In FIG. 9, for example, among the learning models #0 to #7, learning models #5 and #6 are indicated as candidate learning models.

[0064] Monitoring-measurement-time-r19 is used to indicate the processing time of the candidate learning model indicated in candidatemodebitmap-r19. The processing time of the candidate learning model may be the time from when information is input to the candidate learning model until the candidate learning model outputs an inference result. When multiple candidate learning models are indicated in candidatemodebitmap-r19, monitoring-measurement-time-r19 indicates the processing time for each candidate learning model. In FIG. 9, the processing time for learning model #5 is indicated as N1, and the processing time for learning model #6 is indicated as N2. N1 and N2 indicate numerical values ​​of the processing time.

[0065] Monitoring-metric-r19 is used to indicate a performance metric that can be calculated when the UE 40 calculates the performance metric. In other words, monitoring-metric-r19 indicates a performance metric that the UE 40 supports.

[0066] 10 shows the flow of the learning model switching process. First, the gNB 30 configures the UE 40 with Set A resources and the ability to report measurement results using the Set A resources (S51). The gNB 30 may periodically execute the process of step S51. Alternatively, the gNB 30 may execute the process of step S51 in response to the reception of UECapabilityInformation in step S42 of FIG. 7. Alternatively, the process of step S51 may be omitted because it has already been executed in step S31 of FIG. 6.

[0067] Next, gNB30 configures UE40 to report inference results using candidate learning model #5 (S52). Furthermore, gNB30 configures UE40 to report inference results using candidate learning model #6 (S53). In other words, gNB30 sends a message to UE40 indicating that it will report inference results using candidate learning models #5 and #6.

[0068] Candidate learning models #5 and #6 are learning models indicated as candidate learning models in the UECapabilityInformation. Alternatively, the gNB 30 may determine a candidate learning model for which it requests the UE 40 to report an inference result based on the processing time of the candidate learning model. For example, the gNB 30 may determine a candidate learning model whose processing time is shorter than a predetermined standard as a candidate learning model for which it requests the UE 40 to report an inference result. In this case, the gNB 30 may configure the UE 40 to report the inference result using the determined candidate learning model.

[0069] In steps S52 and S53, for example, the gNB 30 transmits a message including a CSI-reportconfig IE (Information Element) to the UE 40. Here, FIG. 11 shows the CSI-reportconfig IE transmitted in step S52. The CSI-reportconfig IE shown in FIG. 11 indicates that reportConfigid is 0. The identification information of resourcesForChannelMeasurement, 0, indicates Set B resources. The identification information of candidatemodelsForMonitoring, 5, indicates candidate learning model #5. In other words, FIG. 11 shows that inference of candidate learning model #5 is performed using Set B resources.

[0070] Fig. 12 shows the CSI-reportconfig IE transmitted in step S53. The CSI-reportconfig IE shown in Fig. 12 indicates that reportConfigid is 1. The identification information of resourcesForChannelMeasurement, 0, indicates Set B resources. The identification information of candidatemodelsForMonitoring, 6, indicates candidate learning model #6. In other words, Fig. 12 shows that inference of candidate learning model #6 is performed using Set B resources.

[0071] 10, UE 40 measures the quality of the CSI-RS beams A1 to A16 using resources in Set A (S54). For example, UE 40 measures the RSRP using the CSI-RS transmitted in each of the beams A1 to A16.

[0072] Next, UE 40 performs inference by inputting the quality measurement results for the CSI-RS beams B1 to B4 into candidate learning model #5 using Set B resources (S55).Next, UE 40 performs inference by inputting the quality measurement results for the CSI-RS beams B1 to B4 into candidate learning model #6 using Set B resources (S56).

[0073] Next, UE 40 transmits the results of measuring the quality of the CSI-RS beams A1 to A16 to gNB 30 (S57). Next, UE 40 transmits the results of the inference performed in candidate learning model #5 to gNB 30 (S58). Next, UE 40 transmits the results of the inference performed in candidate learning model #6 to gNB 30 (S59).

[0074] Since the processing time using candidate learning model #5 in UE 40 is N1, it takes N1 or more time from when gNB 30 transmits the CSI-reportconfig IE in step S52 until UE 40 transmits the inference result to gNB 30 in step S58. Also, the processing time using candidate learning model #6 in UE 40 is N2. Therefore, it takes N2 or more time from when gNB 30 transmits the CSI-reportconfig IE in step S53 until UE 40 transmits the inference result to gNB 30 in step S59. For example, gNB 30 allocates resources to UE 40 for transmitting the messages of steps S58 and S59. In this case, gNB 30 may allocate resources at a time when N1 or N2 or more have elapsed since transmitting the CSI-reportconfig IE in steps S52 and S53 as resources for transmitting the messages of steps S58 and S59.

[0075] Next, the gNB 30 performs performance monitoring for the candidate learning models #5 and #6 (S60). For example, the gNB 30 may calculate KPIs related to beam prediction accuracy as performance metrics. The gNB 30 compares the Top-1 beam prediction accuracy, Top-K / 1 beam prediction accuracy, and Top-1 / K beam prediction accuracy calculated as performance metrics for each of the candidate learning models #5 and #6. The gNB 30 may determine the candidate learning model with the larger performance metric as the learning model to switch from the active learning model as a result of the comparison. Alternatively, the gNB 30 may determine not to switch the active learning model if the values ​​calculated as performance metrics for each of the candidate learning models #5 and #6 do not exceed a predetermined value.

[0076] Next, the gNB 30 transmits an LCM indication message including identification information of the candidate learning model to be switched to the UE 40 (S61). If the gNB 30 determines not to switch the active learning model, the gNB 30 may transmit an LCM indication message including identification information of the active learning model, or may not transmit an LCM indication message. The LCM indication message may be an RRC message.

[0077] Fig. 13 shows a modified example of the flow of the learning model switching process in Fig. 10. Step S71 in Fig. 13 is the same as step S51 in Fig. 10, and therefore detailed description thereof will be omitted.

[0078] Next, the gNB 30 configures the UE 40 to report the inference results using the candidate learning models #5 and #6 (S72). Here, FIG. 14 shows the CSI-reportconfig IE transmitted from the gNB 30 to the UE 40 in step S72. The CSI-reportconfig IE shown in FIG. 14 indicates that the reportConfigid is 0. The identification information of resourcesForChannelMeasurement, 0, indicates Set B resources. The candidate learning models #5 and #6 are indicated in candidatemodelsForMonitoring. In other words, FIG. 14 shows that inference of the candidate learning models #5 and #6 is performed using Set B resources.

[0079] Returning to FIG. 13, steps S73 to S76 are similar to steps S54 to S57 in FIG. 10, and therefore detailed description thereof will be omitted.

[0080] Next, UE40 transmits the results of the inferences performed in candidate learning models #5 and #6 to gNB30 (S77). Here, UE40 may include the results of the inferences performed in candidate learning models #5 and #6 in a single message transmitted to gNB30.

[0081] Steps S78 and S79 are similar to steps S60 and S61 in FIG. 10, and therefore detailed description thereof will be omitted.

[0082] FIG. 15 shows a modified example of the flow of the learning model switching process in FIGS. 10 and 13. First, UE 40 configures UE 40 to report Set A resources, quality measurement results using Set A resources, and inference results using candidate learning models #5 and #6 (S81). In step S81, gNB 30 transmits a message including a CSI-reportconfig IE (Information Element) to UE 40. For example, the CSI-reportconfig IE transmitted from gNB 30 to UE 40 may be the same as that shown in FIG. 14. Here, in step S81, Set A resources and Set B resources may be indicated in resourcesForChannelMeasurement, whose identification information is set to 0 in FIG. 14. The inclusion of Set A resources in resourcesForChannelMeasurement may indicate that measurement results using Set A resources will be reported.

[0083] Steps S82 to S84 are similar to steps S73 to S75 in FIG. 13, and therefore detailed description thereof will be omitted.

[0084] Next, UE 40 transmits the results of measuring the quality of the CSI-RS beams A1 to A16 and the results of the inference performed in the candidate learning models #5 and #6 to gNB 30 (S85). Here, UE 40 may include the results of measuring the quality of the CSI-RS beams A1 to A16 and the results of the inference performed in the candidate learning models #5 and #6 in a single message transmitted to gNB 30.

[0085] Steps S86 and S87 are similar to steps S60 and S61 in FIG. 10, and therefore detailed description thereof will be omitted.

[0086] As described above, gNB30 calculates a performance metric using the results of measuring the quality of CSI-RS beams A1 to A16 in UE40 and the results of inference performed in candidate learning models #5 and #6. Furthermore, gNB30 uses the calculated performance metric to determine the learning model to which the active learning model used in UE40 will be switched. As a result, if degradation occurs in the inference results of the active learning model used in UE40, gNB30 can determine a learning model that is expected to improve the accuracy of the inference results. As a result, UE40 can continue to use a learning model that can maintain or improve quality by switching the active learning model to a learning model that is expected to improve the accuracy of the inference results.

[0087] Third Embodiment Fig. 16 shows the flow of the learning model switching process. Fig. 16 illustrates an example in which the Set A resources and the Set B resources used when calculating the performance metrics for candidate learning models #5 and #6 are different.

[0088] Specifically, when calculating performance metrics for candidate learning model #5, Set A1 resources and Set B1 resources are used. Furthermore, when calculating performance metrics for candidate learning model #6, Set A2 resources and Set B2 resources are used. Set A1 resources and Set A2 resources are used to measure the quality of CSI-RS beams A1 to A16.

[0089] First, the UE 40 configures the UE 40 to report the Set A1 resource and the quality measurement result using the Set A1 resource, and to report the Set B1 resource and the inference result using the Set B1 resource and candidate learning model #5 (S91). In step S91, the gNB 30 transmits a message including a CSI-reportconfig IE (Information Element) to the UE 40. For example, the CSI-reportconfig IE transmitted from the gNB 30 to the UE 40 may be the same as that shown in FIG. 14. Here, in step S91, the Set A1 resource and the Set B1 resource may be indicated in resourcesForChannelMeasurement, whose identification information is set to 0 in FIG. 14. The inclusion of the Set A1 resource in resourcesForChannelMeasurement may indicate that the measurement result using the Set A1 resource will be reported.

[0090] Next, UE 40 configures UE 40 to report Set A2 resources and measurement results using the Set A2 resources, and to report Set B2 resources and inference results using Set B2 resources and candidate learning model #6 (S92). For example, the CSI-reportconfig IE transmitted from gNB 30 to UE 40 may include resourcesForChannelMeasurement whose identification information is set to 1. The Set A2 resources and Set B2 resources may be indicated in resourcesForChannelMeasurement whose identification information is set to 1. The inclusion of Set A2 resources in resourcesForChannelMeasurement may indicate that measurement results using Set A2 resources will be reported.

[0091] Steps S93 to S96 are basically the same as steps S54 to S56 in Fig. 10. While Fig. 10 shows that measurement is performed using Set A resources in step S54, Fig. 16 shows that measurement is performed using Set A1 resources in step S93. Furthermore, Fig. 16 shows that measurement is performed using Set A2 resources in step S94.

[0092] Furthermore, steps S97 to S100 are basically the same as steps S57 to S59 in Fig. 10. While Fig. 10 shows that measurement results using Set A resources are transmitted in step S57, Fig. 16 shows that measurement results using Set A1 resources are transmitted in step S97. Furthermore, Fig. 16 shows that measurement results using Set A2 resources are transmitted in step S98.

[0093] Steps S101 and S102 are similar to steps S60 and S61 in FIG. 10, and therefore detailed description thereof will be omitted.

[0094] Figure 17 shows a modified example of the learning model switching process in Figure 16. Steps S111 to S116 are similar to steps S91 to S96 in Figure 16, and therefore detailed description will be omitted. In Figure 17, step S117 executes the processes of steps S97 and S99 in Figure 16. Furthermore, step S118 executes the processes of steps S98 and S100 in Figure 16. In other words, steps S117 and S118 indicate that the measurement result and the inference result are transmitted at the same time and in the same message.

[0095] Steps S119 and S120 are similar to steps S101 and S102 in FIG. 16, and therefore detailed description thereof will be omitted.

[0096] As described above, even when different Set A resources and Set B resources are used for each learning model, gNB30 can determine an appropriate learning model to be used in UE40 based on the performance metric for each candidate learning model.

[0097] (Fourth embodiment) Fig. 18 shows the flow of the learning model switching process. Steps S121 to S126 in Fig. 18 are the same as steps S51 to S56 in Fig. 10, and therefore detailed description thereof will be omitted.

[0098] After step S126, UE 40 transmits to gNB 30 the performance metric calculated using the quality measurement results using Set A resources and the inference results of candidate learning models #5 and #6 (S127). Regarding what value to calculate as the performance metric, for example, gNB 30 may notify the performance metric to be calculated in steps S122 and S123. Alternatively, gNB 30 may notify the performance metric to be calculated in step S121.

[0099] The UE 40 calculates performance metrics for each of the candidate learning models #5 and #6. The UE 40 may calculate, for example, KPIs related to beam prediction accuracy as the performance metrics.

[0100] Steps S128 and S129 are similar to steps S58 and S59 in FIG. 10, and therefore detailed description thereof will be omitted.

[0101] Next, gNB30 performs performance monitoring for candidate learning models #5 and #6 (S130). gNB30 may compare the performance metrics for candidate learning models #5 and #6 and decide to switch from the active learning model to the candidate learning model with the larger performance metric as the learning model to switch to. Alternatively, gNB30 may decide not to switch the active learning model if the values ​​of the performance metrics for each of candidate learning models #5 and #6 do not exceed a predetermined value.

[0102] Step S131 is similar to step S61 in FIG. 10, and therefore a detailed description thereof will be omitted.

[0103] Figure 19 shows a modified example of the flow of the learning model switching process in Figure 18. Steps S141 to S145 are similar to steps S71 to S75 in Figure 13, so detailed explanations will be omitted. Also, steps S146 to S149 are similar to steps S127 to S131 in Figure 18. However, in Figure 19, in step S147, the inference results of candidate learning models #5 and #6 are sent using a single message.

[0104] Figure 20 shows a modified example of the flow of the learning model switching process in Figure 18. Steps S151 to S154 are similar to steps S81 to S84 in Figure 15, so detailed explanations will be omitted. Also, steps S155 to S158 are similar to steps S146 to S149 in Figure 19.

[0105] As described above, gNB30 can determine an appropriate learning model to be used in UE40 based on the performance metrics for each candidate learning model calculated in UE40.

[0106] Fifth Embodiment Fig. 21 shows the flow of learning model switching processing. Steps S161 to S166 in Fig. 21 are similar to steps S91 to S96 in Fig. 16, and therefore detailed description thereof will be omitted.

[0107] After step S166, UE 40 transmits to gNB 30 a performance metric calculated using the quality measurement results using Set A1 resources and the inference results of candidate learning model #5 (S167). Furthermore, UE 40 transmits to gNB 30 a performance metric calculated using the quality measurement results using Set A2 resources and the inference results of candidate learning model #6 (S168).

[0108] Steps S169 and S170 are similar to steps S99 and S100 in Fig. 16, and therefore detailed description thereof will be omitted. Also, steps S171 and S172 are similar to steps S130 and S131 in Fig. 18, and therefore detailed description thereof will be omitted.

[0109] Fig. 22 shows a modification of the learning model switching process in Fig. 21. Steps S181 to S186 in Fig. 22 are similar to steps S161 to S166 in Fig. 21, and therefore detailed description thereof will be omitted.

[0110] Next, UE 40 transmits to gNB 30 the results of measuring the quality of CSI-RS beams A1 to A16 using Set A1 resources and the results of inference performed in candidate learning model #5 using Set B1 resources (S187). Here, UE 40 may include the results of measuring the quality of CSI-RS beams A1 to A16 using Set A1 resources and the results of inference performed in candidate learning model #5 using Set B1 resources in a single message transmitted to gNB 30.

[0111] Next, UE 40 transmits to gNB 30 the results of measuring the quality of the CSI-RS beams A1 to A16 using the Set A2 resources and the results of the inference performed in candidate learning model #6 using the Set B2 resources (S188). Here, UE 40 may include the results of measuring the quality of the CSI-RS beams A1 to A16 using the Set A2 resources and the results of the inference performed in candidate learning model #6 using the Set B2 resources in a single message transmitted to gNB 30.

[0112] Steps S189 and S190 are similar to steps S171 and S172 in FIG. 21, and therefore detailed description thereof will be omitted.

[0113] As described above, even when different Set A resources and Set B resources are used for each learning model, gNB30 can determine an appropriate learning model to be used in UE40 based on the performance metric for each candidate learning model.

[0114] (Embodiment 6) Figure 23 shows the flow of a process for detecting deterioration in the accuracy of the inference result of a learning model. Steps S191 to S194 are similar to steps S31 to S34 in Figure 6, so detailed description will be omitted. Here, in Figure 6, an example is shown in which gNB 30 performs performance monitoring in step S37, but in Figure 23, UE 40 performs performance monitoring in step S195. By performing performance monitoring, UE 40 detects whether deterioration has occurred in the accuracy of the inference result of the active learning model.

[0115] 24 shows the flow of the learning model switching process. First, the UE 40 transmits a scheduling request to the gNB 30 (S201). When resources for transmitting a PUSCH (Physical Uplink Shared Channel) are not allocated by the gNB 30, the UE 40 transmits the scheduling request to the gNB 30. For example, the UE 40 transmits the scheduling request using a PUCCH (Physical Uplink Control Channel).

[0116] Next, the gNB 30 transmits to the UE 40 an uplink grant (UL grant) including information on the PUSCH allocation resource for the UE 40 (S202). Here, if the resource for transmitting the PUSCH has already been allocated to the UE 40, steps S201 and S202 may be omitted.

[0117] Next, the UE 40 transmits a MAC (Medium Access Control) CE (Control Element) indicating at least one candidate learning model using resources allocated for transmitting the PUSCH (S203). The candidate learning model may be indicated, for example, in the format shown in FIG. 25. FIG. 25 shows that identification information for one learning model is indicated in one octet. FIG. 25 shows that N (N is an integer equal to or greater than 1) candidate learning models are set. The identification information for the learning model may be referred to as a model ID. For example, the identification information for the learning model may be indicated in a 6-bit area indicated as D1. Furthermore, information regarding the periodicity when resources used in performing performance monitoring are notified from the gNB 30 to the UE 40 may be indicated in a 2-bit area indicated as D2. The information regarding the periodicity may be, for example, information identifying whether the resource is notified periodically, aperiodically, or periodically at a predetermined period.

[0118] Returning to FIG. 24, the gNB 30 selects a candidate learning model to be subjected to performance monitoring from among the candidate learning models. The gNB 30 transmits a PDCCH (Physical Downlink Control Channel) including DCI (Downlink Control Information) indicating the selected candidate learning model to the UE 40 (S204). The gNB 30 transmits Set A resources and Set B resources to the UE 40 together with the candidate learning model to be subjected to performance monitoring. In step S204, the gNB 30 selects, for example, candidate learning models #0 and #1 as the candidate learning models to be subjected to performance monitoring.

[0119] Next, UE 40 performs performance monitoring for candidate learning models #0 and #1 (S205). Next, UE 40 selects a learning model to which the active learning model will be switched based on the results of the performance monitoring (S206). UE 40 may transmit information indicating the selected learning model to gNB 30. At this time, UE 40 may transmit a MAC CE indicating the selected learning model to gNB 30.

[0120] Fig. 26 shows a modified example of the flow of the learning model switching process of Fig. 24. Steps S211 to S213 are similar to steps S201 to S203 of Fig. 24, and therefore detailed description thereof will be omitted.

[0121] Next, the gNB 30 transmits to the UE 40 a PDCCH including DCI indicating the candidate learning model to be subjected to performance monitoring and the Set A1 resource and the Set B1 resource to be used for calculating the performance metric (S214). Here, it is assumed that the gNB 30 selects the candidate learning model #0 as the candidate learning model to be subjected to performance monitoring.

[0122] Next, the UE 40 performs performance monitoring for the candidate learning model #0 (S215).

[0123] Next, the gNB 30 transmits to the UE 40 a PDCCH including DCI indicating the candidate learning model to be subjected to performance monitoring and the Set A2 resource and the Set B2 resource to be used for calculating the performance metric (S216). Here, it is assumed that the gNB 30 selects the candidate learning model #1 as the candidate learning model to be subjected to performance monitoring.

[0124] Next, UE 40 performs performance monitoring for candidate learning model #1 (S217).

[0125] Next, UE 40 selects a learning model to which the active learning model is switched based on the results of the performance monitoring in steps S215 and S217 (S218). UE 40 may transmit information indicating the selected learning model to gNB 30. At this time, UE 40 may transmit a MAC CE indicating the selected learning model to gNB 30.

[0126] As described above, the gNB 30 and the UE 40 can perform the learning model switching process using a physical channel. This allows the learning model switching process in the physical layer, which corresponds to the lower layer, to be performed in a shorter time than the learning model switching process in the upper layer.

[0127] Seventh Embodiment Fig. 27 shows the flow of a process for transmitting UEAssistanceInformation, which may be performed instead of the process for transmitting UECapabilityInformation in Fig. 5 .

[0128] First, RRC Reconfiguration is performed between the UE 40 and the gNB 30 (S221). When a deterioration in the accuracy of the inference result of the active learning model in the UE 40 is detected, the RRC Reconfiguration may be performed. In the RRC Reconfiguration, the UE 40 receives OtherConfig including candidatemodelsReporting, as shown in FIG. 28 .

[0129] candidatemodelsReporting is used to request transmission of information about the learning models managed in UE 40.

[0130] Next, when the UE 40 receives the candidatemodelsReporting, the UE 40 transmits a UEAssistanceInformation message, which is an RRC message, to the gNB 30 (S222). The UEAssistanceInformation message includes a parameter or field, candidatemonitoring-Preference-r19, as shown in FIG. 29. The candidatemonitoring-Preference-r19 includes candidatemodebitmap-r19, monitoring-measurement-time-r19, and monitoring-metric-r19.

[0131] As described above, UE40 can transmit information regarding candidate learning models to gNB30 by transmitting UEAssistanceInformation instead of transmitting UECapabilityInformation.

[0132] FIG. 30 is a block diagram showing an example configuration of a base station 10 and a gNB 30 (hereinafter referred to as the base station 10, etc.). Referring to FIG. 30, the base station 10, etc. includes an RF transceiver 1001, a network interface 1003, a processor 1004, and a memory 1005. The RF transceiver 1001 performs analog RF signal processing to communicate with UEs. The RF transceiver 1001 may include multiple transceivers. The RF transceiver 1001 is coupled to an antenna 1002 and a processor 1004. The RF transceiver 1001 receives modulation symbol data (or OFDM symbol data) from the processor 1004, generates a transmit RF signal, and provides the transmit RF signal to the antenna 1002. The RF transceiver 1001 also generates a baseband receive signal based on the receive RF signal received by the antenna 1002 and provides the baseband receive signal to the processor 1004.

[0133] The network interface 1003 is used to communicate with network nodes (e.g., other core network nodes) and may include, for example, a network interface card (NIC) compliant with the IEEE 802.3 series.

[0134] The processor 1004 performs data plane processing and control plane processing, including digital baseband signal processing for wireless communication.

[0135] The processor 1004 may include multiple processors. For example, the processor 1004 may include a modem processor (e.g., DSP) that performs digital baseband signal processing and a protocol stack processor (e.g., CPU or MPU) that performs control plane processing.

[0136] The memory 1005 is configured by a combination of volatile memory and non-volatile memory. The memory 1005 may include multiple physically independent memory devices. The volatile memory may be, for example, static random access memory (SRAM), dynamic RAM (DRAM), or a combination thereof. The non-volatile memory may be mask read only memory (MROM), electrically erasable programmable ROM (EEPROM), flash memory, a hard disk drive, or any combination thereof. The memory 1005 may include storage located remotely from the processor 1004. In this case, the processor 1004 may access the memory 1005 via the network interface 1003 or an I / O interface (not shown).

[0137] The memory 1005 may store software modules (computer programs) including instructions and data for performing processing by the base station 10, etc., described in the above-described embodiments. In some implementations, the processor 1004 may be configured to read and execute the software modules from the memory 1005, thereby performing processing by the base station 10, etc., described in the above-described embodiments.

[0138] FIG. 31 is a block diagram showing an example configuration of a communication terminal 20 and a UE 40 (hereinafter referred to as communication terminal 20, etc.). A radio frequency (RF) transceiver 1101 performs analog RF signal processing to communicate with a gNB 40. The analog RF signal processing performed by the RF transceiver 1101 includes frequency up-conversion, frequency down-conversion, and amplification. The RF transceiver 1101 is coupled to an antenna 1102 and a baseband processor 1103. That is, the RF transceiver 1101 receives modulation symbol data (or OFDM symbol data) from the baseband processor 1103, generates a transmit RF signal, and supplies the transmit RF signal to the antenna 1102. The RF transceiver 1101 also generates a baseband receive signal based on the receive RF signal received by the antenna 1102 and supplies the baseband processor 1103.

[0139] The baseband processor 1103 performs digital baseband signal processing (data plane processing) and control plane processing for wireless communication. Digital baseband signal processing includes (a) data compression / decompression, (b) data segmentation / concatenation, (c) transmission format (transmission frame) generation / decomposition, (d) transmission path coding / decoding, (e) modulation (symbol mapping) / demodulation, and (f) generation of OFDM symbol data (baseband OFDM signal) using Inverse Fast Fourier Transform (IFFT). Meanwhile, control plane processing includes communication management for Layer 1, Layer 2, and Layer 3.

[0140] The baseband processor 1103 may include a modem processor (e.g., a Digital Signal Processor (DSP)) that performs digital baseband signal processing and a protocol stack processor (e.g., a Central Processing Unit (CPU) or a Micro Processing Unit (MPU)) that performs control plane processing. In this case, the protocol stack processor that performs control plane processing may be shared with the application processor 1104, which will be described later.

[0141] The application processor 1104 is also referred to as a CPU, MPU, microprocessor, or processor core. The application processor 1104 may include multiple processors (multiple processor cores). The application processor 1104 executes a system software program (operating system (OS)) and various application programs (e.g., a call application, a web browser, a mailer, a camera operation application, and a music playback application) read from the memory 1106 or a memory not shown, thereby realizing various functions of the communication terminal 20, etc.

[0142] In some implementations, the baseband processor 1103 and the application processor 1104 may be integrated on a single chip, as indicated by the dashed line (1105) in Figure 31. In other words, the baseband processor 1103 and the application processor 1104 may be implemented as a single System on Chip (SoC) device 1105. An SoC device may also be called a system Large Scale Integration (LSI) or chipset.

[0143] The memory 1106 is volatile memory, nonvolatile memory, or a combination thereof. The memory 1106 may include multiple physically independent memory devices. Volatile memory is, for example, static random access memory (SRAM), dynamic RAM (DRAM), or a combination thereof. Nonvolatile memory is, for example, mask read only memory (MROM), electrically erasable programmable ROM (EEPROM), flash memory, or a hard disk drive, or any combination thereof. For example, the memory 1106 may include an external memory device accessible from the baseband processor 1103, the application processor 1104, and the SoC 1105. The memory 1106 may also include an internal memory device integrated within the baseband processor 1103, the application processor 1104, or the SoC 1105. Furthermore, the memory 1106 may include memory within a Universal Integrated Circuit Card (UICC).

[0144] The memory 1106 may store a software module (computer program) including instructions and data for performing processing by the communication terminal 20, etc., described in the above-described embodiments. In some implementations, the baseband processor 1103 or the application processor 1104 may be configured to read and execute the software module from the memory 1106, thereby performing processing by the communication terminal 20, etc., described in the above-described embodiments.

[0145] In the above examples, the program includes instructions (or software code) that, when loaded into a computer, cause the computer to perform one or more functions described in the embodiments. The program may be stored on a non-transitory computer-readable medium or a tangible storage medium. By way of example and not limitation, computer-readable medium or tangible storage medium includes random-access memory (RAM), read-only memory (ROM), flash memory, solid-state drive (SSD) or other memory technology, CD-ROM, digital versatile disc (DVD), Blu-ray disc or other optical disk storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage device. The program may also be transmitted on a transitory computer-readable medium or communication medium. By way of example and not limitation, transitory computer-readable medium or communication medium includes electrical, optical, acoustic, or other forms of propagated signals.

[0146] Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each embodiment can be combined with other embodiments as appropriate.

[0147] Each drawing is merely an example for describing one or more embodiments. Each drawing may not relate to only one particular embodiment, but may also relate to one or more other embodiments. As will be understood by those skilled in the art, various features or steps described with reference to any one drawing can be combined with features or steps shown in one or more other drawings to create, for example, an embodiment not explicitly shown or described. Not all features or steps shown in any one drawing are necessary to describe an exemplary embodiment, and some features or steps may be omitted. The order of steps described in any drawing may be changed as appropriate.

[0148] Some or all of the above embodiments can be described as, but are not limited to, the following supplementary notes. (Supplementary Note 1) A base station comprising: a transmitter that, when switching a learning model used by a communication terminal to determine a beam to be used in communication with a base station to another learning model, transmits a first RRC message to the communication terminal inquiring about information on at least one candidate learning model to be switched to; and a receiver that receives a second RRC message from the communication terminal including information about the at least one candidate learning model. (Supplementary Note 2) The base station according to Supplementary Note 1, further comprising a controller that determines a learning model to be switched to from the at least one candidate learning model. (Supplementary Note 3) The base station according to Supplementary Note 2, wherein the controller determines the learning model to be switched to based on first quality information on a plurality of candidate beams to be used in communication between the communication terminal and the base station and an output result of the at least one candidate learning model. (Supplementary Note 4) The base station according to Supplementary Note 3, wherein the transmitter transmits, to the communication terminal, resource information indicating resources to be used by the communication terminal when measuring the first quality information. (Supplementary Note 5) The base station according to Supplementary Note 3 or 4, wherein the control unit determines the learning model to switch to based on a comparison result between a first beam determined based on the first quality information and a second beam included in the output result. (Supplementary Note 6) The base station according to Supplementary Note 5, wherein the receiving unit receives information indicating the comparison result from the communication terminal. (Supplementary Note 7) The base station according to Supplementary Note 3 or 4, wherein the control unit determines the learning model to switch to based on a comparison result between the first quality information and second quality information related to the second beam included in the output result. (Supplementary Note 8) The base station according to Supplementary Note 7, wherein the receiving unit receives information indicating the comparison result from the communication terminal.(Supplementary Note 9) A communication terminal comprising: a receiver that, when switching a learning model used to determine a beam to be used for communication with a base station to another learning model, receives from the base station a first RRC message inquiring about information on at least one candidate learning model to be switched to; and a transmitter that transmits to the base station a second RRC message including information on the at least one candidate learning model. (Supplementary Note 10) The communication terminal according to Supplementary Note 9, wherein the receiver receives from the base station information indicating a learning model selected from the at least one candidate learning model. (Supplementary Note 11) A communication terminal comprising: a transmitter that, when switching a learning model used to determine a beam to be used for communication with a base station to another learning model, transmits to the base station a MAC CE (Media Access Control Control Element) including information on at least one candidate learning model to be switched to. (Supplementary Note 12) A control method executed in a base station, which, when switching a learning model used by a communication terminal to determine a beam to be used in communication with a base station to another learning model, transmits to the communication terminal a first RRC message inquiring about information on at least one candidate learning model to be switched to, and receives from the communication terminal a second RRC message including information on the at least one candidate learning model. (Supplementary Note 13) A communication method executed in a communication terminal, which, when switching a learning model used to determine a beam to be used in communication with a base station to another learning model, receives from the base station a first RRC message inquiring about information on at least one candidate learning model to be switched to, and transmits to the base station a second RRC message including information on the at least one candidate learning model. (Supplementary Note 14) A communication method executed in a communication terminal, when switching a learning model used to determine a beam to be used in communication with a base station to another learning model, transmits to the base station a MAC CE (Media Access Control Control Element) including information on at least one candidate learning model to be switched to.(Supplementary Note 15) A program that causes a computer to execute the following steps: when switching a learning model that a communication terminal uses to determine a beam to be used for communication with a base station to another learning model, sending to the communication terminal a first RRC message inquiring about information on at least one candidate learning model to be switched to, and receiving from the communication terminal a second RRC message including information on the at least one candidate learning model. (Supplementary Note 16) A program that causes a computer to execute the following steps: when switching a learning model that a communication terminal uses to determine a beam to be used for communication with a base station to another learning model, receiving from the base station a first RRC message inquiring about information on at least one candidate learning model to be switched to, and sending to the base station a second RRC message including information on the at least one candidate learning model. (Supplementary Note 17) A program that causes a computer to execute the following steps: when switching a learning model used to determine a beam to be used for communication with a base station to another learning model, the program transmits to the base station a MAC CE (Media Access Control Control Element) that includes information about at least one candidate learning model to be switched to.

[0149] Some or all of the elements (e.g., configurations and functions) described in Supplementary Notes 2 to 8 that are dependent on Supplementary Note 1 may also be dependent on Supplementary Notes 12 and 15 in the same dependency relationship as Supplementary Notes 2 to 8. Some or all of the elements described in any Supplementary Note may be applied to various hardware, software, recording means for recording software, systems, and methods.

[0150] Although the present invention has been described above with reference to the embodiments, the present invention is not limited to the above. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the invention.

[0151] This application claims priority based on Japanese Patent Application No. 2024-127795, filed August 2, 2024, the disclosure of which is incorporated herein by reference in its entirety.

[0152] REFERENCE SIGNS LIST 10 Base station 11 Transmitter 12 Receiver 13 Control unit 20 Communication terminal 21 Receiver 22 Transmitter 30 gNB 40 UE 100 Base station

Claims

1. A base station comprising: a transmitting means for transmitting a first RRC message to a communication terminal inquiring about information on at least one candidate learning model to be used as a switching target when the learning model used by the communication terminal to determine a beam to be used for communication between the communication terminal and a base station is switched to another learning model; and a receiving means for receiving a second RRC message from the communication terminal including information on the at least one candidate learning model.

2. The base station according to claim 1, further comprising control means for determining a learning model to switch to from among said at least one candidate learning model.

3. The base station described in claim 2, wherein the control means determines the learning model to switch to based on first quality information regarding multiple candidate beams that are candidates for use in communication between the communication terminal and the base station and the output result of at least one candidate learning model.

4. The base station according to claim 3, wherein said transmitting means transmits to said communication terminal resource information indicating resources used by said communication terminal when said first quality information is measured.

5. A base station as described in claim 3 or 4, wherein the control means determines the learning model to be switched to based on the comparison result between a first beam determined based on the first quality information and a second beam included in the output result.

6. The base station according to claim 5, wherein said receiving means receives information indicating said comparison result from said communication terminal.

7. A base station as described in claim 3 or 4, wherein the control means determines the learning model to switch to based on the comparison result between the first quality information and second quality information regarding the second beam included in the output result.

8. The base station according to claim 7, wherein said receiving means receives information indicating said comparison result from said communication terminal.

9. A communications terminal comprising: a receiving means for receiving a first RRC message from the base station inquiring about information on at least one candidate learning model to switch to when switching the learning model used to determine the beam to be used for communication with the base station to another learning model; and a transmitting means for transmitting a second RRC message to the base station including information on the at least one candidate learning model.

10. The communication terminal according to claim 9, wherein said receiving means receives information indicating a learning model selected from said at least one candidate learning model from said base station.

11. A communications terminal comprising a transmitting means for transmitting a MAC CE (Media Access Control Control Element) to a base station, the MAC CE including information about at least one candidate learning model to be used as a switching target when switching the learning model used to determine the beam to be used for communications with the base station to another learning model.

12. A control method executed in a base station, which, when switching a learning model used by a communication terminal to determine a beam to be used for communication with a base station to another learning model, sends a first RRC message to the communication terminal inquiring about information on at least one candidate learning model to switch to, and receives a second RRC message from the communication terminal including information on the at least one candidate learning model.

13. A communication method executed in a communication terminal, which, when switching a learning model used to determine a beam to be used for communication with a base station to another learning model, receives a first RRC message from the base station inquiring about information on at least one candidate learning model to switch to, and transmits a second RRC message to the base station including information on the at least one candidate learning model.

14. A communication method executed in a communication terminal, in which, when switching the learning model used to determine the beam to be used for communication with a base station to another learning model, a MAC CE (Media Access Control Control Element) including information about at least one candidate learning model to switch to is transmitted to the base station.

15. A program that causes a computer to execute the following steps: when switching a learning model that a communication terminal uses to determine a beam to be used for communication with a base station to another learning model, the program sends a first RRC message to the communication terminal inquiring about information about at least one candidate learning model to switch to; and receives a second RRC message from the communication terminal that includes information about the at least one candidate learning model.

16. A program that causes a computer to execute the following steps: when switching a learning model used to determine a beam to be used for communication with a base station to another learning model, receive a first RRC message from the base station inquiring about information about at least one candidate learning model to switch to; and send a second RRC message to the base station that includes information about the at least one candidate learning model.

17. A program that causes a computer to transmit a MAC CE (Media Access Control Control Element) to a base station containing information about at least one candidate learning model to be used as a switching target when switching from a learning model used to determine a beam to be used for communication with a base station to another learning model.

Images