Handover optimization method based on artificial intelligence (AI) / ML model

Abstract

A handover optimization method based on an artificial intelligence (AI) / ML model. In the method, a report message based on AI / ML model inference is put forwards. A prediction result comprised in the report message at least comprises at least one of the following: a predicted handover probability, predicted handover time, a predicted handover failure rate, and prediction result acquisition time. In addition, in the method, event triggering for and a reporting cycle of the report message are configured. In addition, in the method, a handover decision based on AI / ML model inference is enabled. In addition, in the method, one AI / ML model / a category of AI / ML models / a group of AI / ML models is / are activated or deactivated by means of a broadcast message. In addition, in the method, a fallback operation on an AI / ML model at a user equipment side is executed, and the fallback operation is reported or fed back.

Description

Switching optimization method based on artificial intelligence AI / ML model Technical Field

[0001] The present application relates to the field of wireless communication technology, and in particular to a method, user equipment, and base station for handover optimization based on artificial intelligence. Background Art

[0002] Currently, the integration of artificial intelligence (AI) and communication systems has become a key research focus for 5.5G (3GPP Release 18) and 6G. The application of AI machine learning (ML) in wireless communications has been widely discussed, and its application scope includes the physical layer, high layer, and system architecture layer. Applications at the physical layer include AI-based channel coding and decoding, channel estimation, interference cancellation, channel state information (CSI) prediction and estimation, positioning, and beam management. Applications at the high layer include AI-based network energy saving, mobility management, load balancing, and radio resource management (RRM) algorithms. Applications at the system architecture layer include NWADF data analysis and application.

[0003] AI / ML can assist L3-based mobility operations, such as optimizing handover decisions through AI / ML-based RRM management and event prediction. Technical issues:

[0004] The introduction of AI / ML-assisted methods requires the improvement of at least the following technical issues:

[0005] 1. For network-triggered L3 handover operations, it is necessary to propose model-related configuration and reporting mechanisms for AI / ML-assisted mobility management.

[0006] 2: More efficient activation / deactivation of the same or similar AI / ML models;

[0007] 3: Mechanism for falling back from AI / ML operations to non-AI / ML operations;

[0008] 4: AI / ML-assisted mobility management requires relevant performance monitoring.

[0009] Therefore, in order to better support AI / ML-assisted use cases, a switching optimization method based on artificial intelligence AI / ML models is needed.

[0010] Summary of the Invention

[0011] One objective of the present disclosure is to propose a method, user equipment, and base station for handover optimization based on an artificial intelligence (AI) / ML model.

[0012] In a first aspect, the present invention provides a handover optimization method based on an artificial intelligence (AI) / ML model, executed in a user equipment, characterized by comprising:

[0013] Report prediction results based on AI / ML model reasoning;

[0014] The prediction result includes at least one of the following: a predicted switching probability, a predicted switching time, a predicted switching failure rate, and a time of the prediction result.

[0015] In a second aspect, the present invention provides a handover optimization method based on an artificial intelligence (AI) / ML model, executed in a base station, and characterized by comprising:

[0016] Receive prediction results based on AI / ML model reasoning;

[0017] The prediction result includes at least one of the following: a predicted switching probability, a predicted switching time, a predicted switching failure rate, and a time of the prediction result.

[0018] In a third aspect, an embodiment of the present invention provides a user equipment, comprising a processor and a memory, wherein the processor is configured to call and execute a computer program stored in the memory so that the device equipped with the processor performs the disclosed method.

[0019] In a fourth aspect, an embodiment of the present invention provides a network node, comprising a processor and a memory, wherein the processor is configured to call and execute a computer program stored in the memory so that a device equipped with the processor performs the disclosed method.

[0020] The disclosed method can be programmed as computer-executable instructions stored in a non-transitory computer-readable medium. The non-transitory computer-readable medium, when loaded into a computer, instructs the processor of the computer to execute the disclosed method.

[0021] The non-transitory computer-readable medium may include at least one of the group consisting of a hard disk, a CD-ROM, an optical storage device, a magnetic storage device, a read-only memory, a programmable read-only memory, an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), and a flash memory.

[0022] The disclosed method can be programmed as a computer program product, which causes a computer to perform the disclosed method.

[0023] The disclosed method may be programmed as a computer program, which causes a computer to perform the disclosed method.

[0024] Technical effects:

[0025] The method proposes a reporting message of model reasoning of an AI / ML model. The prediction result contained in the reporting message includes at least one of the following: predicted switching probability, predicted switching time, predicted switching failure rate, and time of the prediction result. In addition, the method proposes configuration of event triggering and reporting cycle of the reporting message. In addition, the method proposes to enable switching decision-making based on model reasoning of the AI / ML model. In addition, the method proposes to activate or deactivate one / class / group of AI / ML models through a broadcast message. In addition, the method proposes to perform a fallback operation on the AI / ML model on the user equipment side, and report or feedback the fallback operation.

[0026] During the AI-based handover preparation phase, the user equipment (UE) can infer one or more pieces of information using the AI ​​model, such as the predicted candidate / target cell, the predicted target / candidate beam, the predicted reference signal receiving power (RSRP) and reference signal receiving quality (RSRQ) for each cell, the signal-to-interference plus noise ratio (SINR), the predicted handover failure rate, the predicted handover time, and the predicted handover probability. The UE reports this predicted information to the source base station to assist the source base station in making a handover decision.

[0027] Specifically, the UE can select the information to report based on actual conditions. For example, reporting only candidate cells or target cells can reduce signaling overhead. By using artificial intelligence models to predict candidate / target cells for a period of time in the future, the source base station can pre-configure access information for the candidate or target cells, such as resources and cell identifiers, to reduce handover interruption time. Information such as the predicted handover probability and predicted handover failure rate can help the source base station reduce the handover failure rate and improve handover robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] One or more embodiments are exemplarily illustrated by the figures in the corresponding drawings. These exemplifications do not limit the embodiments. Elements with the same reference numerals in the drawings represent similar elements. Unless otherwise specified, the figures in the drawings are not intended to be proportional. The following divisions of the various embodiments are for ease of description and are not intended to limit the specific implementation of the present invention. The various embodiments may be combined and referenced with each other as long as there is no contradiction.

[0029] FIG1 is a schematic diagram showing the intra-AMF / UPF handover operation.

[0030] FIG2 is a schematic diagram showing a communication system.

[0031] FIG3 shows a schematic diagram of the method of the present invention.

[0032] FIG4 is a schematic diagram showing the model prediction function in a UE-side model scenario.

[0033] FIG5 is a schematic diagram showing the model activation / deactivation, fallback, and performance monitoring functions in a UE-side model scenario.

[0034] FIG6 is a schematic diagram showing a method for reporting prediction results triggered by an event.

[0035] FIG7 is a schematic diagram showing a method for reporting periodic prediction results.

[0036] FIG8 is a schematic diagram showing a switching operation for enabling model inference based on the AI / ML model.

[0037] FIG9 is a schematic diagram illustrating a switching operation for enabling model inference based on the AI / ML model according to network configuration.

[0038] FIG10 is a schematic diagram showing a method in which a UE proactively enables an AI / ML model.

[0039] FIG11 is a schematic diagram showing the process of activating a certain / class / group of AI / ML models through broadcast signaling.

[0040] FIG12 is a schematic diagram showing the process of performing a fallback mechanism on the UE side.

[0041] FIG13 is a schematic diagram showing the process of implementing a fallback mechanism on the network side.

[0042] FIG14 is a schematic diagram showing the monitoring process.

[0043] FIG15 is a schematic diagram showing a monitoring process at the UE side.

[0044] FIG16 is a schematic diagram showing the monitoring process at the network side.

[0045] FIG17 is a schematic diagram showing a user equipment according to the present invention.

[0046] FIG18 is a schematic diagram showing a network node according to the present invention.

[0047] FIG19 is a schematic diagram showing a chip of the present invention.

[0048] FIG20 is a schematic diagram showing a chip of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of this application more clear, some embodiments of this application are further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain this application and are not used to limit this application.

[0050] The relevant technical terms in this article are explained as follows:

[0051] Table 1

[0052] The present invention allows wireless networks to use artificial intelligence to optimize the switching process, thereby improving the stability of the switching and reducing the network burden.

[0053] The introduction of AI / ML-assisted methods requires the improvement of at least the following technical issues:

[0054] 1. For network-triggered L3 handover operations, AI / ML-assisted mobility management requires research on the configuration and reporting of model inference during handover preparation, as well as the reporting mechanism for model inference results.

[0055] 2: Activate / deactivate the same or similar AI / ML models by broadcasting (Paging, SIB);

[0056] 3: Mechanism for falling back from AI / ML operations to non-AI / ML operations;

[0057] 4: For network-triggered L3 switching operations, AI / ML-assisted mobility management requires a related performance monitoring process.

[0058] In the RRC_CONNECTED state, the user equipment (UE) measures at least one or more beams of a cell and averages the measurement results (e.g., power values) to assess the cell quality. During this process, the UE is configured to consider only a subset of the detected beams.

[0059] The network has the ability to configure a UE in the RRC_CONNECTED state to perform measurements. The network can configure the UE to report measurement results based on the measurement configuration, or to perform a conditional reconfiguration evaluation under specific conditions. The measurement configuration is provided via dedicated signaling, such as RRCReconfiguration or RRCResume.

[0060] The measurement configuration includes the following parameters: 1. Measurement object; 2. Report configuration.

[0061] 1. Measurement Object (MO): The measurement object can be represented by a list of objects that the UE needs to measure.

[0062] For both intra-frequency and inter-frequency measurements, the measurement object indicates the frequency / time position and subcarrier spacing of the reference signal (RS) to be measured. Associated with this measurement object, the network can configure a cell-specific offset list, a list of 'exclude-listed' cells, and a list of 'allow-listed' cells. Excluded-list cells are not applicable for event evaluation or measurement reporting. Allowed-list cells are cells applicable for event evaluation or measurement reporting.

[0063] The measObjectId of the MO corresponding to each serving cell is indicated by servingCellMO in the serving cell configuration.

[0064] 2. Report configuration: Report configuration can be represented by a report configuration list, where each measurement object can have one or more report configurations. Each measurement report configuration includes the following:

[0065] 2.1. Reporting Condition: The condition that triggers the UE to send a measurement report. This can be described as periodic or a single event.

[0066] 2.2 RS type: RS (SS / PBCH block or CSI-RS) used by the UE for beam and cell measurement results.

[0067] 2.3. Report format: The UE includes in the measurement report the number of each cell and each beam (e.g. RSRP) and other relevant information, such as the maximum number of cells to be reported and the maximum number of beams per cell.

[0068] Detailed measurement configuration information and measurement results are contained in MeasConfig and MeasResult.

[0069] Control Surface (C-Plane) Processing:

[0070] Network-controlled mobility for user equipment (UE) in the RRC_CONNECTED state is divided into two types of mobility: cell-level mobility and beam-level mobility. Beam-level mobility includes intra-cell beam-level mobility and inter-cell beam-level mobility.

[0071] Intra-NR RAN handovers perform both the preparation and execution phases of the handover process without involving the 5GC. Handover preparation messages are exchanged directly between gNBs. During the handover completion phase, the target gNB triggers the source gNB to release resources. Figure 1 illustrates a basic handover scenario, where neither the access and mobility management function (AMF) nor the user plane function (UPF) remain unchanged:

[0072] Step 0. The UE context in the source gNB contains information about roaming and access restrictions, which was provided during connection establishment or during the last tracking area (TA) update.

[0073] Step 1. The source gNB configures a measurement procedure for the UE. The UE performs measurements and submits measurement reports according to the configuration of the measurement procedure.

[0074] Step 2. The source gNB decides to hand over the UE based on the measurement report (e.g., MeasurementReport) and radio resource management (RRM) information.

[0075] Step 3. The source gNB sends a Handover Request message to the target gNB, delivering a transparent Radio Resource Control (RRC) container containing the necessary information required by the target gNB to prepare for the handover operation. This information includes at least the target cell ID, KgNB, the UE's cell radio network temporary identifier (C-RNTI) in the source gNB, RRM configuration (including UE inactivity time), basic access stratum (AS) configuration (including antenna information and downlink carrier frequency), the mapping rule applied to the UE's current Quality of Service flow (QoS flow) to Data Radio Bearers (DRBs), SIB1 information of the source gNB, UE capabilities under different Radio Access Technologies (RATs), Protocol Data Unit (PDU) session related information, and measurement information (including beam-related information) that may be reported by the UE. PDU session related information includes slice information and QoS profile at the QoS flow level. The source gNB may also request a Dual Active Protocol Stack (DAPS) switch for one or more DRBs.

[0076] It should be understood that after issuing a handover request, the source gNB should not reconfigure the UE, including performing reflective QoS flow to DRB mapping.

[0077] Step 4. The target gNB may perform admission control. If slice information is sent to the target gNB, slice-aware admission control shall be performed. If a PDU session is associated with an unsupported slice, the target gNB shall reject such PDU session.

[0078] Step 5. The target gNB prepares for handover at the L1 / L2 layer and sends a Handover Request Acknowledgement (Handover Request Acknowledgement) to the source gNB. This HACKER message contains a transparent container and is sent to the UE as an RRC message to perform the handover. The target gNB also indicates whether it accepts DAPS handover.

[0079] It should be understood that data forwarding may be initiated once the source gNB receives the Handover Request Acknowledge or upon transmission of a downlink Handover Initiate Command.

[0080] It is important to understand that for DRBs configured with DAPS, downlink Packet Data Convergence Protocol (PDCP) Service Data Units (SDUs) will be forwarded using the sequence numbers (SNs) assigned by the source gNB until the SN assignment is switched to the target gNB in ​​step 8b, after which normal data forwarding as defined in the relevant communication standards will be followed.

[0081] Step 6. The source gNB triggers Uu handover by sending an RRC Reconfiguration message to the UE. This message contains the information required to access the target cell: at least the target cell ID, the new C-RNTI, and the target gNB security algorithm identifier for the selected security algorithm. It may also include a set of dedicated random access channel (RACH) resources, the association between RACH resources and synchronization signal blocks (SSBs), the association between RACH resources and UE-specific CSI-RS configurations, common RACH resources, and system information for the target cell.

[0082] It should be understood that for DRBs configured with DAPS, the source gNB will not stop sending downlink data packets until it receives a handover success message (Handover Success) from the target gNB.

[0083] In traditional handover (HO) methods, network-controlled mobility is primarily applied in the RRC_CONNECTED state, covering both cell-level and beam-level mobility. Taking intra-network RAN ​​handover as an example, typically, the user equipment (UE) uses the measurement results (such as RSRP, RSRQ, and SINR) reported in the measurement configuration. The serving gNB (S-gNB) then makes a handover decision based on the received measurement reports and radio resource management (RRM) information. It then initiates the handover and issues a handover request (HANDOVER REQUEST) over the Xn interface.

[0084] However, unlike the traditional method of obtaining cell quality by measuring reference signals during handover preparation, in AI / ML-based solutions, historical data can be used as input to the AI / ML model. Through the AI / ML model, one or more pieces of information can be inferred to assist in the execution of handover decisions. This information includes predicted alternative / target cells, predicted target / alternative beams, predicted RSRP, RSRQ, SINR (per cell), predicted handover failure rate, and predicted handover ping-pong, etc. Because its existing configuration does not define configurations related to the AI / ML model, especially for scenarios with enhanced handover decisions. The AI / ML model cannot directly reuse the traditional measurement reporting mechanism.

[0085] According to the UE-side model discussion in 3GPP Release 18 (R18), the UE-side data input to the UE-side model prediction function (except for assistant info, such as information elements related to data quality) can be obtained through traditional methods. However, as a new handover use case introduced in Release 19 (R19), it is necessary to study whether the configuration and reporting of the model inference function (model inference function) requires additional information elements or mechanisms.

[0086] Furthermore, performance monitoring, fallback, and model activation / deactivation are all important components of AI / ML-based lifecycle management. In AI / ML-assisted network-triggered L3 handovers, the processes and / or signaling also require corresponding enhancements.

[0087] Referring to Figure 2 , one or more core network devices (e.g., core network device 30) that execute core network network functions are connected to multiple base stations, including base stations 20a, 20b, ..., 20m. Base stations 20a and 20b may be referred to as the first base station and the second base station. In some examples, base station 20a may serve as a source base station, and base station 20b may serve as a target base station. The multiple base stations are connected to multiple user equipment (UEs) 10a, 10b, ..., 10n via wireless channels. m and n may be positive integers.

[0088] The network entity device 30 may be a node in a CN. The CN may include an LTE CN or a 5G core (5GC), which includes a user plane function (UPF), a session management function (SMF), an access and mobility management function (AMF), a unified data management (UDM), a policy control function (PCF), a control plane (CP) / user plane (UP) separation (CUPS), an authentication server (AUSF), a network slice selection function (NSSF), and a network exposure function (NEF).

[0089] Examples of UEs described herein may include one of the UEs 10a, UE 10b, ..., or UE 10n. Examples of base stations described herein may include the base stations 20a or 20b. A network may be a network node of the base station or a core network. Examples of base stations may be gNBs, eNBs, or other base stations. Uplink (UL) transmission of control signals or data may be a transmission operation from a UE to a base station. Downlink (DL) transmission of control signals or data may be a transmission operation from a base station to a UE. DL control signals may include medium access control (MAC) control elements (CE), downlink control information (DCI), or radio resource control (RRC) signals from a base station to a UE.

[0090] Model inference is performed on the UE side (e.g., at least one of UEs 10a-10n). The UE generates a prediction based on the model (including candidate / target cell reports) and sends it to the source base station (e.g., base station 20a). The source base station then makes a handover decision based on the reported prediction results. The configuration process for the model inference function on the UE side is shown in Figure 4.

[0091] Referring to FIG3 , a method for handover optimization based on an artificial intelligence (AI) / ML model is described below. User equipment 10a reports a prediction result 111 based on AI / ML model reasoning to base station 20a ( S10 ). Base station 20a receives prediction result 111 ( S11 ). The prediction result includes at least one of the following: a predicted handover probability, a predicted handover time, a predicted handover failure rate, and a predicted result time.

[0092] In some embodiments of the present invention, the base station 20a sends a configuration message about the AI / ML model, and the user equipment 10a receives the configuration message about the AI / ML model.

[0093] In some embodiments of the present invention, the reporting method of the prediction result includes at least one of the following:

[0094] Based on event triggering, reporting the prediction results;

[0095] Based on the period, the prediction result is reported.

[0096] Accordingly, the base station 20a receives the report of the prediction result based on at least one of the following methods:

[0097] Based on event triggering, receiving a report of the prediction result;

[0098] Based on a period, a report of the prediction result is received.

[0099] In some embodiments of the present invention, the event trigger is based on configuration or the event trigger is predefined. The period is based on configuration or the period is predefined.

[0100] The event triggering includes at least one of the following:

[0101] When the predicted handover probability is greater than a first threshold after being adjusted by the first offset value, the event triggering condition is met;

[0102] When the predicted handover failure rate is less than the second threshold after being adjusted by the second offset value, the event triggering condition is met; or

[0103] The time reaches the time position or time range indicated by the time indication information of the predicted switching time.

[0104] In some embodiments of the present invention, the method further includes: the user equipment 10a sending a reporting message according to a reporting period to report the prediction result of the AI / ML model reasoning, and the base station 20a receiving the reporting message according to the reporting period.

[0105] The configuration of the reporting period includes at least one of the following:

[0106] Reporting interval, reporting quantity, maximum number of reporting cells, inference amount of reporting cells, index of reported reference signals, maximum number of reference signals, and allowed cell list.

[0107] In some embodiments of the present invention, the conditions for enabling the model inference function of the AI / ML model include at least one of the following:

[0108] receiving configuration information of a base station;

[0109] Satisfy predefined or configured conditions; or

[0110] The performance monitoring results of the AI / ML model meet predefined or configured thresholds.

[0111] In some embodiments of the present invention, the activation method of the AI / ML model includes: activating the AI / ML model through a broadcast message, wherein the AI / ML model is a single / class / group AI / ML model.

[0112] In some embodiments of the present invention, the deactivation method of the AI / ML model includes: deactivating the AI / ML model through a broadcast message, wherein the AI / ML model is a single / class / group of AI / ML models.

[0113] In some embodiments of the present invention, the broadcast message includes at least one of the following: a paging message, a short paging message, and a system information block (SIB) message. The paging message, short paging message, or system information block (SIB) message includes at least one of the following information about the one / class / group of AI / ML models:

[0114] Identification information;

[0115] AI / ML features; and

[0116] AI / ML status.

[0117] In some embodiments of the present invention, the AI / ML model fallback mechanism includes: the user equipment 10a reporting indication information about the fallback operation to instruct the system to fall back from the operation based on the AI / ML model to the operation based on the non-AI / ML model. The base station 20a receives the indication information about the fallback operation.

[0118] In some embodiments of the present invention, the indication information of the fallback operation is fallback indication information, wherein a first value of the fallback indication information indicates that fallback is enabled, and a second value of the fallback indication information indicates that fallback is not enabled.

[0119] In some embodiments of the present invention, the indication information of the fallback operation is an AI / ML model reporting message on the user device side, wherein the prediction result in the reporting message is empty, indicating that the fallback is enabled, and the prediction result in the reporting message is non-empty, indicating that the fallback is not enabled.

[0120] In some embodiments of the present invention, the user equipment makes a fallback decision based on monitoring results of performance monitoring or based on calculated statistics of measurements.

[0121] In some embodiments of the present invention, the monitoring mechanism of the AI / ML model includes: the user equipment 10a receiving a configuration for performance monitoring of the AI / ML model, generating a monitoring result of the performance monitoring, and reporting the monitoring result of the performance monitoring related to the AI / ML model. The base station 20a sends the configuration for performance monitoring of the AI / ML model to the user equipment and receives the monitoring result of the performance monitoring related to the AI / ML model.

[0122] In some embodiments of the present invention, the configuration of the performance monitoring includes at least one of the following information:

[0123] Performance indicators;

[0124] Performance comparison benchmark / reference;

[0125] performance parameter thresholds;

[0126] Monitoring start time;

[0127] Monitor execution time;

[0128] Monitoring reporting time;

[0129] Monitor object.

[0130] In some embodiments of the present invention, the performance indicator includes at least one of the following:

[0131] Key performance indicators (KPIs) related to candidate / target cell prediction accuracy;

[0132] Key performance indicators (KPIs) related to link quality;

[0133] Performance metrics based on AI / ML input / output data distribution; and

[0134] The difference between the candidate / target cells indicates the difference between the candidate / target cells evaluated by comparing the benchmark candidate / target cells for performance comparison with the predicted candidate / target cells.

[0135] In some embodiments of the present invention, the performance parameter threshold includes at least one of the following:

[0136] Thresholds for key performance indicators (KPIs) related to candidate / target cell prediction accuracy;

[0137] Thresholds for key performance indicators (KPIs) related to link quality; and

[0138] The threshold for the difference between candidate and target cells.

[0139] In some embodiments of the present invention, the performance monitoring is performed in a user equipment, and reporting the monitoring results of the performance monitoring of the AI / ML model includes: the user equipment 10a transmitting a performance monitoring reporting message. The base station receives the performance monitoring reporting message. The performance monitoring reporting message includes at least one of the following information:

[0140] performance indicators / methods;

[0141] The monitoring results; and

[0142] Lifecycle management operations.

[0143] In some embodiments of the present invention, the performance monitoring is performed in a base station, and the monitoring mechanism of the AI / ML model includes: receiving a monitoring result of the performance monitoring related to the AI / ML model.

[0144] The specific steps in the process of Figure 4 are described as follows:

[0145] Step A1: The UE receives a configuration message regarding model inference sent from the source base station (e.g., base station 20a). The configuration message includes a configuration for model inference and a configuration for reporting a prediction result of the model inference. In this document, the prediction result of the model inference may be referred to as a prediction result or a model inference result. The configuration message includes at least one of the following:

[0146] 1. Predicted candidate / target cells,

[0147] 2. Predicted reference signal received power RSRP,

[0148] 3. Predicted (optimal) beam,

[0149] 4. Predicted reference signal received power RSRQ,

[0150] 5. Predicted signal-to-interference-and-noise ratio (SINR),

[0151] 6. Predicted handover failure rate,

[0152] 7. Predicted Ping Pong,

[0153] 8. Predicted measurement events,

[0154] 9. Predicted switching probability,

[0155] 10. Predicted switching time,

[0156] 11. The time of the predicted results, and

[0157] 12. Methods for reporting prediction results, etc.

[0158] The time of the prediction result can be used to indicate the time during which the prediction result is valid. The configuration message can include a reporting configuration for the prediction result of the model inference. The reporting configuration for the prediction result includes at least one of the following:

[0159] 1. Predicted candidate / target cells,

[0160] 2. Predicted (optimal) beam,

[0161] 3. Predicted RSRP,

[0162] 4. Predicted RSRQ,

[0163] 5. Predicted SINR,

[0164] 6. Predicted handover failure rate,

[0165] 7. Predicted Ping Pong,

[0166] 8. Predicted measurement events,

[0167] 9. Predicted switching probability,

[0168] 10. Predicted switching time,

[0169] 11. Time to predict the results,

[0170] 12. How to report the prediction results.

[0171] 13. Rollback instructions, and

[0172] 14. Report incidents, etc.

[0173] Step A2: The UE performs a model inference function. Specifically, the UE runs one or more appropriate models based on the model input data to perform model inference and generate prediction results. The model inference results / prediction results include at least one or any combination of the following:

[0174] 1. Predicted candidate / target cells,

[0175] 2. Predicted (optimal) beam,

[0176] 3. Predicted RSRP,

[0177] 4. Predicted RSRQ,

[0178] 5. Predicted SINR,

[0179] 6. Predicted handover failure rate,

[0180] 7. Predicted Ping Pong,

[0181] 8. Predicted measurement events,

[0182] 9. Predicted switching probability,

[0183] 10. Predicted switching time, and

[0184] 11. Time to predict results.

[0185] The prediction results are detailed in Example 3.

[0186] Step A3: The UE reports the prediction result of the model reasoning. Based on the configuration of the base station, a reporting message can be sent through an RRC message to carry the prediction result. Optionally, if the reporting content is small, such as only reporting the target cell, it can also be reported through a Medium Access Control (MAC) control element (CE) or uplink control information (UCI), see Example 1 for details. In addition, the reporting method can be periodic reporting or event-triggered reporting, see Example 2 for details.

[0187] Step A4: The source base station (eg, base station 20a) receives the prediction results of the model inference sent by the UE, and / or the measurement reporting amount in the traditional manner, and evaluates and makes a switching decision based on the above prediction results and / or the measurement reporting amount in the traditional manner.

[0188] Step 5: The source base station (e.g., base station 20a) sends a handover request message to the target base station (e.g., base station 20b). The handover request message may include a prediction result. The prediction result includes at least one or any combination of the following:

[0189] 1. Predicted candidate / target cells,

[0190] 2. Predicted (optimal) beam,

[0191] 3. Predicted RSRP,

[0192] 4. Predicted RSRQ,

[0193] 5. Predicted SINR,

[0194] 6. Predicted handover failure rate,

[0195] 7. Predicted Ping Pong,

[0196] 8. Predicted measurement events,

[0197] 9. Predicted switching probability,

[0198] 10. Predicted switching time, and

[0199] 11. Time to predict results.

[0200] The predicted switching time is a predicted switching execution time, and the predicted switching time includes at least one of the following:

[0201] A time offset T of the switching indication, used to indicate that the switching is performed after the time offset T; and

[0202] The time window W of the switching indication is used to indicate that the switching is performed within the time window W.

[0203] 5 , in other embodiments, the above process may further include step 6. Steps 6-7 are optional.

[0204] Step 6: The base station executes model activation / deactivation via broadcast message, see Example 4 for details.

[0205] In other embodiments, step 1, and / or step 3, and / or other step 7 of the above process may also include fallback indication information: the fallback indication information is optional.

[0206] Step 7: The UE performs a fallback operation (fallback), see Example 5 for details.

[0207] In other embodiments, the above process may further include step 8: step 8 is optional.

[0208] Step 8: The UE or base station performs performance monitoring to determine the adaptability of the model through model / performance monitoring and comparison. Specifically, model / performance monitoring involves performance monitoring parameters, monitoring comparison benchmarks, and monitoring procedures, as detailed in Example 6.

[0209] It is important to understand that there is no logical order for the above steps. All of the above steps can be performed independently.

[0210] AI / ML aided mobility for network triggered L3-based handover involves model reasoning based on past historical measurement results, or measurement results, or statistical calculation results. The historical measurement results, or measurement results, or statistical calculations used as inputs to the model reasoning can be the measured RSRP, RSRQ, SINR, the related cells of the measured RSRP, RSRQ, SINR, beam information, historical handover failure rate (HO failure rate), handover ping-pong, UE behavior trajectory, UE location, etc.

[0211] The definitions of terms used in the description of the embodiments of the present invention are as follows:

[0212] (A) Model ID: The Model ID is used to identify the model. There are two types:

[0213] 1) Global AI / ML model ID: used to identify the AI / ML model. It can be a globally unique ID, a Public Land Mobile Network (PLMN)-specific ID, an operator-specific ID, or a unique ID specific to the AI / ML management platform.

[0214] 2) Logical AI / ML model ID: It is used to identify the AI / ML model used in the network and has a certain mapping relationship with the global ID. The logical ID can be a globally unique ID, a PLMN-specific unique ID, an operator-specific unique ID, a cell-specific ID, a link-specific ID, a TA-specific ID, a CU-specific ID, a DU-specific ID, a UPF-specific ID, an AMF-specific ID, an RRC-specific ID, or a network slice-specific ID. The term "cell-specific" means that each AI / ML model has a unique ID in a specific cell. Similarly, "link-specific", "TA-specific", "CU-specific", "DU-specific", "UPF-specific", "AMF-specific", and "network slice-specific" refer to the unique IDs of AI / ML models in a specific environment within the network.

[0215] (B) Functionality: The functionality indicates the functions or features supported by the AI / ML model, such as AI / ML-based radio resource management (RRM) prediction, AI / ML-based event prediction (including measurement event prediction, handover failure rate prediction, link failure prediction), AI / ML-based candidate / target cell prediction, or AI / ML-based candidate or target beam prediction.

[0216] (C) Metadata information: The metadata information is used to describe / identify the model and includes at least one of the following: model input information; model output information; model version information; model format information; required artificial intelligence capabilities; supplier information; applicable scenarios, configuration, and location information; computational complexity; model complexity; model size; model performance; and model functionality.

[0217] The computational complexity may include floating-point operations per second (FLOPS), and the level of pre- / post-processing.

[0218] The model complexity may include: the number of real-valued model parameters, the number of real-valued operations, and the model size.

[0219] The model performance may include: model accuracy, model bias, and model variance.

[0220] Example 1: Model inference output reported by UE

[0221] Background: The reported inference results can include predicted candidate / target cells / beams, time information, model accuracy, etc.

[0222] For the AI / ML aided mobility for network triggered L3-based handover use case, the reported inference results can also include: predicted handover probability, predicted handover time, predicted handover failure rate, and the time of the prediction result.

[0223] The prediction result is based on the switching probability of the cell, or based on the switching probability of the beam, or based on the switching probability of the entire switching event. For example, the predicted switching probability (predicted HO probability): one of the prediction results inferred by the model is the predicted switching probability. The predicted switching probability can be based on the switching probability of the cell, or based on the switching probability of the beam, or based on the switching probability of the entire switching event. Similarly, the predicted switching time and / or the predicted switching failure rate can be based on the switching probability of the cell, or based on the switching probability of the beam, or based on the switching probability of the entire switching event.

[0224] If the predicted handover probability is based on a handover event, based on the predicted handover probability, the result of the handover probability (predicted HO probability) can be used to suggest to the source base station (e.g., base station 20a) whether handover is required. If the UE moves to the cell edge, the predicted handover probability can be used to suggest to the source base station (e.g., base station 20a) whether handover is required.

[0225] If the predicted switching probability is based on the cell, then the predicted switching probability serves as auxiliary information for the source base station (e.g., base station 20a) to evaluate switching to a target base station (e.g., base station 20b), for example, assisting the source base station (e.g., base station 20a) in determining a suitable target base station (e.g., base station 20b) for the switching operation.

[0226] In some embodiments, if the predicted handover probability is based on a handover event, the handover event probability may correspond to the handover probability of one or more cells. For example, if the predicted handover probability is 80%, the corresponding handover probabilities of three cells are 90%, 80%, and 70%. The UE may then report the predicted handover probability of the handover event and the predicted handover probability of the handover cell to the source base station (e.g., base station 20a). The source base station (e.g., base station 20a) may make a judgment based on the predicted handover probability, such as selecting the cell with the highest predicted handover probability as the target base station (e.g., base station 20b), and thus initiate a handover request.

[0227] The predicted handover probability may be expressed in at least one of the following or a combination thereof:

[0228] Method 1: Use 1 bit to indicate the predicted handover probability. For example, if the bit is set to 1, it indicates that the source base station (e.g., base station 20a) is recommended to perform handover; if the bit is set to 0, it indicates that the source base station (e.g., base station 20a) is not recommended to perform handover.

[0229] Method 2: Rough Representation: Define the predicted handover probability as high, medium, or low. Each level has a range, roughly representing the predicted handover probability. For example, "high" is defined as a handover probability of 70%-100%, medium is defined as a handover probability of 40%-70%, and low is defined as a handover probability of 0-40%.

[0230] The high, medium and low can be indicated by 2 bits. For example, 00 represents that the predicted switching probability is high, 01 represents that the predicted switching probability is medium, and 10 represents that the predicted switching probability is low. An example is: when the predicted switching probability is high, the UE reports 00. The network (NW) side receives an indication that the predicted switching probability is high, indicating that it is recommended that the source base station (such as base station 20a) perform switching, or switch to which target base station (such as base station 20b). According to the indication, the source base station (such as base station 20a) may perform switching. When the predicted switching probability is low, the UE reports 10. The network (such as base station 20a or 20b) side receives an indication that the predicted switching probability is low, indicating that it is recommended that the source base station (such as base station 20a) does not need to perform switching. According to the indication, the source base station (such as base station 20a) may not perform switching.

[0231] Method 3: Precise Representation: The specific value of the predicted handover probability inferred by the model is directly expressed as a percentage. For example, the model infers a predicted handover probability of 63%. The UE reports 63%. The network (NW) receives an indication of the predicted handover probability of 63%. Based on this information, the network can at least decide whether to perform a handover or which target base station (e.g., base station 20b) to handover to.

[0232] The above-mentioned method for expressing the predicted handover probability may be configured by the network side (eg, base station 20a or 20b), or predefined. The signaling may be RRC, MAC CE, or DCI signaling.

[0233] The prediction result is a prediction result representing one candidate / target cell or multiple candidate / target cells. For example, the predicted HO probability and the candidate / target cell relationship can be at least one of the following:

[0234] 1) One-to-one: one predicted handover probability corresponds to one candidate / target cell, or one handover event;

[0235] 2) One-to-many: one predicted handover probability corresponds to multiple candidate / target cells;

[0236] It should be understood that the terms “predicted HO probability”, “predicted HO accuracy”, “predicted HO success probability”, or “predicted HO failure rate” in the present invention may be used interchangeably in the embodiments.

[0237] Predicted handover failure rate: One of the prediction results derived from the model is the predicted handover failure rate. The predicted handover failure rate can be based on a cell or on the entire handover event.

[0238] In some examples, such as when the predicted handover failure rate is indicated by 1 bit, the UE reports the predicted handover probability as "handover recommended". The reported information may also include predicted alternative / target cell information, or predicted RSRP, or predicted RSRQ, or predicted SINR, or alternative cell information. The base station (e.g., base station 20a or 20b) may make a handover decision (HO decision) based on the reported prediction results, such as selecting one or more of the predicted alternative / target cells in the reported information to send a handover request signaling. The UE reports the predicted handover probability as "handover not recommended", and the reported information may or may not include the predicted alternative / target cell information. The base station (e.g., base station 20a or 20b) decides whether handover is required based on the reported prediction results. This does not exclude the base station (e.g., base station 20a or 20b) from using the results of a traditional measurement solution to perform handover.

[0239] In some examples, if each candidate cell is associated with its predicted handover probability, the UE reports the candidate cells and their predicted probabilities. The base station (e.g., base station 20a or 20b) makes a handover decision (HO decision) based on the reported prediction results, such as selecting one or more of the predicted candidate / target cells in the reported information to send handover request signaling.

[0240] This embodiment can bring the following technical benefits. Since the channel conditions change very quickly and the prediction is unstable, an additional dimension of prediction results is added to assist the source base station (e.g., base station 20a), such as suggesting that the source base station (e.g., base station 20a) switch or not switch. Alternatively, the source base station (e.g., base station 20a) can use the reported predicted alternative / target cell and its related predicted switching probability to make an evaluation and decide whether to switch or select a more suitable target base station (e.g., base station 20b) to optimize the switching decision (HO decision). In addition, if the result inferred by the model indicates that switching will not occur, then only an indication that switching will not occur is reported, which can reduce signaling overhead.

[0241] Predicted HO time: This is used to represent one of the prediction results derived from the model, which is the predicted execution time of the handover. There are two possible implementations of the predicted handover time:

[0242] Mode 1: Introducing a time switching indication time offset T to indicate that switching is performed after the time offset T. For example, if the time switching indication is 1 minute, it means that the network (e.g., base station 20a or 20b) performs switching one minute after receiving the information reported by the UE.

[0243] Method 2: Introducing a handover indication time window W. If the handover indication time window is 1 minute, it means that the network (eg, base station 20a or 20b) performs handover within one minute after receiving the information reported by the UE.

[0244] Mode 3: A handover indication T and a handover indication time window W are introduced to indicate that handover is performed within the time window W. For example, if the handover indication T is 1 minute and the handover indication time window W is 1 minute, it means that the network (e.g., base station 20a or 20b) performs handover one minute after receiving the information reported by the UE, and the handover is performed within 1 minute.

[0245] The technical effects are as follows: by predicting the switching time, the network side can be assisted to perform the switching at the appropriate time, reducing the switching failure rate and the ping-pong effect.

[0246] Prediction result time: This time information is used to indicate that the prediction result is valid within a certain time period / multiple time periods in the future. The time information can be configured by the network or determined by the model's capabilities. The relationship between the prediction result (predicted candidate / target cell) and the prediction result time can be one-to-one, one-to-many, or many-to-one. For example, "candidate cell#3at T0, candidate cell#3, cell#4, cell#5at T1" means that candidate cell#3 is valid within T0 and T1, and candidate cells#4 and #5 are valid within T1.

[0247] The time information of the prediction result can be any combination of the following: start time, duration, end time, offset time;

[0248] 1. Start time: indicates the start time when the prediction result is valid.

[0249] 2. Duration: indicates the duration during which the prediction result is valid.

[0250] 3. End time: indicates the end time when the prediction result is valid.

[0251] 4. Bias time: represents the bias time during which the prediction result is valid.

[0252] The technical effects are as follows:

[0253] 1) Optimize handover decision-making, allowing the source base station (e.g., base station 20a) to request resources and access information from the target base station (e.g., base station 20b) in advance, thereby reducing handover preparation delay.

[0254] 2) Assisting handover decision-making: If the AI / ML model can predict a handover cell in the future, the cell can be recommended to the source base station (e.g., base station 20a) as a candidate / target cell;

[0255] 3) Auxiliary model / performance monitoring, by configuring the same or similar time, the monitoring accuracy can be improved.

[0256] Example 2: Reporting mechanism of model inference results (prediction results) reported by UE:

[0257] The reporting mechanism of the model inference result can be based on a period or based on event triggering. The reporting mechanism can be configured by model configuration signaling. The model configuration signaling can be an RRC message.

[0258] (1) Event trigger reporting method:

[0259] 6 , the following describes a method for reporting prediction results triggered by an explanation element.

[0260] Step B001: The network sends a configuration message, wherein the configuration message includes an event-triggered configuration regarding a reporting message of the AI / ML model, and the reporting message includes a prediction result generated by model reasoning of the AI / ML model, which is used to assist in switching decisions.

[0261] Step B002: The UE executes model reasoning of the AI / ML model to generate prediction results to assist in handover decision-making.

[0262] Step B003: When an event triggering a reporting message occurs, the UE sends the reporting message to report the prediction result of the model inference. When an event triggering a reporting message occurs, the network receives the reporting message to obtain the prediction result of the model inference.

[0263] In the AI / ML aided mobility for network triggered L3-based handover use case, the threshold for triggering reporting can be different from the traditional reporting threshold (for example, reporting thresholds based on RSRP, RSRQ, and SINR). Since the prediction results can directly include at least one of the following: predicted alternative / target cell, predicted RSRP, predicted RSRP, predicted SINR, alternative cell, predicted switching probability, predicted switching failure rate, or predicted switching time. The alternative cell can be an alternative cell based on predicted RSRP / or predicted RSRP / or predicted SINR. The event that triggers reporting can be based on the value of any one of the various prediction results or a combination thereof. The following examples are provided.

[0264] Event 1: The predicted switching probability is greater than the threshold. Alternatively, the predicted switching probability meets the condition for triggering the event when it is greater than the first threshold after being adjusted by the first bias value. Based on the predicted switching probability (predicted HO probability), a threshold for the predicted switching probability is defined. If the predicted switching probability generated by the AI / ML model exceeds the threshold, the UE is triggered to send a reporting message to report the model reasoning result (i.e., the prediction result). The reporting message carries the prediction result, such as the predicted alternative / target cell, beam, etc. The threshold can be based on the cell, based on the switching event, or based on both.

[0265] In some examples, if the predicted handover probability of all cells or most cells is greater than the predicted handover probability threshold, the UE is triggered to send a reporting message to report the prediction result. The reporting message carries the prediction result, such as the predicted cell, or candidate cells related to the prediction result, or predicted RSRP / RSRQ / SINR, or predicted handover probability, or predicted handover time, etc.

[0266] In some examples, if the predicted handover probability of the handover event is greater than the predicted handover probability threshold, the UE is triggered to send a reporting message to report the prediction result. The reporting message carries the prediction result, such as the predicted cell, or candidate cells related to the prediction result, or predicted RSRP / RSRQ / SINR, and / or predicted handover probability, or predicted handover time, etc.

[0267] For example, the specific representation of event 1 can be as follows:

[0268] Predicted switching probability - bias 1 > threshold 1 (1)

[0269] The formula (1) is explained as follows:

[0270] Predicted handover probability: indicates the result of model inference, which can be based on handover events or cells. The predicted handover probability can be uncompensated.

[0271] Threshold 1: Indicates the threshold parameter for event 1.

[0272] Offset 1: indicates the offset parameter of event 1. Offset 1 is optional.

[0273] The predicted handover probability, threshold 1, and offset 1 may be expressed as a percentage. The above parameters may be configured, preconfigured, or predefined by the network side (eg, base station 20a or 20b).

[0274] Event 2: The predicted handover failure rate is less than the threshold. Alternatively, the predicted handover failure rate meets the condition for triggering the event when it is less than the second threshold after being adjusted by the second bias value. Based on the predicted handover failure rate (predicted HO probability), a threshold for the predicted handover failure rate is defined. If the predicted handover failure rate generated by the AI / ML model is lower than the threshold, the UE is triggered to send a reporting message to report the model reasoning result (i.e., the prediction result). The reporting message carries the prediction result. The threshold can be based on the cell or based on the handover event.

[0275] In some examples, if the predicted switching failure rate of all cells or most cells is less than the predicted switching failure rate threshold, the UE is triggered to send a reporting message to report the prediction result to the network side (e.g., base station 20a or 20b).

[0276] In some examples, if the predicted handover failure rate of the handover event is less than the predicted handover failure rate threshold, the UE is triggered to report the prediction result to the network side.

[0277] For example, the specific representation of the relevant event 2 can be as follows:

[0278] Predicted handover failure rate + offset 2 < threshold 2 (2)

[0279] The formula (2) is explained as follows:

[0280] Predicted handover failure rate: refers to the handover failure rate predicted by AI / ML models or other means. The predicted handover failure rate can be time-based, number-based, or a combination thereof. For example, the predicted number or proportion of handover failures within a period of time (such as 1 hour); or the cumulative predicted number or proportion of handover failures within the past n times (n is an integer greater than 1).

[0281] Threshold 2: Indicates the threshold parameter for event 2.

[0282] Offset 2: indicates the offset parameter of event 2. Offset 2 is optional.

[0283] The predicted handover failure rate, threshold 1, and offset 1 may be expressed as a percentage. The above parameters may be configured, preconfigured, or predefined by the network side (eg, base station 20a or 20b).

[0284] Event 3: Define time indication information based on the predicted switching time. The time indication information is used to adjust the reporting time of the prediction result. The time indication information can indicate a time offset based on the predicted switching time, or a time window based on the predicted switching time.

[0285] For example, if the AI / ML model predicts that a handover will occur at time T1, the UE can be triggered to send a reporting message t times before T1 to report the model inference result (i.e., the prediction result). t can be a time offset, a time window, or an indicator to adjust the reporting time. The absolute value of the time difference T1-t is greater than the transmission time of the reporting message, such as 5 milliseconds (ms).

[0286] The time indication information may be expressed in time units such as time slots, sub-time slots, microseconds, milliseconds, seconds, or minutes. Parameters related to the time indication information may be configured, preconfigured, or predefined by the network side (eg, base station 20a or 20b).

[0287] The technical effects of this embodiment are described as follows: Only when certain conditions are met can reporting information be sent to report the model inference result (ie, prediction result), thereby saving resources.

[0288] (2) Periodic reporting method:

[0289] 7 , the following describes a method for reporting periodic prediction results.

[0290] Step B011: The network sends a configuration message, wherein the configuration message includes a configuration of a reporting period of the AI / ML model, and the reporting message includes a prediction result generated by model reasoning of the AI / ML model, which is used to assist in switching decisions.

[0291] Step B012: The UE performs model inference of the AI / ML model to generate a prediction result to assist in handover decision-making; and

[0292] Step B012: The UE sends the reporting message according to a reporting period to report the prediction result of the model reasoning. The network receives the reporting message according to a reporting period, and the reporting message includes the prediction result of the model reasoning of the AI / ML model.

[0293] The configuration of the reporting period includes at least one of the following:

[0294] Reporting interval, reporting quantity, maximum number of reporting cells, inference amount of reporting cells, index of reported reference signals, maximum number of reference signals, and allowed cell list.

[0295] The periodic reporting of model inference results (i.e., prediction results) can be based on configuration or predefined. The configuration of periodic reporting includes at least one of the following information: reporting interval, reporting number, maximum number of reporting cells, inference amount of reporting cells, index of reported reference signals, maximum number of reference signals, and allowed cell list. The configuration information is described as follows:

[0296] Reporting interval: used to indicate the interval between two periodic reports.

[0297] Report quantity: used to indicate the number of reports for periodic reporting. Similarly, the report quantity can also be configured for event triggering.

[0298] Maximum number of reported cells: used to indicate the maximum number of non-serving cells included in the reporting message.

[0299] Reporting cell inference amount: used to indicate the inference amount to be included in the reporting message.

[0300] Reported reference signal index (index): used to indicate the index of the reference signal to be included in the reporting message.

[0301] Maximum number of reference signals: used to indicate the maximum number of reference signals included in the reporting message.

[0302] Allowed cell list: The number of cells allowed for measurement or AI / ML inference.

[0303] The technical effects of this embodiment are described as follows: The configuration is simple, but periodic measurement and reporting are required after configuration.

[0304] Example 3: Conditions for enabling AI / ML model methods.

[0305] There are two methods for handover in L3-based mobility management: one is based on the results of traditional measurement reports, and the other is based on the reasoning / results of AI / ML model methods. In the current protocol, the traditional measurement method is enabled by default.

[0306] 8 , an example of a switching operation for enabling model reasoning based on the AI / ML model is described as follows.

[0307] Step B2: The UE performs model inference of the AI / ML model to generate prediction results to assist in handover decision-making.

[0308] Step B3: The UE starts a switching operation of model reasoning based on the AI / ML model according to the configuration of the AI / ML model or according to the prediction result of the AI / ML model.

[0309] In some embodiments of the present invention, the user equipment may send an uplink signaling to the network side to indicate that the UE side has enabled the switching method of the AI / ML model.

[0310] When the AI / ML model is located on the UE side (e.g., at least one of UEs 10a-10n), the scheme / condition for enabling the AI / ML model method may be any one of the following or a combination thereof:

[0311] (1) Solution 1: Network configuration:

[0312] 9 , the switching operation of enabling model reasoning based on the AI / ML model in Solution 1 is illustrated as follows.

[0313] Step B11: The network (eg, base station 20a or 20b) sends the configuration of the AI / ML model, and the UE receives the configuration of the AI / ML model.

[0314] Step B12: The UE executes model inference of the AI / ML model to generate prediction results to assist in handover decision-making.

[0315] Step B13: The UE starts a switching operation of model reasoning based on the AI / ML model according to the configuration of the AI / ML model.

[0316] Since the results of AI / ML model inference need to be reported to the network (e.g., base station 20a or 20b), in order to obtain appropriate reporting content, the network can enable the AI / ML model through network configuration and send the configured signaling to the UE. The configured signaling can be RRC information, DCI, MAC CE, system message, etc. There are several options:

[0317] Option 1: Implicit Indication. When the prediction result reporting configuration includes relevant parameter values ​​for the AI / ML model, this serves as an implicit indication to indicate that the AI / ML model is enabled. The relevant parameter values ​​for the AI / ML model include "predicted candidate / target cell," "values ​​required for handover prediction," model identification (model ID), functionality information (including functionality ID), and meta information.

[0318] Solution Option 2: Explicit Indication: The network issues an explicit indication to enable the AI / ML model. Upon receiving the indication, the UE enables AI / ML-based prediction.

[0319] Option 3: Explicit Indication: The network issues an explicit indication to enable the AI / ML model. Upon receiving the indication, the UE begins AI / ML-based prediction and suspends traditional measurement methods.

[0320] Option 4: Periodic Enable. The network configures the method for periodically enabling the AI / ML model, the time of the period, and the interval.

[0321] (2) Solution 2: Introduce a threshold 3 (threshold 3). When the prediction result is greater than the threshold 3, the AI / ML-based method is enabled. For example, the threshold 3 is the threshold of the predicted handover failure rate. In the past period of time, when the actual handover failure rate through traditional measurement is greater than the threshold, it indicates that the handover method through the traditional measurement scheme is inaccurate, and the UE enables the AI / ML-based method for prediction.

[0322] For example, the specific expression of the conditions for enabling AI / ML model-based methods is as follows:

[0323] HO failure rate + offset 3 > threshold 3 (3)

[0324] The formula (3) is explained as follows:

[0325] Handover failure rate: This refers to the probability of handover failure as measured by traditional methods. The handover failure rate can be time-based, number-based, or a combination thereof. For example, it can be the predicted number or proportion of handover failures over a period of time (e.g., one hour); or the cumulative predicted number or proportion of handover failures over the past n times (n is an integer greater than 1).

[0326] Threshold 3: indicates the threshold parameter that triggers the start of the AI / ML method.

[0327] Bias 3: indicates the bias parameter that triggers the start of the AI / ML method. Bias 3 is optional.

[0328] The handover failure rate, threshold 3, and offset 3 may be expressed in percentages. The above parameters may be configured, preconfigured, or predefined by the network side.

[0329] It should be understood that in the description of the embodiments of the present invention, the handover failure rate can also be equivalent to the number of handover failures. The number of handover failures, threshold 3, and offset 3 can be expressed as a number. The handover failure rate can be based on a cell or on the entire handover event. If a successful handover to any cell is considered a handover success. In solution 3, since the handover failure rate can be obtained by the network side or the UE root statistics, the AI / ML model method can be triggered to start by the network side or actively started by the UE side (for example, at least one of UE 10a-10n). If the AI / ML model method is triggered to start by the network side, the network sends model configuration information or model activation signaling, the content of which refers to solution 1. The UE receives the configuration signaling and can reply or not reply to the confirmation message. If it is actively turned on by the UE, an uplink signaling can be sent to the network side to indicate that the UE side (for example, at least one of UE 10a-10n) has turned on the AI / ML model method, and the auxiliary network (for example, base station 20a or 20b) performs appropriate configuration for the UE, as shown in Figure 10.

[0330] Specifically, the uplink signaling may be a new signaling. When the network (eg, base station 20a or 20b) receives the uplink signaling, it considers that the UE has enabled the AI / ML model method. The uplink signaling may be UCI, MAC CE, or RRC signaling.

[0331] Optionally, the UE can reuse the original protocol signaling, such as UE assistance information (UAI), MAC CE, etc., as the above-mentioned uplink signaling, but it is necessary to add an indication information to the original signaling to indicate that the UE has enabled the AI / ML model method. The indication information can be implemented in the following ways:

[0332] Method 1: Use 1 bit to indicate, a value of 1 means it is enabled, and a value of 0 means it is not enabled.

[0333] Method 2: Use 1 bit to indicate, the value of present means it is enabled, and the value of absent means it is not enabled.

[0334] Method 3: Use 1 bit to indicate, the value is true to indicate that it is enabled, and the value is false to indicate that it is not enabled.

[0335] Solution 3: Initiate an AI / ML model approach based on performance monitoring results. Performance monitoring can be performed by obtaining measured values ​​or statistical values ​​through traditional methods and comparing them with predicted values ​​derived through AI / ML reasoning. See Example 6 for a detailed description. When the monitoring results meet certain conditions / thresholds, the AI / ML model approach is initiated.

[0336] Solution 4: When the UE moves to the cell boundary, the quality (RSRP, RSRQ, SINR) of multiple cells can be measured by traditional measurement methods to meet the event-based reporting conditions or periodic reporting conditions. Since the quality of these cells is the same or similar, the network (such as base station 20a or 20b) may select an inappropriate cell based on its own evaluation, resulting in handover failure. Therefore, Solution 4 can introduce a threshold 4, and the threshold 4 can be based on the number of cells. Assuming that threshold 4 is configured to be 5 cells, when the number of cells with the same or similar cell quality is measured to be 7 by traditional methods, which is greater than the configured threshold 4, the AI / ML model is turned on for prediction to assist in handover decisions. The threshold 4 can be configured on the network side or predefined.

[0337] For example, the specific expression of the conditions for enabling AI / ML model-based methods is as follows:

[0338] Number of cells with the same or similar cell quality + offset 4 > threshold 4 (4)

[0339] The formula (4) is explained as follows:

[0340] Number of cells with the same or similar cell quality: refers to the number of cells with the same or similar cell quality obtained through traditional measurement methods.

[0341] Threshold 4: indicates the threshold parameter for triggering the start of the AI / ML method, which can be expressed in the number of cells.

[0342] Bias 4: indicates the bias parameter for triggering the AI / ML method. Bias 4 is optional.

[0343] After traditional measurements, the UE determines that the quality of multiple cells is the same or similar. Identical quality means that the quality of the cells is equal. Similar quality means that the quality of multiple cells is within the same range. There are two possible scenarios for similar quality:

[0344] Solution 1: The network configures or predefines a reference value and an offset value. Cells whose quality differs from the reference value by no more than the offset value are considered to have similar quality. Cells with similar cell quality are considered to have similar quality.

[0345] Solution 2: The network configures or predefines an offset value. Cells with quality differences no greater than the offset value are considered to have similar quality. Cells with similar cell quality are considered to have similar quality.

[0346] Example 4: Process of activating / deactivating AI / ML models via broadcast

[0347] Multiple UEs may carry the same AI / ML model, or AI / ML models with similar functionality. This is because AI / ML models are trained based on cells, areas, or configurations. Therefore, the network (e.g., base station 20a or 20b) can activate or deactivate a specific AI / ML model for multiple UEs, class, or group, through broadcasts (e.g., SIB messages, paging messages), thereby reducing signaling overhead.

[0348] In the current standard, neither paging nor System Information Block (SIB) messages define operations related to the AI / ML model. For example: the main purpose of paging messages is to page idle UEs and inactive UEs. Short paging is used to notify UEs in RRC_IDLE, RRC_INACTIVE and RRC_CONNECTED states. In addition, the use cases currently executed by the AI / ML model include supporting beam prediction for UEs in RRC active state, positioning enhancement for UEs in RRC inactive state, and cell selection / reselection for UEs that may be in RRC idle state. Therefore, the AI / ML model can work in all three states of the UE.

[0349] Figure 11 illustrates the process of activating a specific AI / ML model / class / group via broadcast signaling. The network (CN or base station) issues a broadcast message, where the broadcast message includes activation information for activating a specific AI / ML model / class / group, or deactivation information for deactivating a specific AI / ML model / class / group. UEs 10a-10n receive the broadcast message. When the broadcast message includes the activation information, one or more AI / ML models belonging to the specific AI / ML model / class / group are activated. When the broadcast message includes the deactivation information, one or more AI / ML models belonging to the specific AI / ML model / class / group are deactivated.

[0350] The network (CN or base station) sends a broadcast message to the UEs 10a-10n. The broadcast message includes identification information of the AI / ML model. The identification information of the AI / ML model includes at least one or any combination of the following information:

[0351] 1. Global Model ID: This is the unique identifier of the AI / ML model within the global scope. The global scope can represent a network, a network slice, or an operator.

[0352] 2. Local model ID: This is the unique identifier of the AI / ML model in the local network.

[0353] 3. Functionality information: includes a functionality ID and is used to indicate the functional type of the AI / ML model.

[0354] 4. Meta information: This information includes other information about the AI / ML model, such as model version and model creation time.

[0355] 5. Tracking area (TA): indicates the TA where the UE of the AI / ML model is located.

[0356] 6. Cell ID: Indicates the ID of the cell where the UE of the AI / ML model is located.

[0357] 7. Beam ID: Indicates the ID of the beam of the UE serving the AI / ML model.

[0358] 8. AI-A (AI-specific area): represents the area of ​​the AI / ML model.

[0359] 9. Base station ID (gNB ID): The ID of the base station (e.g., base station 20a or 20b) that serves the beam of the UE serving the AI / ML model.

[0360] The broadcast message may include a paging message, a short paging message, or a system information block (SIB) message. The paging message, short paging message, or system information block (SIB) message may include at least one of the following information about the AI / ML model(s):

[0361] Identification information;

[0362] AI / ML features; and

[0363] AI / ML status.

[0364] Activate / deactivate AI / ML models by broadcasting messages. Specific signaling options include:

[0365] (1) Mode 1: Multiplexing a short paging message or SIB message as activation information for activating an AI / ML model, or as deactivation information for deactivating an AI / ML model. The activation information includes an activation indication, which is associated with identification information of one or more AI / ML models. Similarly, the deactivation information includes a deactivation indication, which is associated with identification information of one or more AI / ML models. The UE 10a-10n receives a short paging message or SIB, which contains a model activation / deactivation indication (e.g., information element IE) and / or identification information of an AI / ML model, which is used to indicate activation / deactivation of the AI / ML model associated with the identification information of the AI / ML model.

[0366] An example is as follows:

[0367] Table 2: Short paging information

[0368] (2) Method 2: A new message (such as a short paging message, paging, or SIB message) is introduced. This message is specifically used to indicate the activation / deactivation of the AI / ML model. It carries an activation / deactivation indicator. For example, if it is set to 0, it deactivates the model, and if it is set to 1, it deactivates the model. In addition, the message also contains the identification information of the AI / ML model. An example is as follows:

[0369] Table 3

[0370] AI / ML-info: AI / ML-info is AI / ML information, which indicates the identification information of the AI / ML model.

[0371] AI / ML-status: AI / ML-status is the AI / ML status, indicating whether the AI / ML model is activated or deactivated. The AI / ML model is related to the AI / ML model identification information. If it is set to 0, it means the relevant AI / ML model is deactivated. If it is set to 1, it means the relevant AI / ML model is activated.

[0372] Method 3: Multiplex the short paging message and include an activation / deactivation indication associated with one or more messages related to the AI / ML model. The UE 10a-10n receives a paging message containing a model activation / deactivation information element (IE) and / or AI / ML model identification information, indicating activation / deactivation and the AI / ML model identifier. The AI / ML model identifier indicates the AI / ML model.

[0373] Example 5: Fallback to non-AI / ML operation

[0374] Fallback refers to the system's reversion from AI-based operations to non-AI-based operations. In the AI-related specifications of Release 18 (R18) of the 3GPP communication standard, the fallback operation is the monitoring output of AI / ML-based reasoning or the monitoring metrics of AI / ML-based reasoning. The fallback operation can be determined by the UE (e.g., at least one of UEs 10a-10n) or by the network (e.g., base station 20a or 20b).

[0375] Performance monitoring of AI / ML models affects fallback operations. Different model deployment locations (UE side or NW side) require different monitoring mechanisms to adapt to different network environments and user needs.

[0376] For the model on the UE side (e.g., at least one of UE 10a-10n), the UE can send different types of signaling, such as instructions, requests, or reports, to the base station (e.g., base station 20a or 20b) based on its own performance status. The UE can also autonomously decide when to use, activate, deactivate, switch, or fallback to the model based on its own evaluation. In order to make these decisions, the UE needs to have a way to detect whether the model is suitable for the current situation, such as by comparing the model output and the actual observation value. The base station (e.g., base station 20a or 20b) can guide the UE to perform performance measurements and / or reports through configuration or signaling, so that the base station (e.g., base station 20a or 20b) can also understand the model status of the UE. When designing such a mechanism, the complexity and power consumption of the UE should be taken into account.

[0377] For the network side model, the UE needs to report beam measurement values ​​to the base station (e.g., base station 20a or 20b) according to the instructions of the base station (e.g., base station 20a or 20b) so that the base station (e.g., base station 20a or 20b) can use the model for network optimization. This reporting can be done through RRC or L1 signaling, but this may require some modifications to existing specifications. The base station (e.g., base station 20a or 20b) is responsible for monitoring the performance indicators of the model and deciding when to use, activate, deactivate, switch, or fallback the model based on its own evaluation. When designing this mechanism, the performance and efficiency of the network should be taken into account.

[0378] A fallback mechanism is a special performance monitoring mechanism used to restore the model to normal operation when a problem occurs. This embodiment focuses on the fallback mechanism and related information elements for the candidate / target cell prediction use case in L3-based mobility.

[0379] For a UE side model (eg, at least one of the UEs 10a - 10n) (ie, the reasoning function is located on the UE side (eg, at least one of the UEs 10a - 10n)), the fallback decision may be determined by the UE or the network side.

[0380] When the UE side (e.g., at least one of UE 10a-10n) decides to fallback, the UE may send a message to the network side, which is used to notify / indicate the network side that the UE side (e.g., at least one of UE 10a-10n) has performed a fallback operation (including deactivating all AI / ML models). Optionally, it can also be used to notify / instruct the network side to close / suspend configuration and reporting related to AI-based methods, specifically, such as configuration and reporting of model reasoning, or configuration and reporting of model monitoring, or configuration and reporting related to AI-related data collection. Optionally, it can also be used to notify / instruct the network side to enable configuration and reporting related to traditional methods (such as RRM measurement reporting, LTM measurement reporting). To achieve the above functions of the message, there are several potential solutions:

[0381] (1) Solution 1: The UE has performed a fallback operation by explicitly indicating through a fallback indication message. The information is used to indicate to the network (e.g., base station 20a or 20b) that the UE side (e.g., at least one of UE 10a-10n) has performed a fallback operation; the specific implementation methods may be as follows. If the network (e.g., base station 20a or 20b) receives a report message from the UE containing fallback indication information, the network (e.g., base station 20a or 20b) determines whether the UE side (e.g., at least one of UE 10a-10n) has performed a fallback operation based on the indication information. The fallback indication information may be predefined or configured based on the network (e.g., base station 20a or 20b).

[0382] Mode 1: 1 bit indication, present means fallback is enabled, absent means fallback is not enabled

[0383] Method 2: 1 Bit indication, true means the fallback is enabled, false means the fallback is not enabled.

[0384] Mode 3: 1 Bit indication, set = 1 means the fallback is enabled, set = 0 means the fallback is not enabled,

[0385] (2) Solution 2: Through a reporting message, the UE explicitly or implicitly indicates that it has performed a fallback operation. The prediction results output by all model reasoning functions are set to empty or 0. For example, if the reporting amount in the reporting configuration reported by the network side to the UE is the predicted alternative / target cell, the predicted RSRP, or the predicted switching probability, etc. When the UE side (for example, at least one of UE 10a-10n) decides to perform a fallback operation, all reporting amounts are set to empty / or 0. When the network (for example, base station 20a or 20b) receives the predicted reporting amount in the predicted reporting message as empty / 0, it can be considered that the UE has performed a fallback operation, then the network (for example, base station 20a or 20b) can turn off or suspend the AI-related configuration, and / or send the traditional measurement configuration. The configured AI-based reporting information is as described in steps 1, 2 and 3 of the aforementioned general scheme.

[0386] (3) Solution 3: The reporting message includes fallback indication information (such as Solution 1) and the prediction result output by the model reasoning function. The reporting information output by the model reasoning function is the prediction result, see step 2 of the aforementioned general solution. The network (such as base station 20a or 20b) receives the reporting message from the UE including fallback indication information and the prediction result output by the model reasoning function. The network (such as base station 20a or 20b) determines whether the UE side (such as at least one of UE 10a-10n) has performed the fallback operation based on the indication information. Optionally, it is not excluded that the prediction result output by the model reasoning function is set to empty / 0.

[0387] The fallback indication information of the above-mentioned solution 1 and / or the prediction result output by the model reasoning function of solution 2 can be configured by the network side and can be carried on the RRC, or on the MAC CE, or on the DCI. Specifically, the prediction result output by the model reasoning function and the fallback indication information are shown in step 1 of the above-mentioned general solution.

[0388] Optionally, to save signaling overhead, the UE proactively sends a message via UCI or MAC CE. This message may be a dedicated fallback message. Upon receiving this message, the network (e.g., base station 20a or 20b) deems that the UE (e.g., at least one of UEs 10a-10n) has made a fallback decision. This message may also include a fallback indication, as described in Solution 1 above.

[0389] The process of the UE side (e.g., at least one of UEs 10a-10n) performing the fallback mechanism is shown in FIG12 , and the specific steps are as follows:

[0390] Step C1: The network (e.g., base station 20a or 20b) sends a configuration message related to the fallback operation to the UE. The configuration message can be a configuration message for model inference or a dedicated configuration message for fallback. The configuration message includes at least one of the following information: reporting configuration information including prediction results, and a configuration message including fallback indication information. The configuration message can be an RRC message, a MAC CE, or a DCI, etc.

[0391] Step C2: The UE may make a fallback decision based on monitoring results (e.g., model / performance monitoring) or based on calculation statistics of traditional measurements. The UE performs a fallback operation on the AI / ML model on the user equipment side.

[0392] Step C3: The UE reports indication information regarding the fallback operation. The UE sends a fallback-related reporting message, wherein the reporting message packet includes at least one of the following information: a fallback indication and an inferred reporting amount. The reporting message may be an RRC message (such as a UAI message), a MAC CE, or a UCI.

[0393] 13 , the process of performing the fallback mechanism on the network side is described below.

[0394] Step C12: The network performs a fallback operation on the AI / ML model on the network side.

[0395] Step C13: The network sends instruction information about the rollback operation.

[0396] It is important to understand that there is no logical order for all the above steps; all the above steps can be decoupled.

[0397] Example 6: Performance Monitoring

[0398] In order to ensure the adaptability of the AI / ML model, a performance monitoring function is introduced in the AI-related functions of the air interface (AI over air interface) of R18. In the performance monitoring, the inference output of the AI / ML model and the calculation results of the traditional measurement are obtained as the true value (ground truth). For example, by comparing the inference output of the model and the true value (ground truth), the accuracy of the result of the AI / ML model is judged, thereby judging the adaptability of the AI / ML model. In the use case of AI / ML aided mobility for network triggered L3-based handover for network triggered L3 layer switching operations, the new performance monitoring issues brought about by the introduction of new use cases are discussed, including the performance indicators (Performance metric(s)) of monitoring, the benchmark of monitoring, the process of monitoring, etc.

[0399] The following uses the candidate / target cell prediction in L3-based mobility use case as an example to illustrate the performance metric(s) of model monitoring and the performance benchmark for performance comparison.

[0400] The following options are available for performance monitoring, performance metric(s), and methods for the candidate / target cell prediction in L3-based mobility use case:

[0401] (1) Option 1: Key Performance Indicators related to candidate / target cell prediction accuracy, such as candidate / target cell prediction accuracy, HO failure rate prediction accuracy, and HO probability prediction accuracy.

[0402] (2) Option 2: KPIs related to link quality, such as throughput, RSRP, RSRQ, SINR, and hypothetical block error rate (BLER).

[0403] (3) Option 3: Performance metric(s) based on AI / ML input / output data distribution.

[0404] (4) Option 4: Evaluate the difference between the candidate / target cells by comparing the benchmark candidate / target cells (eg, those obtained by conventional measurements) for performance comparison with the predicted candidate / target cells.

[0405] Performance benchmarks used for performance comparison include:

[0406] (1) Option 1: The performance comparison benchmark for Option 1 may be the KPIs of candidate / target cells obtained by measuring a group of cells indicated by a base station (e.g., base station 20a or 20b). For example, in performance monitoring, the KPIs of candidate / target cells obtained through traditional measurements are compared with the KPIs of candidate / target cells predicted by the AI / ML model.

[0407] (2) Option 2: The performance comparison benchmark for Option 2 can be measured using KPIs related to link quality. For example, in performance monitoring, the KPIs related to link quality obtained through measurement are compared with the KPIs related to link quality predicted by the AI / ML model.

[0408] (3) Option 3: Option 3 does not require a performance benchmark.

[0409] (4) Option 4: The performance comparison benchmark for Option 4 can be a suitable candidate / target cell corresponding to the model output. For example, in performance monitoring, the actual candidate cell is compared with the candidate / target cell predicted by the AI / ML model.

[0410] For example, if the performance metric(s) adopts the above-mentioned solution option 1:

[0411] When the relevant key performance indicator (KPI) is the HO failure rate prediction accuracy, the HO failure rate prediction accuracy can be judged by comparing the actual number of HO failures with the predicted number of HO failures over a period of time. The accuracy can be accurately expressed in terms of accuracy level or percentage. For example, the accuracy level includes high, medium and low levels, and high, medium and low can correspond to a range of percentages. For example, in the past 10 days, the actual number of HO failures was 15 times, and the number of predicted HO failures predicted by the AI / ML model was 16 times. When the KPI requires an error value of 2, since the difference between the actual number of HO failures and the predicted number of HO failures is less than 2 times, it can be considered that the accuracy of the handover predicted by the AI / ML model is high. Alternatively, its accuracy is directly calculated as (1-1 / 15)%, or (1-1 / 16)%.

[0412] When the relevant key performance indicator (KPI) is the HO probability prediction accuracy (HO probability prediction accuracy), the HO probability prediction accuracy (HO probability prediction accuracy) can be judged by comparing the actual number of switches with the predicted number of switches over a period of time in the past. The accuracy can be accurately expressed in terms of accuracy level, or percentage. For example, the accuracy level includes high, medium, and low levels, and high, medium, and low can correspond to a range of percentages. For example, in the past 10 days, the actual number of switches was 15 times, and the number of switches predicted by the AI / ML model was 16 times. When the KPI requires an error value of 2, since the difference between the actual number of switches and the predicted number of switches is less than 2 times, it can be considered that the accuracy of the AI / ML model prediction is high. Alternatively, its accuracy is directly calculated as (1-1 / 15)%, or (1-1 / 16)%.

[0413] Optionally, a comparison is performed based on the cells that the UE passes through. For example, if the UE passes through 300 cells, the actual number of handovers is 15, and the predicted number of handovers is 16. When the KPI requires an error value of 2, since the difference between the actual number of handovers and the predicted number of handovers is less than 2, it can be considered that the accuracy of the AI / ML model prediction is high. Alternatively, its accuracy can be directly calculated as (1-1 / 15)%, or (1-1 / 16)%.

[0414] When the relevant key performance indicator (KPI) is candidate / target cell prediction accuracy, the candidate / target cell prediction accuracy can be judged by comparing the actual measured candidate / target cell with the predicted candidate / target cell over a period of time in the past or the number of predictions. Accuracy can be accurately expressed in terms of accuracy level or percentage. For example, the accuracy level includes high, medium and low levels, and high, medium and low can correspond to a range of percentages. For example, in the past 15 measurements, the actual candidate / target cell and the candidate / target cell predicted by the AI / ML model were the same 13 times. When the KPI requires an error value of 2, since the difference between the actual number of switching and the predicted number of switching is less than 2 times, it can be considered that the accuracy of the AI / ML model prediction is high. Alternatively, its accuracy is directly calculated as (1-1 / 15)%.

[0415] For example, if the performance metric(s) adopts option 4 above: the quality of all predicted candidate / target cells can be measured using non-AI methods to compare the accuracy of the predicted optimal candidate / target cell; or a performance benchmark or reference information can be obtained using non-AI methods and compared with the predicted candidate / target cell. For example, the monitoring result of the KPI is the candidate / target cell prediction accuracy. If it exceeds 90%, the AI / ML model result is considered accurate.

[0416] Note that for the above-mentioned performance monitoring parameters, one or more thresholds are configured or predefined on the network side, such as the threshold for candidate / target cell prediction accuracy, the threshold for HO failure rate prediction accuracy, the threshold for HO probability prediction accuracy, and the threshold for comparing the predicted candidate / target cell with the measured candidate / target cell. The AI / ML model method can be enabled based on the threshold. For example, a threshold for HO failure rate prediction accuracy is defined. If the threshold is 95%, and the percentage of the actually monitored HO failure rate prediction accuracy is greater than the defined / configured threshold, it can generally be considered that the AI / ML model method is very effective, and the AI / ML model method can be enabled in the system. The same applies to other thresholds. The network side configuration can be through RRC, DCI, MAC CE, etc.

[0417] 14 , the process of performance monitoring is described as follows.

[0418] Step D101: The network node or UE performs performance monitoring on the AI / ML model according to the configuration of performance monitoring on the AI / ML model, and generates a monitoring result of the performance monitoring.

[0419] Step D102: The network node and the UE exchange monitoring results of the performance monitoring of the AI / ML model.

[0420] The monitoring is performed on the UE side (e.g., at least one of the UEs 10a-10n). The monitoring process is shown in FIG15 . The specific steps are as follows:

[0421] Step D1: The network (e.g., base station 20a or 20b) sends a performance monitoring configuration message to a user equipment (UE). The configuration message includes at least one of the following information:

[0422] (1) Performance indicators: used to indicate the standards for measuring performance, such as the KPI.

[0423] (2) Benchmark / reference: used to indicate the standard for comparison with performance indicators, such as time obtained by traditional measurements, historical data, industry standards, etc.

[0424] (3) Performance parameter threshold: used to indicate the allowable range of performance indicators.

[0425] (4) Monitoring start time: used to indicate the time when monitoring starts.

[0426] (5) Monitoring execution time: used to indicate the duration of monitoring.

[0427] (6) Monitoring reporting time: used to indicate the time when the monitoring results are reported.

[0428] (7) Monitoring object: used to indicate the object that needs to be monitored, such as AI / ML model, functionality, etc.

[0429] The monitoring object may include at least one of the following:

[0430] (1) AI / ML model ID (model ID); and

[0431] (2) Functionality ID.

[0432] The performance monitoring configuration message may be DCI, MAC CE, or RRC.

[0433] Step D2: The user equipment (UE) performs performance monitoring, calculation, or comparison. The UE performs performance monitoring based on the performance indicators, methods, and performance comparison benchmarks configured in step 1 and generates a performance monitoring result. The performance monitoring result (referred to as the monitoring result) indicates whether the monitored AI / ML model or function is suitable.

[0434] Step D3: The user equipment (UE) reports the monitoring result of the performance monitoring and sends a performance monitoring report message to the network (e.g., base station 20a or 20b). The performance monitoring report message includes at least one of the following information: (1) performance metric(s) / method, (2) monitoring result, or (3) lifecycle management operation.

[0435] (1) Performance metric(s) / method: Performance metric(s) / method is used to represent the performance of the monitored object. Detailed description is given above.

[0436] (2) Monitoring results: The monitoring results indicate whether the monitored object meets the requirements. The results of monitoring, calculation, and comparison are used to display whether the AI / ML model / function is adapted. The monitoring result is a result obtained based on a performance metric(s) / method. For example, if the HO failure rate prediction accuracy of one or more AI / ML models is low, the monitoring result indicates that the monitored AI / ML model or AI function is incompatible, and a corresponding incompatibility indication of the AI / ML model is reported. The incompatibility indication can be mapped one-to-one or one-to-many to the AI / ML model or AI / ML function.

[0437] (3) Lifecycle management operations: Lifecycle management operations are used to represent the state or operation of the monitored object. The lifecycle management operations include at least one of the following:

[0438] A. Model activation: Deploy the model to the UE and start using it.

[0439] B. Model deactivation: Deactivate the model from the UE and stop using it.

[0440] C. Model update: Deploy the latest version of the model to the UE, replacing the existing version.

[0441] D. Model switching: Switching from one model to another to deal with situations such as model performance degradation or failure.

[0442] E. Model selection: Select a model from multiple models for UE to meet the needs of different scenarios.

[0443] F. Rollback: Switch the model to a previous version or switch to traditional measurement methods to solve problems that occur after the model is launched or updated.

[0444] For example, if the reporting message includes at least one of any of the above operation instructions or related information (such as the fallback operation in Example 5), it can implicitly indicate whether the monitored AI / ML model or AI function is adapted or not.

[0445] For example, if the handover failure rate prediction accuracy percentage is greater than a defined / configured threshold, the AI / ML model method may be enabled. The UE sends a notification indicating the AI / ML model is enabled to the network (e.g., base station 20a or 20b). The notification indicating the AI / ML model is enabled is optional.

[0446] The monitoring is located on the network side (e.g., base station 20a or 20b). The monitoring process is shown in Figure 16 below. The specific steps are as follows:

[0447] Step 1: The network (e.g., base station 20a or 20b) collects performance monitoring data. The data may include data obtained through traditional measurement methods, data predicted through AI methods, or data calculated by the UE.

[0448] Step 2: The network (eg, base station 20a or 20b) performs performance monitoring. The network (eg, base station 20a or 20b) may perform performance monitoring based on implementation. The performance monitoring may include calculation / comparison.

[0449] Step 3: The network (e.g., base station 20a or 20b) sends the performance monitoring result to the UE as a feedback message. The content of the feedback message is the same as step 3 in the monitoring process performed by the UE (e.g., at least one of UEs 10a-10n). The feedback message includes at least one of the following information:

[0450] Performance metric(s) / methods;

[0451] The monitoring results; and

[0452] Lifecycle management operations.

[0453] Referring to Figure 17, UE 10 may include a processor 11a, a memory 12a, and a transceiver 13a. The processor 11a is configured to invoke and execute a computer program stored in the memory 12a, causing the UE 10, in which the processor 11 is installed, to perform the disclosed methods, steps, and / or UE functions. UE 10 is an example of a UE described herein (e.g., one of UEs 10a-10n). The transceiver 13a may include baseband circuitry and radio frequency (RF) circuitry.

[0454] Referring to Figure 18 , network node 200 is a network device that may include a processor 21a, a memory 22a, and a transceiver 23a. The processor 21a is configured to invoke and execute a computer program stored in the memory 22a, causing the network node 200, in which the processor 21a is installed, to perform methods, steps, and / or functions of a network node. Network node 200 is an example of a CN network entity, network node, radio node, base station, or gNB described herein. The transceiver 23a may include baseband circuitry and radio frequency (RF) circuitry.

[0455] 19 , the present embodiment further provides a chip 70. This chip 70 may correspond to the user device 10 in the present embodiment, and the chip 70 may implement the corresponding processes implemented by the user device 10 in the various methods in the present embodiment. The chip 70 includes a processor 71, which may call and execute computer programs from memory to implement the methods in the present embodiment.

[0456] Optionally, the chip 70 may further include a memory 72. The processor 71 may call and execute a computer program from the memory 72 to implement the method in the embodiment of the present application.

[0457] The memory 72 may be a separate device independent of the processor 71 , or may be integrated into the processor 71 .

[0458] Optionally, the chip 70 may further include an input interface 73. The processor 71 may control the input interface 73 to communicate with other devices or chips, and specifically, may obtain information or data sent by other devices or chips.

[0459] Optionally, the chip 70 may further include an output interface 74. The processor 71 may control the output interface 74 to communicate with other devices or chips, and specifically, may output information or data to other devices or chips.

[0460] 20 , another embodiment of the present application further provides another chip 80. This chip 80 may correspond to the network node 200 in the embodiment of the present application, and this chip 80 may implement the corresponding processes implemented by the network node 200 in each method in the embodiment of the present application. This chip 80 includes a processor 81, which may call and execute computer programs from a memory 82 to implement the methods in the embodiment of the present application.

[0461] Optionally, the chip 80 may further include a memory 82. The processor 81 may call and execute a computer program from the memory 82 to implement the method in the embodiment of the present application.

[0462] The memory 82 may be a separate device independent of the processor 81 , or may be integrated into the processor 81 .

[0463] Optionally, the chip 80 may further include an input interface 83. The processor 81 may control the input interface 83 to communicate with other devices or chips, and specifically, may obtain information or data sent by other devices or chips.

[0464] Optionally, the chip may further include an output interface 84. The processor 81 may control the output interface 84 to communicate with other devices or chips, and specifically, may output information or data to other devices or chips.

[0465] Those skilled in the art will appreciate that the above-mentioned embodiments are specific embodiments for implementing the present application, and that in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present application.

Claims

1. A handover optimization method based on an artificial intelligence (AI) / ML model, executed in a user device, characterized in that: The method includes: reporting prediction results based on AI / ML model reasoning; The prediction result includes at least one of the following: a predicted switching probability, a predicted switching time, a predicted switching failure rate, and a time of the prediction result.

2. The method according to claim 1, characterized in that The method further comprises: Receive a configuration message about the AI / ML model sent from the base station, where the configuration message may include a reporting configuration for a prediction result of the AI / ML model reasoning, and the configuration message may include at least one of the following: Predicted candidate / target cells, Predicted beam, The predicted reference signal received power RSRP, The predicted reference signal received power RSRQ, Predicted signal-to-interference-and-noise ratio SINR, Predicted handover failure rate, Predicted switching ping-pong, Predicted measurement events, Predicted switching probability, Predicted switching time, The time to predict the outcome, and How to report prediction results, etc.

3. The method according to claim 1, characterized in that The predicted handover probability is based on the handover probability of the cell, or based on the handover probability of the beam, or based on the handover probability of the entire handover event.

4. The method according to claim 1, wherein The prediction result is a prediction result representing one candidate / target cell or multiple candidate / target cells.

5. The method according to claim 1, wherein The predicted switching time is a predicted switching execution time, and the predicted switching time includes at least one of the following: A time offset T of the switching indication, used to indicate that the switching is performed after the time offset T; and The time window W of the switching indication is used to indicate that the switching is performed within the time window W.

6. The method according to claim 1, wherein The reporting method of the prediction result includes at least one of the following: Based on event triggering, reporting the prediction results; Based on the period, the prediction result is reported.

7. The method according to claim 6, characterized in that The event trigger is based on configuration or the event trigger is predefined.

8. The method according to claim 6, characterized in that The period is based on configuration or the period is predefined.

9. The method according to claim 6, wherein the event triggering comprises at least one of the following: When the predicted handover probability is greater than a first threshold after being adjusted by the first offset value, the condition for triggering the event is met; The predicted handover failure rate meets the event triggering condition when it is less than the second threshold after being adjusted by the second offset value; or The time reaches the time position or time range indicated by the time indication information of the predicted switching time.

10. The method according to claim 1, characterized in that include: Sending a reporting message according to a reporting period to report the prediction result inferred by the AI / ML model; The configuration of the reporting period includes at least one of the following: Reporting interval, reporting quantity, maximum number of reporting cells, inference amount of reporting cells, index of reported reference signals, maximum number of reference signals, and allowed cell list.

11. The method according to claim 1, characterized in that The conditions for enabling the model inference function of the AI / ML model include at least one of the following: receiving configuration information of a base station; Satisfy predefined or configured conditions; or The performance monitoring results of the AI / ML model meet predefined or configured thresholds.

12. The method according to claim 1, characterized in that The activation methods of the AI / ML model include: The AI / ML model is activated by broadcasting a message, wherein the AI / ML model is a single / class / group of AI / ML models.

13. The method according to claim 1, wherein Methods for deactivating the AI / ML model include: Deactivate the AI / ML model by broadcasting a message, wherein the AI / ML model is a single / class / group of AI / ML models.

14. The method according to claim 12 or 13, characterized in that The broadcast message includes at least one of the following: paging information, short paging information and system information block SIB message.

15. The method according to claim 14, characterized in that The paging message, short paging message or system information block (SIB) message includes at least one of the following information of the one / class / group of AI / ML models: Identification information; AI / ML features; and AI / ML status.

16. The method according to claim 1, wherein The fallback mechanism of the AI / ML model includes: Report instruction information about the fallback operation to instruct the system to fall back from the operation based on the AI / ML model to the operation based on the non-AI / ML model.

17. The method according to claim 16, characterized in that The indication information of the fallback operation is fallback indication information, wherein a first value of the fallback indication information indicates that fallback is enabled, and a second value of the fallback indication information indicates that fallback is not enabled.

18. The method according to claim 16, characterized in that The indication information of the fallback operation is about the AI / ML model reporting message on the user equipment side, wherein the prediction result in the reporting message is empty, indicating that the fallback is enabled, and the prediction result in the reporting message is non-empty, indicating that the fallback is not enabled.

19. The method according to claim 16, wherein The user equipment makes a fallback decision based on a monitoring result of performance monitoring or based on calculated statistics of measurements.

20. The method according to claim 1, wherein The monitoring mechanism of the AI / ML model includes: Receive performance monitoring configurations for AI / ML models and generate performance monitoring results; and Report the monitoring results of performance monitoring related to the AI / ML model.

21. The method according to claim 20, characterized in that The performance monitoring configuration includes at least one of the following information: Performance indicators; Performance comparison benchmark / reference; performance parameter thresholds; Monitoring start time; Monitor execution time; Monitoring reporting time; Monitor object.

22. The method according to claim 21, characterized in that The performance indicators include at least one of the following: Key performance indicators (KPIs) related to candidate / target cell prediction accuracy; Key performance indicators (KPIs) related to link quality; Performance metrics based on AI / ML input / output data distribution; and The difference between the candidate / target cells indicates the difference between the candidate / target cells evaluated by comparing the benchmark candidate / target cells for performance comparison with the predicted candidate / target cells.

23. The method according to claim 21, characterized in that The performance parameter threshold includes at least one of the following: Thresholds for key performance indicators (KPIs) related to candidate / target cell prediction accuracy; Thresholds for key performance indicators (KPIs) related to link quality; and The threshold for the difference between candidate and target cells.

24. The method according to claim 20, characterized in that The performance monitoring is performed in the user equipment, and the reporting of the monitoring result of the performance monitoring of the AI / ML model includes: Transmit a performance monitoring report message, wherein the performance monitoring report message includes at least one of the following information: performance indicators / methods; The monitoring results; and Lifecycle management operations.

25. The method according to claim 1, wherein The performance monitoring is performed in the base station. The monitoring mechanism of the AI / ML model includes: Receive monitoring results of performance monitoring related to the AI / ML model.

26. A handover optimization method based on artificial intelligence (AI) / ML model, executed in a base station, characterized in that: The method comprises: Receive prediction results based on AI / ML model reasoning; The prediction result includes at least one of the following: a predicted switching probability, a predicted switching time, a predicted switching failure rate, and a time of the prediction result.

27. The method according to claim 26, characterized in that The method further comprises: Send a configuration message about the AI / ML model. The configuration message may include a configuration for reporting prediction results of the AI / ML model inference. The configuration message may include at least one of the following: Predicted candidate / target cells, Predicted beam, The predicted reference signal received power RSRP, The predicted reference signal received power RSRQ, Predicted signal-to-interference-and-noise ratio SINR, Predicted handover failure rate, Predicted switching ping-pong, Predicted measurement events, Predicted switching probability, Predicted switching time, The time to predict the outcome, and How to report prediction results, etc.

28. The method according to claim 26, characterized in that The predicted handover probability is based on the handover probability of the cell, or based on the handover probability of the beam, or based on the handover probability of the entire handover event.

29. The method according to claim 26, wherein The prediction result is a prediction result representing one candidate / target cell or multiple candidate / target cells.

30. The method according to claim 26, wherein The predicted switching time is a predicted switching execution time, and the predicted switching time includes at least one of the following: A time offset T of the switching indication, used to indicate that the switching is performed after the time offset T; and The time window W of the switching indication is used to indicate that the switching is performed within the time window W.

31. The method according to claim 26, wherein Receiving the report of the prediction result is based on at least one of the following methods: Based on event triggering, receiving a report of the prediction result; Based on a period, a report of the prediction result is received.

32. The method according to claim 31, characterized in that The event trigger is based on configuration or the event trigger is predefined.

33. The method according to claim 31, wherein The period is based on configuration or the period is predefined.

34. The method according to claim 31, wherein the event triggering comprises at least one of the following: When the predicted handover probability is greater than a first threshold after being adjusted by the first offset value, the condition for triggering the event is met; The predicted handover failure rate meets the event triggering condition when it is less than the second threshold after being adjusted by the second offset value; or The time reaches the time position or time range indicated by the time indication information of the predicted switching time.

35. The method according to claim 26, wherein include: Receiving a report message according to a reporting period, wherein the report message includes the prediction result of the AI / ML model reasoning; The configuration of the reporting period includes at least one of the following: Reporting interval, reporting quantity, maximum number of reporting cells, inference amount of reporting cells, index of reported reference signals, maximum number of reference signals, and allowed cell list.

36. The method according to claim 26, wherein The conditions for enabling the model inference function of the AI / ML model include at least one of the following: Configuration information of the above base station; Satisfy predefined or configured conditions; The performance monitoring results of the AI / ML model meet predefined or configured thresholds.

37. The method according to claim 26, wherein The activation methods of the AI / ML model include: The AI / ML model is activated by broadcasting a message, wherein the AI / ML model is a single / class / group of AI / ML models.

38. The method according to claim 26, wherein Methods for deactivating the AI / ML model include: Deactivate the AI / ML model by broadcasting a message, wherein the AI / ML model is a single / class / group of AI / ML models.

39. The method according to claim 37 or 38, characterized in that The broadcast message includes at least one of the following: paging information, short paging information and system information block SIB message.

40. The method according to claim 39, wherein The paging message, short paging message or system information block (SIB) message includes at least one of the following information of the one / class / group of AI / ML models: Identification information; AI / ML features; and AI / ML status.

41. The method according to claim 26, wherein The fallback mechanism of the AI / ML model includes: Instruction information about the fallback operation is received, wherein the instruction information about the fallback operation instructs the system to fall back from the operation based on the AI / ML model to the operation based on the non-AI / ML model.

42. The method according to claim 41, wherein The indication information of the fallback operation is fallback indication information, wherein a first value of the fallback indication information indicates that fallback is enabled, and a second value of the fallback indication information indicates that fallback is not enabled.

43. The method according to claim 41, wherein The indication information of the fallback operation is about the AI / ML model reporting message on the user equipment side, wherein the prediction result in the reporting message is empty, indicating that the fallback is enabled, and the prediction result in the reporting message is non-empty, indicating that the fallback is not enabled.

44. The method according to claim 26, wherein The monitoring mechanism of the AI / ML model includes: Sending configurations related to AI / ML model performance monitoring to user devices; and Receive monitoring results of performance monitoring related to the AI / ML model.

45. The method according to claim 44, wherein The performance monitoring configuration includes at least one of the following information: Performance indicators; Performance comparison benchmark / reference; performance parameter thresholds; Monitoring start time; Monitor execution time; Monitoring reporting time; Monitor object.

46. The method according to claim 45, characterized in that The performance indicators include at least one of the following: Key performance indicators (KPIs) related to candidate / target cell prediction accuracy; Key performance indicators (KPIs) related to link quality; Performance metrics based on AI / ML input / output data distribution; and The difference between the candidate / target cells indicates the difference between the candidate / target cells evaluated by comparing the benchmark candidate / target cells for performance comparison with the predicted candidate / target cells.

47. The method according to claim 45, wherein The performance parameter threshold includes at least one of the following: Thresholds for key performance indicators (KPIs) related to candidate / target cell prediction accuracy; Thresholds for key performance indicators (KPIs) related to link quality; and The threshold for the difference between candidate and target cells.

48. The method according to claim 44, wherein The performance monitoring is executed in the user equipment, and the receiving of the monitoring result of the performance monitoring related to the AI / ML model includes: Receive a performance monitoring report message, where the performance monitoring report message includes at least one of the following information: performance indicators / methods; The monitoring results; and Lifecycle management operations.

49. The method according to claim 26, wherein The performance monitoring is performed in the base station. The monitoring mechanism of the AI / ML model includes: Send monitoring results of performance monitoring related to the AI / ML model.

50. A wireless communication device, characterized in that: include: A processor configured to call and execute a computer program stored in a memory so that a device equipped with the processor executes the method of any one of claims 1 to 49.

51. A chip, characterized in that: include: A processor configured to call and execute a computer program stored in a memory so that a device equipped with the processor executes the method of any one of claims 1 to 49.

52. A computer-readable storage medium, characterized in that A computer program is stored therein, wherein the computer program causes a computer to execute the method of any one of claims 1 to 49.

53. A computer program product, characterized in that comprising a computer program, wherein the computer program causes a computer to execute the method of any one of claims 1 to 49.

Images