Method and apparatus for implementing AI-ML in wireless network

Abstract

A method, apparatus, and computer readable medium are described for storing instructions for a user terminal and a base station to update an AI / ML configuration in the event of a handover. A method performed by a user equipment includes operating a first AI / ML configuration in a coverage area of a first base station; receiving AI / ML configuration information indicating a second AI / ML configuration; a second AI / ML configuration indicated by the AI / ML configuration information is operated in the coverage area of the second base station. Operating the first AI / ML configuration includes operating the first AI / ML model or a first AI / ML model having the first AI / ML model configuration in a coverage area of the first base station. The first AI / ML model is associated with a first AI / ML model identifier, and the first AI / ML model configuration is associated with a first AI / ML model configuration identifier. Operating the second AI / ML configuration includes operating the second AI / ML model or the first AI / ML model with the second configuration in a coverage area of the second base station. The second AI / ML model is associated with a second AI / ML model identifier, and the second AI / ML model configuration is associated with a second AI / ML model configuration identifier. Indicating the second AI / ML configuration includes indicating a second AI / ML model identifier and / or a second AI / ML model configuration identifier.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of Indian Application No. 202211047358 filed on August 19, 2022, the contents of which are incorporated herein by reference as if fully set forth. Technical Field

[0002] RP-213599, TSG RAN 94e Meeting, December 6-17, 2021, available at: https: / / www.3gpp.orq / ftp / TSG_RAN / TSG_RAN / TSGR_94e / Docs / RP-213599.zip

[0003] TR 38.901V17.0.0

[0004] O-RAN Massive MIMO Use Case Technical Report 1.0, July 2022

[0005] The draft report of the meeting minutes is TSG RAN TSGR1_109-e, available at: https: / / www.3gpp.org / ftp / TSG_RAN / WG1_RL1 / TSGR1_109-e / Report / Draft_Minutes_report_ RAN1%23109-e_v030.zip

[0006] Deep Learning-Based CSI Feedback Method for Time-Varying Massive MIMO Channels, Tianqi Wang, Chaokai Wen, Jin Shi, and Yeli Li (Published on July 31, 2018)

[0007] WO2021051362A1

[0008] US20210345134A1

[0009] US20190277957A1 Summary of the invention

[0010] A method, apparatus, and computer-readable medium for storing instructions for updating AI / ML configurations in a user terminal and a base station in the event of a handover are described. In an embodiment, a method performed by a user equipment (UE) includes: operating a first AI / ML model with a first AI / ML model configuration within the coverage of a first base station (BS1); transmitting one or more UE parameters to BS1, the one or more UE parameters including UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of a second base station (BS2); and receiving, by the UE, a first signaling from BS1, the first signaling including information about a second AI / ML model configuration; wherein the first signaling is transmitted by BS1 based on a determination of a potential handover of the UE from the coverage of BS1 to the coverage of BS2; wherein the determination of the handover is based on the UE parameters. The information about the second AI / ML model configuration includes a second AI / ML model configuration identifier. The first signaling is received in a handover command message.

[0011] In an embodiment, a method performed by a user equipment (UE) includes: operating a first AI / ML model with a first AI / ML model configuration within the coverage of a first base station (BS1); transmitting one or more UE parameters to BS1, the one or more UE parameters including UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of a second base station (BS2); receiving, by the UE, first signaling from BS1, the first signaling including information about a handover to a second base station (BS2); wherein the first signaling is transmitted by BS1 based on a determination of a potential handover of the UE from the coverage of BS1 to the coverage of a second base station (BS2); wherein the determination of the handover is based on the UE parameters; transmitting a PRACH to BS2 by the UE; and receiving, by the UE from BS2, a second signaling including information about a second AI / ML model configuration. The information about the second AI / ML model configuration includes a second AI / ML model configuration identifier. The first signaling is received in a handover command message.

[0012] In an embodiment, an apparatus comprises: an AI / ML engine configured to operate a first AI / ML model having a first AI / ML model configuration within a coverage area of ​​a first base station (BS1); a transceiver configured to transmit one or more UE parameters to BS1, the one or more UE parameters comprising UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of a second base station (BS2); the transceiver configured to receive a first signaling from BS1, the first signaling comprising information about a second AI / ML model configuration; wherein the first signaling is transmitted by BS1 based on a determination of a potential handover of the UE from the coverage area of ​​BS1 to the coverage area of ​​BS2; wherein the determination of the handover is made based on the UE parameters; and the information about the second AI / ML model configuration comprises a second AI / ML model configuration identifier. The first signaling is received in a handover command message.

[0013] In an embodiment, an apparatus comprises: an AI / ML engine configured to operate a first AI / ML model having a first AI / ML model configuration within a coverage area of ​​a first base station (BS1); a transceiver configured to transmit one or more UE parameters to BS1, the one or more UE parameters including UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of a second base station (BS2); the transceiver is configured to receive a first signaling from BS1, the first signaling including handover information to a second base station (BS2); wherein the first signaling is transmitted by BS1 based on a determination of a potential handover of the UE from the coverage area of ​​BS1 to the coverage area of ​​the second base station (BS2); wherein the determination of the handover is based on the UE parameters; the transceiver is configured to transmit a PRACH to BS2; and receive a second signaling from BS2, the second signaling including information about a second AI / ML model configuration. The information about the second AI / ML model configuration includes a second AI / ML model configuration identifier. The first signaling is received in a handover command message.

[0014] In an embodiment, a method performed by a user equipment (UE) includes: operating a first AI / ML model or a first group of AI / ML models within the coverage of a first base station (BS1); transmitting one or more UE parameters to BS1, the one or more UE parameters including UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of a second base station (BS2); and receiving, by the UE from BS1, first signaling, the first signaling including information about a second AI / ML model or a second group of AI / ML models; wherein the first signaling is transmitted by BS1 based on a determination of a potential handover of the UE from the coverage of BS1 to the coverage of BS2; wherein the determination of the handover is based on the UE parameters; and the information about the second AI / ML model configuration includes a second AI / ML model identifier or an identifier of the second group of models. The first signaling is received in a handover command message.

[0015] In an embodiment, a method performed by a user equipment (UE) includes: operating a first AI / ML model or a first group of AI / ML models within the coverage of a first base station (BS1); transmitting one or more UE parameters to BS1, the one or more UE parameters including UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of a second base station (BS2); receiving, by the UE, first signaling from BS1, the first signaling including information about a handover to a second base station (BS2); wherein the first signaling is transmitted by BS1 based on a determination of a potential handover of the UE from the coverage of BS1 to the coverage of a second base station (BS2); wherein the determination of the handover is based on the UE parameters; transmitting a PRACH to BS2 by the UE; and receiving, by the UE from BS2, second signaling including information about a second AI / ML model or a second group of AI / ML models. The information about the second AI / ML model configuration includes a second AI / ML model identifier or an identifier of a second group of models. The first signaling is received in a handover command message.

[0016] In an embodiment, an apparatus comprises: an AI / ML engine configured to operate a first AI / ML model or a first set of AI / ML models within a coverage area of ​​a first base station (BS1); a transceiver configured to transmit one or more UE parameters to BS1, the one or more UE parameters including UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of a second base station (BS2); the transceiver configured to receive a first signaling from BS1, the first signaling including information about a second AI / ML model or a second set of AI / ML models; wherein the first signaling is transmitted by BS1 based on a determination of a potential handover of the UE from the coverage area of ​​BS1 to the coverage area of ​​BS2; wherein the determination of the handover is made based on the UE parameters; and the information about the second AI / ML model configuration includes a second AI / ML model identifier or an identifier of the second set of models. The first signaling is received in a handover command message.

[0017] In an embodiment, an apparatus comprises: an AI / ML engine configured to operate a first AI / ML model or a first set of AI / ML models within the coverage of a first base station (BS1); a transceiver configured to transmit one or more UE parameters to BS1, the one or more UE parameters including UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of a second base station (BS2); the transceiver configured to receive a first signaling from BS1, the first signaling including handover information to a second base station (BS2); wherein the first signaling is transmitted by BS1 based on a determination of a potential handover of the UE from the coverage of BS1 to the coverage of a second base station (BS2); wherein the determination of the handover is based on the UE parameters; the transceiver is configured to transmit a PRACH to BS2; and receive a second signaling from BS2, the second signaling including information about a second AI / ML model or a second set of AI / ML models. The information about the second AI / ML model configuration includes a second AI / ML model identifier or an identifier of a second set of models. The first signaling is received in a handover command message.

[0018] In an embodiment, a method performed by a user equipment includes: operating a first AI / ML configuration in a coverage area of ​​a first base station; receiving AI / ML configuration information indicating a second AI / ML configuration; operating the second AI / ML configuration indicated by the AI / ML configuration information in the coverage area of ​​the second base station. Operating the first AI / ML configuration includes operating a first AI / ML model or a first AI / ML model with a first AI / ML model configuration in the coverage area of ​​the first base station. The first AI / ML model is associated with a first AI / ML model identifier, and the first AI / ML model configuration is associated with a first AI / ML model configuration identifier. Operating the second AI / ML configuration includes operating a second AI / ML model or a first AI / ML model with a second configuration in the coverage area of ​​the second base station. The second AI / ML model is associated with a second AI / ML model identifier, and the second AI / ML model configuration is associated with a second AI / ML model configuration identifier. Indicating the second AI / ML configuration includes indicating the second AI / ML model identifier and / or the second AI / ML model configuration identifier. The AI / ML configuration information may be received in a handover command message, or the AI / ML configuration information may be received in a RACH response message. The AI / ML configuration information may be received in a message transmitted by the first base station, or the AI / ML configuration information may be received in a message transmitted by the second base station. The AI / ML configuration information is received after initiating a handover procedure in the first base station for handing over the user equipment to the coverage area of ​​the second base station, or the AI / ML configuration information is received before initiating a handover procedure in the first base station for handing over the user equipment to the coverage area of ​​the second base station.

[0019] In an embodiment, a method performed by a first base station includes: transmitting AI / ML configuration information indicating a second AI / ML configuration to a user equipment operating a first AI / ML configuration in a coverage area of ​​the first base station; wherein the second AI / ML configuration is used by the user equipment in the coverage area of ​​the second base station. Operating the first AI / ML configuration includes operating a first AI / ML model or a first AI / ML model with a first configuration in the coverage area of ​​the first base station. The first AI / ML model is associated with a first AI / ML model identifier, and the first AI / ML model configuration is associated with a first AI / ML model configuration identifier. Operating the second AI / ML configuration includes operating a second AI / ML model or a first AI / ML model with a second AI / ML model configuration in the coverage area of ​​the second base station. The second AI / ML model is associated with a second AI / ML model identifier, and the second AI / ML model configuration is associated with a second AI / ML model configuration identifier. Indicating the second AI / ML configuration includes indicating the second AI / ML model identifier and / or the second AI / ML model configuration identifier. The AI / ML configuration information may be transmitted in a handover command message. The AI / ML configuration information is transmitted after a switching process is initiated in the first base station for switching the user equipment to the coverage area of ​​the second base station, or the AI / ML configuration information is transmitted before a switching process is initiated in the first base station for switching the user equipment to the coverage area of ​​the second base station.

[0020] In an embodiment, an apparatus comprises: a memory; an AI / ML engine configured to operate a first AI / ML configuration in a coverage area of ​​a first base station; and a transceiver configured to receive AI / ML configuration information including information of a second AI / ML configuration; wherein the AI / ML engine operates the second AI / ML configuration indicated by the AI / ML configuration information in the coverage area of ​​the second base station. The AI / ML engine is configured to operate the first AI / ML configuration including executing a first AI / ML model or having a first AI / ML model configuration in the coverage area of ​​the first base station. The first AI / ML model is associated with a first AI / ML model identifier, and the first AI / ML model configuration is associated with a first AI / ML model configuration identifier. The AI / ML engine is configured to operate the second AI / ML configuration including executing a second AI / ML model or a first AI / ML model having a second configuration in the coverage area of ​​the second base station. The second AI / ML model is associated with a second AI / ML model identifier, and the second AI / ML model configuration is associated with a second AI / ML model configuration identifier. The information of the second AI / ML configuration includes a second AI / ML model identifier and / or a second AI / ML model configuration identifier. The AI / ML configuration information may be received in a handover command message, or the AI / ML configuration information may be received in a RACH response message. The AI / ML configuration information may be received in a message transmitted by the first base station, or the AI / ML configuration information may be received in a message transmitted by the second base station. The AI / ML configuration information is received after initiating a handover procedure in the first base station for handing over the user equipment to the coverage area of ​​the second base station, or the AI / ML configuration information is received before initiating a handover procedure in the first base station for handing over the user equipment to the coverage area of ​​the second base station.

[0021] In an embodiment, a first base station comprises: a memory; and a transceiver configured to transmit AI / ML configuration information including information of a second AI / ML configuration to a user equipment; wherein the user equipment is configured to operate the first AI / ML configuration in the coverage area of ​​the first base station; wherein the user equipment is configured to operate the second AI / ML configuration in the coverage area of ​​the second base station based on the transmitted information of the second AI / ML configuration. The user equipment is configured to operate the first AI / ML configuration including operating the first AI / ML model or the first AI / ML model with the first configuration in the coverage area of ​​the first base station. The first AI / ML model is associated with the first AI / ML model identifier, and the first AI / ML model configuration is associated with the first AI / ML model configuration identifier. The user equipment is configured to operate the second AI / ML configuration including operating the second AI / ML model or the first AI / ML model with the second AI / ML model configuration in the coverage area of ​​the second base station. The second AI / ML model is associated with the second AI / ML model identifier, and the second AI / ML model configuration is associated with the second AI / ML model configuration identifier. The information of the second AI / ML configuration includes the second AI / ML model identifier and / or the second AI / ML model configuration identifier. The AI / ML configuration information may be transmitted in a handover command message. The AI / ML configuration information is transmitted after the handover process for handing over the user equipment to the coverage area of ​​the second base station is initiated in the first base station, or the AI / ML configuration information is transmitted before the handover process for handing over the user equipment to the coverage area of ​​the second base station is initiated in the first base station.

[0022] In an embodiment, a first non-transitory computer-readable medium includes instructions operable to cause a processor or processors to receive AI / ML configuration information from a base station, the AI / ML configuration information including information of a second AI / ML configuration; a second non-transitory computer-readable medium includes instructions operable to cause an AI / ML engine to operate a first AI / ML configuration in a coverage area of ​​a first base station, and to operate a second AI / ML configuration indicated by the AI / ML configuration information in a coverage area of ​​a second base station. The instructions operable to cause the AI / ML engine to operate the first AI / ML configuration include instructions for executing a first AI / ML model or a first AI / ML model having a first AI / ML model configuration in a coverage area of ​​the first base station. The first AI / ML model is associated with a first AI / ML model identifier, and the first AI / ML model configuration is associated with the first AI / ML model configuration identifier. The instructions operable to cause the AI / ML engine to operate a second AI / ML configuration include instructions for executing a second AI / ML model or a first AI / ML model having a second configuration in a coverage area of ​​the second base station. The second AI / ML model is associated with a second AI / ML model identifier, and the second AI / ML model configuration is associated with a second AI / ML model configuration identifier. The information of the second AI / ML configuration includes a second AI / ML model identifier and / or a second AI / ML model configuration identifier. The AI / ML configuration information may be received in a switching command message, or the AI / ML configuration information may be received in a RACH response message. The AI / ML configuration information may be received in a message transmitted by the first base station, or the AI / ML configuration information may be received in a message transmitted by the second base station. The AI / ML configuration information is received after a switching process for switching a user equipment to a coverage area of ​​a second base station is initiated in the first base station, or the AI / ML configuration information is received before a switching process for switching a user equipment to a coverage area of ​​a second base station is initiated in the first base station.

[0023] In an embodiment, a first non-transitory computer-readable medium includes instructions operable to cause a processor or processors to transmit AI / ML configuration information including information of a second AI / ML configuration to a user device; wherein the user device is configured to operate the first AI / ML configuration in a coverage area of ​​a first base station; wherein the user device is configured to operate the second AI / ML configuration in a coverage area of ​​a second base station based on the transmitted information of the second AI / ML configuration. The user device is configured to operate the first AI / ML configuration including operating a first AI / ML model or a first AI / ML model with a first configuration in the coverage area of ​​the first base station. The first AI / ML model is associated with a first AI / ML model identifier, and the first AI / ML model configuration is associated with a first AI / ML model configuration identifier. The user device is configured to operate the second AI / ML configuration including operating a second AI / ML model or a first AI / ML model with a second AI / ML model configuration in the coverage area of ​​a second base station. The second AI / ML model is associated with a second AI / ML model identifier, and the second AI / ML model configuration is associated with a second AI / ML model configuration identifier. The information of the second AI / ML configuration includes a second AI / ML model identifier and / or a second AI / ML model configuration identifier. The AI / ML configuration information may be transmitted in a handover command message. The AI / ML configuration information is transmitted after the handover process for handing over the user equipment to the coverage area of ​​the second base station is initiated in the first base station, or the AI / ML configuration information is transmitted before the handover process for handing over the user equipment to the coverage area of ​​the second base station is initiated in the first base station.

[0024] According to an embodiment, a method performed by a user equipment includes: receiving AI / ML model information from a first base station; wherein the AI / ML model information includes first information of a first plurality of AI / ML models to be used for communicating with the first base station and second information of a second plurality of AI / ML models to be used for communicating with a second base station; receiving AI / ML model activation information from the first base station; wherein the AI / ML model activation information includes third information of one or more AI / ML models from the first plurality of AI / ML models to be activated for communicating with the first base station, and fourth information of one or more AI / ML models from the second plurality of AI / ML models to be activated for communicating with the second base station; based on the received third information, activating one or more AI / ML models from the first plurality of AI / ML models; based on the received fourth information, activating one or more AI / ML models from the second plurality of AI / ML models; using the activated one or more AI / ML models from the first plurality of AI / ML models to transmit a signal or channel to / receive a signal or channel from the first base station; and using the activated one or more AI / ML models from the second plurality of AI / ML models to transmit a signal or channel to / receive a signal or channel from the second base station. The AI / ML model information is received in an RRC message. The AI / ML model activation information is received in a physical layer message (e.g., DCI) or a MAC layer message. The first information and the second information may include one or more of the following: an identifier or index of an AI / ML model, an identifier or index of an AI / ML model series, or an identifier or index of an AI / ML model configuration. The third information and the fourth information may include a bit pattern in which each bit indicates the activation / deactivation state of the AI / ML model. In the bit pattern, the bits are arranged according to a predefined order of the identifier or index of the AI / ML model. For example, the LSB corresponds to the AI / ML model with the smallest index, and the MSB corresponds to the AI / ML model with the largest index. Similarly, the third information and the fourth information may include a bit pattern for indicating the AI / ML model configuration.

[0025] According to an embodiment, a method performed by a user equipment includes: receiving AI / ML model information from a first base station; wherein the AI / ML model information includes first information of a first plurality of AI / ML models to be used for communicating with the first base station and second information of a second plurality of AI / ML models to be used for communicating with a second base station; receiving first AI / ML model activation information from the first base station; wherein the first AI / ML model activation information includes information of one or more AI / ML models from the first plurality of AI / ML models to be activated for communicating with the first base station; receiving second AI / ML model activation information from the second base station; wherein the second AI / ML model activation information includes information of one or more AI / ML models from the second plurality of AI / ML models to be activated for communicating with the first base station Information of one or more AI / ML models to be activated for communication with a second base station; based on the received first AI / ML model activation information, activate one or more AI / ML models from a first plurality of AI / ML models; based on the received second AI / ML model activation information, activate one or more AI / ML models from a second plurality of AI / ML models; use the activated one or more AI / ML models from the first plurality of AI / ML models to transmit a signal or channel to / receive a signal or channel from the first base station; and use the activated one or more AI / ML models from the second plurality of AI / ML models to transmit a signal or channel to / receive a signal or channel from the second base station. The AI / ML model information is received in an RRC message. The first and second AI / ML model activation information are received in corresponding physical layer messages (e.g., DCI) or MAC layer messages. The first information and the second information may include one or more of the following: an identifier or index of an AI / ML model, an identifier or index of an AI / ML model family, or an identifier or index of an AI / ML model configuration. The first and second AI / ML model activation information may include a bit pattern in which each bit indicates the activation / deactivation state of the AI / ML model. In the bit pattern, the bits are arranged according to a predefined order of the identifier or index of the AI / ML model. For example, the LSB corresponds to the AI / ML model with the smallest index, and the MSB corresponds to the AI / ML model with the largest index. Similarly, the first and second AI / ML model activation information may include a bit pattern for indicating the AI / ML model configuration.

[0026] According to an embodiment, a method performed by a user equipment includes: receiving first AI / ML model information from a first base station; wherein the first AI / ML model information includes information of a first plurality of AI / ML models to be used for communicating with the first base station; receiving second AI / ML model information from a second base station; wherein the second AI / ML model information includes information of a second plurality of AI / ML models to be used for communicating with the second base station; receiving first AI / ML model activation information from the first base station; wherein the first AI / ML model activation information includes information of one or more AI / ML models from the first plurality of AI / ML models to be activated for communicating with the first base station; receiving second AI / ML model activation information from the second base station; wherein the second AI / ML model activation information The information includes information of one or more AI / ML models from a second plurality of AI / ML models to be activated for communication with a second base station; based on the received first AI / ML model activation information, one or more AI / ML models from the first plurality of AI / ML models are activated; based on the received second AI / ML model activation information, one or more AI / ML models from the second plurality of AI / ML models are activated; using the activated one or more AI / ML models from the first plurality of AI / ML models to transmit a signal or channel to / receive a signal or channel from the first base station; and using the activated one or more AI / ML models from the second plurality of AI / ML models to transmit a signal or channel to / receive a signal or channel from the second base station. The first AI / ML model information and the second AI / ML model information are received in an RRC message. The first and second AI / ML model activation information are received in corresponding physical layer messages (e.g., DCI) or MAC layer messages. The first AI / ML model information and the second AI / ML model information may include one or more of the following: an identifier or index of an AI / ML model, an identifier or index of an AI / ML model family, or an identifier or index of an AI / ML model configuration. The first and second AI / ML model activation information may include a bit pattern in which each bit indicates the activation / deactivation state of the AI / ML model. In the bit pattern, the bits are arranged according to a predefined order of the identifier or index of the AI / ML model. For example, the LSB corresponds to the AI / ML model with the smallest index, and the MSB corresponds to the AI / ML model with the largest index. Similarly, the first and second AI / ML model activation information may include a bit pattern for indicating the AI / ML model configuration.

[0027] According to an embodiment, a method performed by a first base station includes: transmitting AI / ML model information to a user equipment; wherein the AI / ML model information includes first information of a first plurality of AI / ML models to be used for communication with the first base station and the user equipment, and second information of a second plurality of AI / ML models to be used for communication with the second base station and the user equipment; transmitting AI / ML model activation information to the user equipment; wherein the AI / ML model activation information includes third information of one or more AI / ML models to be activated for communication with the user equipment and the first base station from the first plurality of AI / ML models, and fourth information of one or more AI / ML models to be activated for communication with the user equipment and the second base station from the second plurality of AI / ML models; and transmitting to the user equipment a signal or channel to be decoded using one or more AI / ML models activated from the first plurality of AI / ML models, or receiving from the user equipment a signal or channel generated using one or more AI / ML models activated from the first plurality of AI / ML models. The AI / ML model information is transmitted in an RRC message. The AI / ML model activation information is transmitted in a physical layer message (e.g., DCI) or a MAC layer message. The first information and the second information may include one or more of the following: an identifier or index of an AI / ML model, an identifier or index of an AI / ML model series, or an identifier or index of an AI / ML model configuration. The third information and the fourth information may include a bit pattern in which each bit indicates the activation / deactivation state of the AI / ML model. In the bit pattern, the bits are arranged according to a predefined order of the identifier or index of the AI / ML model. For example, the LSB corresponds to the AI / ML model with the smallest index, and the MSB corresponds to the AI / ML model with the largest index. Similarly, the third information and the fourth information may include a bit pattern for indicating the AI / ML model configuration.

[0028] According to an embodiment, a method performed by a first base station includes: transmitting AI / ML model information to a user equipment; wherein the AI / ML model information includes first information of a first plurality of AI / ML models to be used for communication with the user equipment and the first base station, and second information of a second plurality of AI / ML models to be used for communication with the user equipment and the second base station; transmitting first AI / ML model activation information from to the user equipment; wherein the first AI / ML model activation information includes information of one or more AI / ML models to be activated for communication with the first base station from the first plurality of AI / ML models; and transmitting to the user equipment a signal or channel to be decoded using one or more activated AI / ML models from the first plurality of AI / ML models, or receiving from the user equipment a signal or channel generated using one or more activated AI / ML models from the first plurality of AI / ML models. The AI / ML model information is transmitted in an RRC message. The first AI / ML model activation information is transmitted in a physical layer message (e.g., DCI) or a MAC layer message. The first information and the second information may include one or more of the following: an identifier or index of an AI / ML model, an identifier or index of an AI / ML model series, or an identifier or index of an AI / ML model configuration. The first AI / ML model activation information may include a bit pattern in which each bit indicates the activation / deactivation state of the AI / ML model. In the bit pattern, the bits are arranged according to a predefined order of the identifier or index of the AI / ML model. For example, the LSB corresponds to the AI / ML model with the smallest index, and the MSB corresponds to the AI / ML model with the largest index. Similarly, the first AI / ML model activation information may include a bit pattern for indicating the AI / ML model configuration.

[0029] According to an embodiment, a method performed by a first base station includes: transmitting first AI / ML model information to a user equipment; wherein the first AI / ML model information includes information of a first plurality of AI / ML models to be used for communicating with the user equipment and the first base station; transmitting first AI / ML model activation information to the user equipment; wherein the first AI / ML model activation information includes information of one or more AI / ML models to be activated for communicating with the user equipment from the first plurality of AI / ML models; and transmitting to the user equipment a signal or channel to be decoded using one or more activated AI / ML models from the first plurality of AI / ML models, or receiving from the user equipment a signal or channel generated using one or more activated AI / ML models from the first plurality of AI / ML models. The first AI / ML model information is transmitted in an RRC message. The first AI / ML model activation information is transmitted in a physical layer message (e.g., DCI) or a MAC layer message. The first AI / ML model information may include one or more of the following: an identifier or index of an AI / ML model, an identifier or index of an AI / ML model series, or an identifier or index of an AI / ML model configuration. The first AI / ML model activation information may include a bit pattern in which each bit indicates an activation / deactivation state of the AI / ML model. In the bit pattern, the bits are arranged according to a predefined order of an identifier or index of the AI / ML model. For example, the LSB corresponds to the AI / ML model with the smallest index, and the MSB corresponds to the AI / ML model with the largest index. Similarly, the first AI / ML model activation information may include a bit pattern for indicating the AI / ML model configuration.

[0030] According to an embodiment, a user device includes: an AI / ML engine configured to operate multiple AI / ML models; wherein the AI / ML model is associated with an AI / ML model series; wherein the AI / ML model is associated with one or more AI / ML model configurations; a transceiver configured to transmit information containing identifiers of one or more of the supported AI / ML model series, AI / ML models, and AI / ML model configurations. In an embodiment, the information containing the identifier is a bit pattern in which each bit indicates an AI / ML model series, an AI / ML model, and an AI / ML model configuration in a predefined order. For example, the AI / ML models can be indicated in an ascending order in which the LSB bit indicates the AI / ML model with the smallest index (or smallest identifier). Similarly, a separate bit pattern can be used to indicate an AI / ML model series or an AI / ML model configuration.

[0031] According to an embodiment, a user device comprises: an AI / ML engine configured to operate multiple AI / ML models; wherein the AI / ML model is associated with an AI / ML model series; wherein the AI / ML model is associated with one or more AI / ML model configurations; a transceiver configured to receive information from a base station containing identifiers of one or more of the supported AI / ML model series, AI / ML models, and AI / ML model configurations to be used by the user device. In an embodiment, the information containing the identifier is a bit pattern in which each bit indicates an AI / ML model series, an AI / ML model, and an AI / ML model configuration in a predefined order. For example, the AI / ML models can be indicated in an ascending order in which the LSB bit indicates the AI / ML model with the smallest index (or smallest identifier). Similarly, a separate bit pattern can be used to indicate an AI / ML model series or an AI / ML model configuration.

[0032] According to an embodiment, a user device includes: an AI / ML engine configured to perform a processing task by using one or more AI / ML models; a non-AI / ML signal processing module; a transceiver configured to transmit information indicating the switching of the processing task from the AI / ML engine to the non-AI / ML signal processing module. In an embodiment, the information indicating the switching of the processing task from the AI / ML engine to the non-AI / ML signal processing module includes one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating the switching. In an embodiment, if the processing task is switched from the AI / ML engine to the non-AI / ML signal processing module, the flag bit indicating the switching is set to "1".

[0033] According to an embodiment, a user equipment comprises: an AI / ML engine configured to perform a processing task by using one or more AI / ML models; a non-AI / ML signal processing module; a transceiver configured to receive information indicating the switching of the processing task from the AI / ML engine to the non-AI / ML signal processing module from a base station. In an embodiment, the information indicating the switching of the processing task from the AI / ML engine to the non-AI / ML signal processing module includes one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating the switching. In an embodiment, if the processing task is switched from the AI / ML engine to the non-AI / ML signal processing module, the flag bit indicating the switching is set to "1".

[0034] According to an embodiment, a user device comprises: an AI / ML engine; a non-AI / ML signal processing module, which is configured to perform a processing task; a transceiver, which is configured to transmit information indicating the switching of the processing task from the non-AI / ML signal processing module to the AI / ML engine. In an embodiment, the information indicating the switching of the processing task from the non-AI / ML signal processing module to the AI / ML engine includes one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating the switching. In an embodiment, if the processing task is switched from the non-AI / ML signal processing module to the AI / ML engine, the flag bit indicating the switching is set to "0".

[0035] According to an embodiment, a user equipment comprises: an AI / ML engine; a non-AI / ML signal processing module, which is configured to perform a processing task; a transceiver, which is configured to receive information from a base station indicating the switching of the processing task from the non-AI / ML signal processing module to the AI / ML engine. In an embodiment, the information indicating the switching of the processing task from the non-AI / ML signal processing module to the AI / ML engine includes one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating the switching. In an embodiment, if the processing task is switched from the non-AI / ML signal processing module to the AI / ML engine, the flag bit indicating the switching is set to "0".

[0036] According to an embodiment, a user device includes: an AI / ML engine configured to perform a first processing task by using a first AI / ML model and to perform a second processing task by using a second AI / ML model; a non-AI / ML signal processing module; a transceiver configured to transmit information indicating the switching of the first processing task from the AI / ML engine to the non-AI / ML signal processing module; wherein the second processing task is not switched to the non-AI / ML signal processing module. In an embodiment, the information indicating the switching of the processing task from the AI / ML engine to the non-AI / ML signal processing module includes one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating switching. In an embodiment, if the processing task is switched from the AI / ML engine to the non-AI / ML signal processing module, the flag bit indicating the switching is set to "1".

[0037] According to an embodiment, a user equipment comprises: an AI / ML engine configured to perform a first processing task by using a first AI / ML model and to perform a second processing task by using a second AI / ML model; a non-AI / ML signal processing module; a transceiver configured to receive information indicating the switching of the first processing task from the AI / ML engine to the non-AI / ML signal processing module from a base station; wherein the second processing task is not switched to the non-AI / ML signal processing module. In an embodiment, the information indicating the switching of the processing task from the AI / ML engine to the non-AI / ML signal processing module includes one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating switching. In an embodiment, if the processing task is switched from the AI / ML engine to the non-AI / ML signal processing module, the flag bit indicating the switching is set to "1".

[0038] According to an embodiment, a user equipment comprises: an AI / ML engine configured to perform a first processing task by using a first AI / ML model and to perform a second processing task by using a second AI / ML model; a non-AI / ML signal processing module configured to perform a third processing task; a transceiver configured to transmit information indicating the switching of the first processing task from the AI / ML engine to the non-AI / ML signal processing module and information indicating the switching of the third processing task from the non-AI / ML signal processing module to the AI / ML engine; wherein the second processing task is not switched to the non-AI / ML signal processing module. In an embodiment, the information indicating the switching of the first processing task from the AI / ML engine to the non-AI / ML signal processing module and the information indicating the switching of the third processing task from the non-AI / ML signal processing module to the AI / ML engine include one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating switching. In an embodiment, if the first processing task is switched from the AI / ML engine to the non-AI / ML signal processing module, the second processing task is not switched, and the third processing task is switched from the non-AI / ML signal processing module to the AI / ML, the flag bit indicating the task switch is set to "101". In an embodiment, the flag bit indicates the task index defined in a predefined order. For example, the predefined order can be set to an ascending order in which the LSB bit is set for the task with the smallest index. In an embodiment, the flag bit indicating the task switch is set separately for the task associated with the AI / ML engine and the task associated with the non-AI / ML signal processing module. In an embodiment, the flag bit indicating the task switch is set so that the MSB indicates the AI / ML engine or the non-AI / ML signal processing module, and the remaining bits indicate the task switch. For example, "1100" indicates that the AI / ML engine (the MSB set to "1" indicates the AI / ML engine) switches the third task to the AI / ML engine. Similarly, "0001" indicates a non-AI / ML signal processing module (the MSB set to '0' indicates a non-AI / ML signal processing module) to switch the first task to the non-AI / ML signal processing module.

[0039] According to an embodiment, a user equipment comprises: an AI / ML engine configured to perform a first processing task by using a first AI / ML model and to perform a second processing task by using a second AI / ML model; a non-AI / ML signal processing module configured to perform a third processing task; a transceiver configured to receive information indicating the switching of the first processing task from the AI / ML engine to the non-AI / ML signal processing module and information indicating the switching of the third processing task from the non-AI / ML signal processing module to the AI / ML engine from a base station; wherein the second processing task is not switched to the non-AI / ML signal processing module. In an embodiment, the information indicating the switching of the first processing task from the AI / ML engine to the non-AI / ML signal processing module and the information indicating the switching of the third processing task from the non-AI / ML signal processing module to the AI / ML engine include one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating switching. In an embodiment, if the first processing task is switched from the AI / ML engine to the non-AI / ML signal processing module, the second processing task is not switched, and the third processing task is switched from the non-AI / ML signal processing module to the AI / ML, the flag bit indicating the task switch is set to "101". In an embodiment, the flag bit indicates the task index defined in a predefined order. For example, the predefined order can be set to an increasing order in which the LSB bit is set for the task with the smallest index. In an embodiment, the flag bit indicating the task switch is set separately for the task associated with the AI / ML engine and the task associated with the non-AI / ML signal processing module. In an embodiment, the flag bit indicating the task switch is set so that the MSB indicates the AI / ML engine or the non-AI / ML signal processing module, and the remaining bits indicate the task switch. For example, "1100" indicates that the AI / ML engine (the MSB set to "1" indicates the AI / ML engine) switches the third task to the AI / ML engine. Similarly, "0001" indicates a non-AI / ML signal processing module (the MSB set to '0' indicates a non-AI / ML signal processing module) to switch the first task to the non-AI / ML signal processing module.

[0040] According to an embodiment, a base station includes: an AI / ML engine configured to operate multiple AI / ML models; wherein the AI / ML model is associated with an AI / ML model series; wherein the AI / ML model is associated with one or more AI / ML model configurations; a transceiver configured to receive information containing an identifier of one or more of an AI / ML model series, an AI / ML model, and an AI / ML model configuration supported by a user device; and a comparison module for comparing the AI / ML model series, AI / ML model, and AI / ML model configuration supported by the user device and the AI / ML model series, AI / ML model, and AI / ML model configuration supported by the base station. In an embodiment, the information containing the identifier is a bit pattern in which each bit indicates an AI / ML model series, an AI / ML model, and an AI / ML model configuration in a predefined order. For example, the AI / ML model can be indicated in an ascending order, in which the LSB bit indicates the AI / ML model with the smallest index (or smallest identifier). Similarly, a single bit pattern may be used to indicate an AI / ML model family or AI / ML model configuration.

[0041] According to an embodiment, a base station includes: an AI / ML engine configured to operate multiple AI / ML models; wherein the AI / ML model is associated with an AI / ML model series; wherein the AI / ML model is associated with one or more AI / ML model configurations; a transceiver configured to transmit information containing identifiers of one or more of the AI / ML model series, AI / ML models, and AI / ML model configurations supported by the base station. In an embodiment, the information containing the identifier is a bit pattern in which each bit indicates an AI / ML model series, an AI / ML model, and an AI / ML model configuration in a predefined order. For example, the AI / ML models can be indicated in an ascending order in which the LSB bit indicates the AI / ML model with the smallest index (or smallest identifier). Similarly, a single bit pattern can be used to indicate an AI / ML model series or an AI / ML model configuration.

[0042] According to an embodiment, a base station includes: a transceiver configured to receive information indicating the switching of a processing task from an AI / ML engine to a non-AI / ML signal processing module from a user device; wherein the user device includes an AI / ML engine and a non-AI / ML signal processing module, and the AI / ML engine is configured to perform the processing task by using one or more AI / ML models. In an embodiment, the information indicating the switching of the processing task from the AI / ML engine to the non-AI / ML signal processing module includes one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating the switching. In an embodiment, if the processing task is switched from the AI / ML engine to the non-AI / ML signal processing module, the flag bit indicating the switching is set to "1".

[0043] According to an embodiment, a base station comprises: a transceiver configured to transmit information indicating the switching of a processing task from an AI / ML engine to a non-AI / ML signal processing module to a user equipment; wherein the user equipment comprises an AI / ML engine and a non-AI / ML signal processing module, and the AI / ML engine is configured to perform the processing task by using one or more AI / ML models. In an embodiment, the information indicating the switching of the processing task from the AI / ML engine to the non-AI / ML signal processing module comprises one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating the switching. In an embodiment, if the processing task is switched from the AI / ML engine to the non-AI / ML signal processing module, the flag bit indicating the switching is set to "1".

[0044] According to an embodiment, a base station comprises: a transceiver configured to transmit information indicating the switching of a processing task from a non-AI / ML signal processing module to an AI / ML engine to a user equipment; wherein the user equipment comprises an AI / ML engine and a non-AI / ML signal processing module configured to perform the processing task. In an embodiment, the information indicating the switching of the processing task from the non-AI / ML signal processing module to the AI / ML engine comprises one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating the switching. In an embodiment, if the processing task is switched from a non-AI / ML signal processing module to an AI / ML engine, the flag bit indicating the switching is set to "0".

[0045] According to an embodiment, a base station includes: a transceiver configured to receive information indicating the switching of a processing task from a non-AI / ML signal processing module to an AI / ML engine from a user equipment; wherein the user equipment includes an AI / ML engine and a non-AI / ML signal processing module configured to perform the processing task. In an embodiment, the information indicating the switching of the processing task from the non-AI / ML signal processing module to the AI / ML engine includes one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating the switching. In an embodiment, if the processing task is switched from a non-AI / ML signal processing module to an AI / ML engine, the flag bit indicating the switching is set to "0".

[0046] According to an embodiment, a base station comprises: a transceiver configured to transmit information indicating the switching of a first processing task from an AI / ML engine to a non-AI / ML signal processing module to a user equipment; wherein the user equipment comprises an AI / ML engine, the AI / ML engine being configured to perform the first processing task by using a first AI / ML model, and to perform the second processing task by using a second AI / ML model and a non-AI / ML signal processing module. In an embodiment, the information indicating the switching of the processing task from the AI / ML engine to the non-AI / ML signal processing module comprises one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating the switching. In an embodiment, if the processing task is switched from the AI / ML engine to the non-AI / ML signal processing module, the flag bit indicating the switching is set to "1".

[0047] According to an embodiment, a base station includes: a transceiver configured to receive information indicating the switching of a first processing task from an AI / ML engine to a non-AI / ML signal processing module from a user device; wherein the user device includes an AI / ML engine configured to perform the first processing task by using a first AI / ML model, and to perform the second processing task by using a second AI / ML model and a non-AI / ML signal processing module. In an embodiment, the information indicating the switching of the processing task from the AI / ML engine non-AI / ML signal processing module includes one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating the switching. In an embodiment, if the processing task is switched from the AI / ML engine to the non-AI / ML signal processing module, the flag bit indicating the switching is set to "1".

[0048] According to an embodiment, a base station includes: a transceiver configured to receive information indicating the switching of a first processing task from an AI / ML engine to a non-AI / ML signal processing module, and information indicating the switching of a third processing task from a non-AI / ML signal processing module to an AI / ML engine from a user equipment; wherein the user equipment includes an AI / ML engine configured to perform a first processing task by using a first AI / ML model and a second processing task by using a second AI / ML model, and a non-AI / ML signal processing module configured to perform the third processing task. In an embodiment, the information indicating the switching of the first processing task from the AI / ML engine to the non-AI / ML signal processing module and the information indicating the switching of the third processing task from the non-AI / ML signal processing module to the AI / ML engine include one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating switching. In an embodiment, if the first processing task is switched from the AI / ML engine to the non-AI / ML signal processing module, the second processing task is not switched, and the third processing task is switched from the non-AI / ML signal processing module to the AI / ML, the flag bit indicating the task switch is set to "101". In an embodiment, the flag bit indicates a task index defined in a predefined order. For example, the predefined order can be set to an ascending order in which the LSB bit is set for the task with the smallest index. In an embodiment, the flag bit indicating the task switch is set separately for the task associated with the AI / ML engine and the task associated with the non-AI / ML signal processing module. In an embodiment, the flag bit indicating the task switch is set so that the MSB indicates the AI / ML engine or the non-AI / ML signal processing module, and the remaining bits indicate the task switch. For example, "1100" indicates that the AI / ML engine (the MSB set to "1" indicates the AI / ML engine) switches the third task to the AI / ML engine. Similarly, "0001" indicates a non-AI / ML signal processing module (the MSB set to '0' indicates a non-AI / ML signal processing module) to switch the first task to the non-AI / ML signal processing module.

[0049] According to an embodiment, a base station includes: a transceiver configured to transmit to a user equipment information indicating the switching of a first processing task from an AI / ML engine to a non-AI / ML signal processing module, and information indicating the switching of a third processing task from a non-AI / ML signal processing module to an AI / ML engine; wherein the user equipment includes an AI / ML engine configured to perform a first processing task by using a first AI / ML model and to perform a second processing task by using a second AI / ML model, and a non-AI / ML signal processing module configured to perform the third processing task. In an embodiment, the information indicating the switching of the first processing task from the AI / ML engine to the non-AI / ML signal processing module and the information indicating the switching of the third processing task from the non-AI / ML signal processing module to the AI / ML engine include one or more of a processing task identifier (or index), an AI / ML model identifier (or index), an AI / ML model configuration identifier (or index), or a flag bit indicating switching. In an embodiment, if the first processing task is switched from the AI / ML engine to the non-AI / ML signal processing module, the second processing task is not switched, and the third processing task is switched from the non-AI / ML signal processing module to the AI / ML, the flag bit indicating the task switch is set to "101". In an embodiment, the flag bit indicates a task index defined in a predefined order. For example, the predefined order can be set to an ascending order in which the LSB bit is set for the task with the smallest index. In an embodiment, the flag bit indicating the task switch is set separately for the task associated with the AI / ML engine and the task associated with the non-AI / ML signal processing module. In an embodiment, the flag bit indicating the task switch is set so that the MSB indicates the AI / ML engine or the non-AI / ML signal processing module, and the remaining bits indicate the task switch. For example, "1100" indicates that the AI / ML engine (the MSB set to "1" indicates the AI / ML engine) switches the third task to the AI / ML engine. Similarly, "0001" indicates a non-AI / ML signal processing module (the MSB set to '0' indicates a non-AI / ML signal processing module) to switch the first task to the non-AI / ML signal processing module. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1a is an architecture of a wireless radio system according to an embodiment.

[0051] Figure 1b is a diagram of a user plane protocol stack in a wireless radio system according to an embodiment.

[0052] Figure 1c is a diagram of a control plane protocol stack in a wireless radio system according to an embodiment.

[0053] Figure 2 is a block diagram of a first embodiment related to a user equipment.

[0054] Figure 3 is a block diagram of a first embodiment related to a base station.

[0055] Figure 4 is a block diagram of an Al engine according to an embodiment.

[0056] Figure 5a is a block diagram of a second embodiment related to a user equipment.

[0057] Figure 5b is a block diagram of a third embodiment related to a user equipment.

[0058] Figure 6a is a block diagram of a second embodiment related to a base station.

[0059] Figure 6b is a block diagram of a third embodiment related to a base station.

[0060] Figure 7 is a flow chart for implementing one or more AI / ML models on the base station side according to an embodiment.

[0061] Figure 8 is a flow chart for implementing one or more AI / ML models on a user device side according to an embodiment.

[0062] Fig. 9 is a flow chart for implementing one or more AI / ML models on both the base station side and the user equipment side according to an embodiment.

[0063] Fig.10 is a flowchart for training one or more AI / ML models on the base station side according to an embodiment.

[0064] Fig.11 is a flowchart for training one or more AI / ML models on a user device side according to an embodiment.

[0065] Fig.12 is a flow chart for one or more AI / ML model performance monitoring and feedback according to an embodiment.

[0066] Fig.13a is a diagram illustrating one or more AI / ML models and / or one or more AI / ML model configuration updates before moving from one cell to another cell according to an embodiment.

[0067] Fig.13b is a diagram illustrating handover prediction using one or more AI / ML models, according to an embodiment.

[0068] Fig.13c is a flowchart of a method performed by a user equipment for updating an AI / ML model configuration based on signaling from a first base station during handover according to an embodiment.

[0069] Fig.13d is a flowchart of a method performed by a user equipment for updating an AI / ML model or a group of AI / ML models based on signaling from a first base station during handover according to an embodiment.

[0070] Fig.13e is a flowchart of a method performed by a first base station for updating an AI / ML model configuration of a user equipment during handover according to an embodiment.

[0071] Fig.13f is a flowchart of a method performed by a first base station for updating an AI / ML model or a set of AI / ML models of a user equipment during handover according to an embodiment.

[0072] Fig.14a is a diagram illustrating one or more AI / ML models and / or one or more AI / ML model configuration updates after moving from one cell to another cell according to an embodiment.

[0073] Fig.14b is a flowchart of a method performed by a user equipment for updating an AI / ML model configuration based on signaling from a second base station during handover according to an embodiment.

[0074] Fig.14c is a flowchart of a method performed by a user equipment for updating an AI / ML model or a set of AI / ML models based on signaling from a second base station during handover according to an embodiment.

[0075] Fig.14d is a flowchart of a method performed by a second base station for updating an AI / ML model configuration of a user equipment during handover according to an embodiment.

[0076] Fig.14e is a flowchart of a method performed by a second base station for updating an AI / ML model or a set of AI / ML models of a user equipment during handover according to an embodiment.

[0077] Fig.15a is a diagram illustrating one or more AI / ML model configuration updates while moving from one location to another within a cell according to an embodiment.

[0078] Fig.15b is a diagram illustrating one or more AI / ML models and / or one or more AI / ML model configuration updates while moving from one location to another within a cell according to an embodiment.

[0079] Fig.16 is a flow chart illustrating a first embodiment related to signaling exchanges between base stations for updating / downloading one or more AI / ML models and / or one or more AI / ML model configurations.

[0080] Fig.16a is a flow chart illustrating a second embodiment related to signaling exchanges between base stations for updating / downloading one or more AI / ML models and / or one or more AI / ML model configurations.

[0081] Fig.17 is a flow chart illustrating a third embodiment related to signaling exchanges between base stations for updating / downloading one or more AI / ML models and / or one or more AI / ML model configurations.

[0082] Fig.18 is a diagram illustrating a first embodiment related to signaling of a user equipment in a dual connectivity configuration for implementing one or more AI / ML models and / or one or more AI / ML model configurations.

[0083] Fig.18a is a diagram illustrating a second embodiment related to signaling of a user equipment in a dual connectivity configuration for implementing one or more AI / ML models and / or one or more AI / ML model configurations.

[0084] Fig.18b is a diagram illustrating a third embodiment related to signaling of a user equipment in a dual connectivity configuration for implementing one or more AI / ML models and / or one or more AI / ML model configurations.

[0085] Fig.18c is a diagram illustrating a fourth embodiment related to signaling of a user equipment in a dual connectivity configuration for implementing one or more AI / ML models and / or one or more AI / ML model configurations.

[0086] Fig.19 is a diagram illustrating a first embodiment related to handover prediction using one or more AI / ML models and associated signaling.

[0087] Fig. 20 is a diagram illustrating a second embodiment related to handover prediction using one or more AI / ML models and associated signaling.

[0088] Fig.21 is a diagram illustrating a third embodiment related to handover prediction using one or more AI / ML models and associated signaling.

[0089] Fig. 22is a diagram illustrating a fourth embodiment related to handover prediction using one or more AI / ML models and associated signaling.

[0090] Fig.23 is a diagram illustrating a fifth embodiment related to handover prediction using one or more AI / ML models and associated signaling.

[0091] Fig.24 is a flowchart of a method performed by a user equipment for receiving downlink channels on multiple carriers using an AI / ML model or a set of AI / ML models according to an embodiment.

[0092] Fig.25 is a flowchart of a method performed by a user equipment for receiving downlink channels on multiple carriers using corresponding AI / ML model configurations according to an embodiment.

[0093] Fig.26 is a flowchart of a method performed by a base station for transmitting downlink channels on multiple carriers using an AI / ML model or a set of AI / ML models according to an embodiment.

[0094] Fig. 27 is a flowchart of a method performed by a base station for transmitting downlink channels on multiple carriers using corresponding AI / ML model configurations according to an embodiment. DETAILED DESCRIPTION

[0095] Figure 1aThe present invention is a system diagram of a wireless communication system that can be deployed to provide various communication services such as voice services, packet data, audio, video, etc. The wireless communication system may include user equipment (UE) (200a, 200b, 200c, 200d, 200e), RAN (100) (radio access network), and a core including a 5G core (110) and / or an LTE core (120). The RAN (100) includes base stations (300a, 300b, 300c, 300d, 300e, 300f) or cells that communicate with the UE (200a, 200b, 200c, 200d, 200e). The LTE core (120) includes core network components such as MME (121), HSS (122), PGW (124), and SGW (123). The 5G core (110) includes various functions such as UPF (115), AMF (111), SMF (112), AUSF (116), NSSF (114), UDM (117), PCF (113), and other functions (118) such as NEF, NRF, AF, etc. The detailed scope and functionality of the LTE (120) and 5G core (110) network components can be identified from the 3GPP standard specifications (including connections to the Internet (130a, 130b), PSTN (140a, 140b) and other networks (150a, 150b)). UE (200a, 200b, 200c, 200d, 200e) may refer to a combination of Figure 2 , Figure 5a or Figure 5b The UE disclosed in the description, and the base station (300a, 300b, 300c, 300d, 300e, 300f) may refer to a combination Figure 3 , Figure 6a or Figure 6b A base station is disclosed.

[0096] Throughout the patent specification, a user equipment may be an inclusive concept indicating a terminal used in wireless communication, including UE (User Equipment) (200) in Long Term Evolution (LTE), 5G NR, etc.

[0097] A base station (BS) (300) or a cell may generally refer to a station that communicates with a user equipment (UE) (200). A base station (300) may also be referred to as a Node-B, an evolved Node-B (eNb) (300c, 300d), a gNodeB (gNb) (300a, 300b), MeNb, SeNb, HeNb, a sector, a site, a transmission-reception point (TRP) (300f), a base transceiver system (BTS), an access point, a relay node, an integrated access and backhaul (IAB) node, a remote radio head (RRH) (300e), a radio unit (RU), etc.

[0098] In the patent specification, a base station (300) or a cell may have an inclusive concept, indicating a part of an area covered by a Node-B, an evolved Node-B (eNb) (500d), a gNodeB (gNb) (500b), a MeNb, a SeNb, a sector, a site, a base transceiver system (BTS), an access point, a relay node, an integrated access and backhaul (IAB) node, a remote radio head (RRH) (300e), a radio unit (300e), etc. and functions performed by them. The base station (300) or the cell may include various coverage areas such as a macro cell, a macro cell, a micro cell, a pico cell, a femto cell, a communication range of a relay node, an RRU, an RU, etc.

[0099] In the patent specification, BS may also refer to a radio unit (RU) and / or a distributed unit (DU) and / or a centralized unit (CU) according to the required functionality. It should be understood by those skilled in the art that processing may be split between RU, DU, and CU according to 3GPP and / or O-RAN specifications.

[0100] exist Figure 1b for the user plane (aka data plane) protocol stack, and Figure 1c In the present invention, for the control plane protocol stack, exemplary communication between a base station (300) and a UE (200) in a 5G system is disclosed. In an LTE system, a similar protocol stack also exists for UE (200) communication. One difference with respect to the 5G user plane protocol stack is the SDAP layer that only exists in 5G. One difference with respect to the 5G control plane protocol stack is that NAS signaling is between the UE (200) and the AMF (111), while in LTE, NAS signaling is between the UE (200) and the MME (121).

[0101] like Figure 2 As shown, the user equipment (UE) (200) may include a processor (201), a transceiver (203), one or more antennas (202), a speaker (204) / microphone (205), a keyboard (not shown), a display / touchpad / user interface (210), a memory (non-removable memory or removable memory) (206), an AI / ML engine (211), an AI / ML model format conversion module (212), a power supply (208) (or a battery including a charging circuit), a sensor such as an accelerometer (207), an electronic compass, a global positioning system (GPS) chipset, NFC and other devices such as a satellite transceiver, a digital camera (for photos or videos), a universal serial bus (USB) port, a vibration device, a hands-free headset, Other peripheral devices (209) include a module, a frequency modulation (FM) radio unit, a multimedia player, a video game console module, an Internet browser, etc. It should be understood that the user equipment (UE) (200) may include any sub-combination of the aforementioned elements.

[0102] The processor (201) may be coupled to all or a subset of the following: transceiver (203), speaker (204) / microphone (205), keyboard, display / touchpad / user interface (210), non-removable or removable storage (206), power source (208), sensors (207), and other peripherals (209).

[0103] like Figure 3 As shown, the base station (300) may include a processor (301), a transceiver (303-1 ..., 303-n), an antenna (302-1 ... 302-n), a memory (non-removable memory or removable memory) (306), an AI / ML engine (311), an AI / ML model format conversion module (312) and a power supply (308) (or a battery including a charging circuit). The base station (300) may be configured to host modules such as a measurement configuration module (309) for channel measurement for mobility and scheduling, a radio admission control module (310) for UE (200) admission control to the network, a connection mobility control (313) module for handover related processing, a backhaul interface processing module (307) for processing messages received / transmitted to the core network, an Xn interface processing module (305) for processing messages received / transmitted to other base stations, and a scheduler (304) for dynamically allocating resources to UEs in both uplink and downlink. It should be understood that the base station (300) may include any sub-combination of the aforementioned elements. The base station (300) may also host Figure 3 The MIMO module, channel coding and / or modulation module and carrier aggregation module are not shown.

[0104] The following additional modules may also be hosted by the base station (300): radio resource management for inter-cell radio resource management, radio bearer control, IP header compression, encryption and integrity protection of data, selection of AMF (111) at the UE (200) attachment when the route to the AMF (111) cannot be determined based on information provided by the UE (200), routing of user plane data to one or more UPFs, routing of control plane information to the AMF (111), connection establishment and release, scheduling and transmission of paging messages (originated from the AMF (111)), scheduling and transmission of system broadcast information (originated from the AMM (111) or operations and maintenance), transport level packet marking in uplink; session management; support for network slicing, QoS flow management and mapping to data radio bearers, support for UEs (200a, 200b, 200c, 200d and 200e) in RRC_INACTIVE state, distribution function for non-access stratum (NAS) messages, radio access network sharing, dual connectivity, to name just a few.

[0105] The processor in the UE or BS (201 or 301) can be a general-purpose processor, a digital signal processor (DSP), multiple microprocessors, a single-core or multi-core processor, one or more microprocessors associated with a DSP core, a controller, a microcontroller, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) circuit, any other type of integrated circuit (IC), etc. The processor (201 or 201) can perform signal encoding / decoding, data processing, power control, input / output processing, or any other functionality that enables the user equipment to operate in a wireless environment. The processor (201 or 201) can be coupled to a transceiver (203 or 303-1...303-n), which can also be coupled to an antenna (202 or 302-1...302-n). Although the processor (201 or 301) and the transceiver (203 or 303-1...303-n) may be separate components, it should be understood that the processor (201 or 301) and the transceiver (203 or 303-1..303-n) may be integrated in an electronic package or chip.

[0106] The one or more antennas (202 or 302-1 ... 302-n) may include multiple antennas or antenna arrays. The one or more antennas (202 or 302-1 ... 302-n) may be capable of transmitting / receiving over the entire radio spectrum including the millimeter wave spectrum.

[0107] The transceiver (203 or 303-1...303-n) may be configured to modulate signals to be transmitted by one or more antennas (202 or 302-1...302-n) and demodulate signals received by one or more antennas (302 or 302-1...302-n). The memory (206 or 306) may include a non-removable memory or a removable memory. The non-removable memory may include a random access memory (RAM), a read-only memory (ROM), a hard disk, an SSD, or any other type of memory storage device. The removable memory may include a subscriber identity module (SIM) card, a memory stick, a secure digital (SD) memory card, etc. The memory (206 or 306) may be used to store instructions used by the processor (201 or 301) to perform various user equipment functions, including but not limited to cellular transmission and reception. The cellular transmission and reception functions may include transmission and reception of physical channels and signals (e.g., PUSCH, PUCCH, PRACH, SRS, DMRS, PDSCH, PBCH, PDCCH, PSS, SSS, DMRS, CSI-RS, and PTRS), or may include transmission and reception of high-level data and control signaling (e.g., RRC, MAC, RLC, PDCP, NAS, and SDAP).

[0108] Figure 2 and Figure 3 The AI / ML model format conversion module (212 or 312) shown can operate to convert between the formats presented in Table 2 below (and / or variants developed as AI / ML technology advances). The AI / ML model format conversion module can optionally exist at the UE (200) and / or the BS (300), as shown below in combination with Fig.10 , 11 and 12. The AI / ML model format conversion module 212 or 312) can help download / upload and / or transmit one or more AI / ML models between the UE and the BS and / or between the BS and the UE.

[0109] The AI / ML engine (artificial intelligence / machine learning engine) (211 or 311) in the UE and / or BS may be implemented as pure software, pure hardware, or a combination of hardware and software. When the AI / ML engine (211 or 311) is implemented as software, the processor / GPU may be used as hardware, when the AI / ML engine (211 or 311) is implemented as hardware, an additional AI / ML processor / GPU or chipset may be used, and when the AI / ML engine (211 or 311) is implemented as a combination of software and hardware, the processor / GPU and / or an additional AI / ML processor / GPU or chipset may be used as hardware.

[0110] The AI / ML engine (400) may implement Figure 4 The modules disclosed in and corresponding to Figure 2In the present specification, the AI / ML engine (400) may be used interchangeably with the AI / ML engine (211 or 311). The components of the AI / ML engine (400) may include the following:

[0111] Data collection module (401): provides input data to the model training engine (402) and the model inference engine (403). Such input data may include the following types of data: Training data (406): Data that can be used as input to the AI / ML model training engine. Inference Data (407): Data that can be used as input to the AI / ML model inference engine.

[0112] AI / ML model-specific data preparation (e.g., data preprocessing and cleaning, formatting, and transformation) may not be performed in the data acquisition module (401).

[0113] Examples of data (405) collected by the data collection module (401) may include measurements from one or more UEs, one or more BSs, and / or one or more different network entities / servers, feedback from an execution engine, and / or output from an AI / ML model. The one or more network entities may include one or more core network entities. The one or more servers may include one or more network operator's servers or one or more application servers, such as a location server or a map server (e.g., Google Maps or Apple Maps). A description of the term "AI / ML model" may be identified from Table 1.

[0114] The model training engine (402) is a function that can perform AI / ML model training, validation and / or testing, and can generate AI / ML model performance metrics as part of the AI / ML model testing program. The model training engine can also be responsible for data preparation (e.g., data preprocessing and cleaning, formatting and / or conversion) based on the training data (406) delivered by the data acquisition module. Model deployment / update (408): can be used to initially deploy a trained, validated, and / or tested AI / ML model to the model inference engine (403) and / or deliver an updated AI / ML model to the model inference engine. Model training may involve one or more model training methods, including, for example, supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, neural networks, federated learning, dictionary learning, and active learning.

[0115] The model inference engine (403) is a function that can run an AI / ML model and generate an AI / ML model inference output (e.g., predicted / estimated data, processed data, or decisions) (410). The model inference engine (403) can also provide model performance feedback (409) to the model training engine (402) where applicable. The model inference engine (403) can also be responsible for data preparation (e.g., data preprocessing and cleaning, formatting, and / or conversion) based on the inference data (407) received from the data acquisition module (401). Output to execution engine (410): The AI / ML model inference output generated by the model inference engine may be provided to the execution engine. Output to model training engine (409): Model performance feedback can be used to monitor the performance of the AI / ML model (where available).

[0116] The performance of a trained AI / ML model can be evaluated by using one or more of the following example metrics: Classification metrics Classification metrics can be used to evaluate the performance of AI / ML models for classification tasks, which may involve predicting discrete class labels. There are many suitable classification metrics that can be used, including, for example, One or more of: Accuracy can be used to measure the percentage of correct predictions across multiple samples (including the total number of samples). Precision can be used to measure the percentage of positive instances among multiple predicted positive instances (including the total number of predicted positive instances) Recall can be used to measure the number of actual positive instances (including the total number of actual positive instances) The percentage of positive cases. The F1 score can be used to combine the contributions of both precision and recall so that performance can be evaluated with one metric. The confusion matrix can be used to measure the true positive (tp), true negative (tn), and false positive (fp) in the prediction and false negatives (fn). The confusion matrix is ​​presented in the form of a matrix where the Y-axis shows the true class and the X-axis shows the predicted class. Regression metrics Regression metrics can be used to evaluate the performance of AI / ML models for regression tasks. Unlike classification tasks, which classify inputs into discrete class labels, regression tasks involve predicting continuous numbers. There are many metrics that can be used. Suitable indicators include, for example, one or more of the following: Mean absolute error (MAE) Mean Square Error (MSE) Root mean square error (RMSE) Normalized Mean Square Error (NMSE) · Decision coefficient (often called R-squared) Adjusted R-squared Metrics for online iterative optimization Unlike classification or regression tasks that use labeled data to train one or more AI / ML models, online iterative optimization tasks using reinforcement learning typically do not have labels for pre-trained AI / ML models. Therefore, a separate set of metrics can be used in evaluating the performance of AI / ML models. For online iterative optimization tasks, reinforcement learning (RL) type techniques can be utilized. In addition to its final performance or average optimized performance across iterations / runs, one of the concerns in utilizing RL is its usability or reliability / stability. To measure reliability, metrics that quantify performance dispersion and risk can be used.

[0117] In addition to the above, other metrics that may be considered when leveraging AI / ML-based models include the complexity of the algorithm, the overhead / cost associated with training and inference, and / or resource requirements.

[0118] The execution engine (404) is a function that can receive output from the model reasoning engine (403) and trigger or execute the corresponding task. For example, the execution engine (404) can trigger feedback / commands (411) directed to other device components (413). Other device components (413) can include components within the device (UE or BS) implementing the AI / ML engine (211 or 311) and / or entities outside the device implementing the AI / ML engine (211 or 211). As shown below in conjunction with Figure 7 , 8 The example arrangement further described in the embodiment shown in and 9 includes the following: AI / ML engine (311) on the BS: The execution engine (404) can provide instructions to the processor (301) to transmit reference signals (such as, for example, CSI-RS, DMRS, The execution engine (404) may also provide instructions to measure reference signals to capture data for the AI / ML model. The execution engine (404) may share trained AI / ML model parameters or AI / ML models with the processor (301) for transmission to the UE. (200). The execution engine (404) may also provide feedback to be shared with the UE (200). The feedback may include data for updating the AI / ML model used at the UE (200). · AI / ML engine (211) on the UE: The execution engine (404) can provide instructions to the processor (201) to prepare measurement reports that can be shared with the BS (300). The execution engine (404) can also provide instructions to measure reference signals (such as, for example, CSI-RS, DMRS, SSB signals, and positioning reference signals) to capture data for the AI / ML model. The execution engine (404) can share trained AI / ML model parameters or AI / ML models with the processor (201) for transmission to the BS (300). The execution engine (400) can also provide feedback data to be shared with the BS (300). The feedback can include data for updating the AI / ML model used at the BS (300).

[0119] Data collection feedback (412): is information that can be used to derive training data (406), inference data (407), and / or monitor the performance of the AI / ML model and / or its impact on the network through the update of KPIs and / or performance counters.

[0120] The AI / ML model used by the AI / ML engine (400) may include a deep learning model based on a neural network. In a deep learning model, multiple network nodes may be arranged in different layers, and data may be sent and / or received according to a convolutional connection relationship. Examples of neural network models may include various deep learning techniques, such as a deep neural network (DNN), a convolutional deep neural network (CNN), a recurrent neural network (RNN), a recursive Boltzmann machine (RBM), a restricted Boltzmann machine (RBM), a deep belief network (DBN), LSTM, and / or a deep Q network. Other variations or combinations may be used as AI / ML models in the AI / ML engine (400).

[0121] according to Figure 5a In the presented embodiment, the UE (200) may include as part of the AI / ML engine (211) (e.g. Figure 2 One or more AI / ML modules (as shown), such as, for example, a CSI module (510), a beamforming module (540), a positioning module (520), a power control module (550), a MIMO module (530), a channel coding and / or modulation module (570) and / or a carrier aggregation module (560). CSI module (510): includes one or more AI / ML models (510-1 ..., 510-N) that use CSI-RS signals and / or their derivatives as input for predicting / estimating future CSI-RS signals and / or channel characteristics. The CSI module (510) may also include one or more AI / ML models among the AI / ML models (510-1 ... 510-N) for compressing CSI measurement reports and / or any specific fields in the CSI measurement reports (such as, for example, PMI) for transmission over a wireless interface. Beamforming module (540): includes one or more AI / ML models (540-1...540-N) that use reference signals (such as CSI-RS and / or SSB signals or their derivatives) as input for predicting / estimating future beam pairs. The beamforming module (540) may also include one or more AI / ML models among the AI / ML models (540-1...540-N) for compressing beam measurement reports and / or any specific fields in the beam measurement reports for transmission over the wireless interface. Positioning module (or UE location prediction module) (520): includes one or more AI / ML models (520-1...520-N) that use reference signals such as positioning reference signals (or BS's SRS signals), and / or their derivatives, and / or UE speed, and / or UE trajectory information as input for predicting / estimating the future UE location in the cellular network. • Power Control Module (550): includes one or more AI / ML models (550-1...550-N) that use reference signals and / or historical power control data and / or their derivatives as input for predicting / estimating future UE transmit power in a cellular network. MIMO module (530): includes one or more AI / ML models (530-1...530-N) that use reference signals and / or historical MIMO channel data and / or corresponding derivative signals as input for predicting / estimating future MIMO channel characteristics (such as, for example, the number of MIMO layers, the number of antennas, the number of codewords). Channel Coding and / or Modulation Module (570): includes one or more AI / ML models (570-1...570-N) that use historical channel coding and / or modulation data and / or corresponding derivative signals as input for predicting / estimating future channel coding and / or modulation to be used. Carrier aggregation module (560): includes one or more AI / ML models (560-1...560-N) that use historical carrier aggregation data and / or their derivative signals as input for predicting / estimating future carrier aggregation characteristics (such as, for example, transmit power of aggregated carriers, channel bandwidth of aggregated carriers, combination of carriers to be aggregated, activation / deactivation of aggregated carriers, addition / release of aggregated carriers, and / or MIMO / channel coding / modulation / beamforming characteristics of aggregated carriers).

[0122] AI / ML modules (such as 510, 520, 530, 540, 550, 560, and 570) may correspond to an AI / ML series, which may include one or more AI / ML models within the series. For example, the CSI module (510), the beamforming module (540), the positioning module (520), and the power control module (550) may include the following models: CSI module (or CSI series) (510) AI / ML Model 1CSI(510-1) o… AI / ML Model NCSI (510-N) Beamforming module (or beamforming series) (540) o AI / ML Model 1 Beamforming (540-1) o… AI / ML Model N Beamforming (540-N) Positioning module (or positioning series) (520) o AI / ML Model 1 Positioning (520-1) o… AI / ML Model N Positioning (520-N) Power control module (or power control series) (550) o AI / ML Model 1 Power Control (550-1) o… AI / ML Model N Power Control (550-N) The number of AI / ML models in a series "N" varies depending on the UE and / or BS configuration. The UE configuration includes the UE processing capabilities, memory capabilities, and the UE requirements for AI / ML models at a given time. The BS configuration includes signaling transmitted to the UE to configure "N" AI / ML models in the series at a time. Separate signaling may be transmitted for each AI / ML model series. The BS may transmit additional signaling to include activation of "M" out of the "N" configured AI / ML models in the series at a given time. Information about AI / ML models. At a given time, out of “N” configured AI / ML models, there may be “M” active AI / ML models, where M<=N. N and M are integers and may include "0". N=0 indicates that no AI / ML model family is configured in the UE. While M=0 and N>0 indicates that no AI / ML model is active at a given time.

[0123] according to Figure 5b In the presented embodiment, the UE (200) may include as part of the AI / ML engine (211) (e.g. Figure 2 One or more AI / ML modules of the wireless communication system (shown) such as, for example, a CSI module (510), a beamforming module (540), a positioning module (520), a power control module (550), a MIMO module (530), a channel coding and / or modulation module (570), and / or a carrier aggregation module (570). The MIMO module (530), the channel coding and / or modulation module (570), and the carrier aggregation module (560) are not shown in the figure, but may be used to utilize AI / ML technology for corresponding use cases according to implementation. The UE (200) may also include one or more modules (580) that do not use any AI / ML model to process enhancements, such as, for example, a non-AI / ML CSI module (510a), a non-AI / ML beamforming module (540a), a non-AI / ML positioning module (520a), a non-AI / ML power control module (550a), a non-AI / ML conventional MIMO module (530a), a non-AI / ML conventional channel coding and / or modulation module (570a), and / or a non-AI / ML conventional carrier aggregation module (560a). The non-AI / ML conventional MIMO module (530a), the non-AI / ML conventional channel coding and / or modulation module (570a), and / or the non-AI / ML conventional carrier aggregation module are not shown in the figure, but may be used according to an embodiment.

[0124] According to an embodiment, the UE (200) may switch between the AI / ML module (510, 520, 530, 540, 550, 560, or 570) and the corresponding non-AI / ML processing module (510a, 520a, 530a, 540a, 550a, 560a, or 570a), including, for example, based on signaling received from the BS (300) or based on AI / ML model performance. For example, the UE (200) may switch from the AI / ML CSI module (510) (e.g., Model 1 CSI (510-1)) to the non-AI / ML CSI model (510a) for transmitting CSI reports to the BS (300). According to an embodiment, the UE (200) may use one or more specific bits to indicate whether it is generated from one or more AI / ML modules (510, 520, 530, 540, 550, 560 or 570) and / or one or more non-AI / ML processing modules (510a, 520a, 530a, 540a, 550a, 560a or 570a) when sending a report / feedback (such as, for example, a CSI report, a beamforming report, a positioning-related parameter report, a power control-related parameter feedback, a MIMO-related parameter report, a channel coding and / or modulation-related parameter report, and / or a carrier aggregation-related parameter report). According to another embodiment, the UE (200) may include a flag bit in the report / feedback to indicate whether the AI / ML engine (211) or the non-AI / ML signal processing module (580) is used when sending the report / feedback.

[0125] According to an embodiment, the UE (200) may switch from one or more AI / ML modules (510, 520, 530, 540, 550, 560, or 570) to corresponding one or more non-AI / ML processing modules (510a, 520a, 530a, 540a, 550a, 560a, or 570a), including, for example, based on signaling received from the BS (300) and / or based on AI / ML model performance measurements. For example, the UE (200) may switch from an AI / ML CSI module (510) (e.g., Model 1 CSI) to a non-AI / ML CSI model (510a) for sending CSI reports to the BS (300) without switching the AI / ML beamforming module (540). According to an embodiment, the UE (200) may operate in a configuration using a subset (510, 520, 530, 540, 550, 560, or 570) of AI / ML modules and a subset (510a, 520a, 530a, 540a, 550a, 560a, or 570a) of non-AI / ML processing modules. According to an embodiment, the UE (200) may use one or more specific bits to indicate whether a report and / or feedback (such as, for example, a CSI report, a beamforming report, a positioning-related parameter report, a power control-related parameter feedback, a MIMO-related parameter report, a channel coding and / or modulation-related parameter report, and / or a carrier aggregation-related parameter report) is generated from one or more AI / ML modules (510, 520, 530, 540, 550, 560, or 570) and / or one or more non-AI / ML processing modules (510a, 520a, 530a, 540a, 550a, 560a, or 570a) when sending the report and / or feedback. According to an embodiment, the UE (200) may use a specific bit pattern to indicate which parameters in the report / feedback are generated from one or more AI / ML modules (510, 520, 530, 540, 550, 560, or 570) and which parameters are generated from one or more non-AI / ML processing modules (510a, 520a, 530a, 540a, 550a, 560a, or 570a) when transmitting a report / feedback (such as, for example, a CSI report, a beamforming report, a positioning-related parameter report, a power control-related parameter feedback, a MIMO-related parameter report, a channel coding and / or modulation-related parameter report, and / or a carrier aggregation-related parameter report), for example, wherein a single bit may indicate an AI / ML model (or AI / ML module) or one or more non-AI / ML processing modules.

[0126] According to an embodiment, the UE (200) may switch from the AI / ML CSI module (510), the beamforming module (540), the positioning module (520) or the power control module (550) to the corresponding non-AI / ML CSI module (510a), the non-AI / ML beamforming module (540a), the non-AI / ML positioning module (520a) or the non-AI / ML power control module (550a) based on physical layer signaling and / or RRC signaling.

[0127] According to an embodiment, the UE (200) may switch from a non-AI / ML CSI module (510a), a non-AI / ML beamforming module (540a), a non-AI / ML positioning module (520a), or a non-AI / ML power control module (550a) to a corresponding AI / ML CSI module (510), a beamforming module (540), a positioning module (520), or a power control module (550) based on physical layer signaling and / or RRC signaling.

[0128] According to an embodiment, the UE (200) may switch from the AI / ML CSI module (510), the AI / ML beamforming module (540), the AI / ML positioning module (520), and / or the AI / ML power control module (550) to the corresponding non-AI / ML CSI module (510a), the non-AI / ML beamforming module (540a), the non-AI / ML positioning module (520a), and / or the non-AI / ML power control module (550a) based on a predefined AI / ML model performance threshold. For example, if the AI / ML model performance drops below the threshold, the UE (200) may switch to the corresponding non-AI / ML processing module (510a, 520a, 530a, 540a, 550a, 560a, or 570a). The UE (200) may use a specific field or one or more bits in the uplink RRC or physical layer signaling to indicate to the BS (300) when it switches from one or more modules to another one or more modules based on one or more threshold comparisons.

[0129] According to an embodiment, the UE (200) switched from the AI / ML module (510, 520, 530, 540, 550, 560 or 570) to the non-AI / ML module (510a, 520a, 530a, 540a, 550a, 560a or 570a) may switch back to the AI / ML module (510, 520, 530, 540, 550, 560 or 570) based on signaling from the BS (300).

[0130] According to an embodiment, the UE (200) may switch from an AI / ML model within an AI / ML module (series) (510, 520, 530, 540, 550, 560, or 570) to another AI / ML model, including based on a predefined AI / ML model performance threshold. The UE (200) may indicate to the BS (300) when it switches from one AI / ML model to another AI / ML model based on a threshold comparison using a specific field or one or more bits in uplink RRC or physical layer signaling. The UE may optionally include an identifier of the switched AI / ML model in the uplink signaling.

[0131] Depending on the embodiment, the AI / ML models within the AI / ML module(s) (510, 520, 530, 540, 550, 560, or 570) may correspond to different configurations. For example, AI / ML Model 1 CSI (510-1) may correspond to a CSI prediction with configuration 1, and AI / ML Model 2 CSI (510-2) may correspond to a CSI prediction with configuration 2. Similarly, AI / ML Model 1 Beamforming (540-1) may correspond to a beamforming prediction with configuration 1, and AI / ML Model 2 Beamforming (540-2) may correspond to a beamforming prediction with configuration 1. Similar configurations may also exist for a positioning module, a power control module, a MIMO module, a channel coding and / or modulation module, or a carrier aggregation module.

[0132] According to an embodiment, the AI / ML model within the AI / ML module (series) (510, 520, 530, 540, 550, 560 or 570) may correspond to different categories of AI / ML models for the same parameter. For example, AI / ML Model 1 CSI (510-1) may correspond to a CSI prediction category, while AI / ML Model 2 CSI (510-2) may correspond to a CSI compression category, with CSI being a parameter. Similarly, AI / ML Model 1 Beamforming (540-1) may correspond to a time domain beamforming prediction category, and AI / ML Model 2 Beamforming (540-2) may correspond to a spatial domain beamforming prediction category, with beamforming being a parameter. Similar configurations may exist for a positioning module, a power control module, a MIMO module, a channel coding and / or modulation module, or a carrier aggregation module.

[0133] according to Figure 6a In the presented embodiment, the BS (300) may include as part of the AI / ML engine (311) (e.g. Figure 3One or more AI / ML modules (610, 620, 630, 640, 650, 660 or 670) for each UE (UE1, ..., UE N) as shown in the figure, such as, for UE 1, a CSI module (610-1), a beamforming module (640-1), a positioning module (620-1), a power control module (650-1), a MIMO module (630-1), a channel coding and / or modulation module (670-1) and / or a carrier aggregation module (660-1). Similarly, for other UEs, such AI / ML modules can be configured. The AI / ML module can correspond to an AI / ML series, which includes one or more AI / ML models within the series. Further, the BS (300) may have one or more of such AI / ML modules associated with one or more UEs (e.g., those UEs supporting the AI / ML modules in the coverage area of ​​the BS), which may include separate configurations for UEs supporting one or more AI / ML models in the coverage area of ​​the BS (300). Figure 6a Details of the functionality of the modules can be found in Figure 5a The descriptions of , which also apply here.

[0134] according to Figure 6b In the presented embodiment, the BS (300) may include as part of the AI / ML engine (311) (e.g. Figure 3 One or more AI / ML modules (610, 620, 630, 640, 650, 660 or 670) for each UE (UE1, ..., UE N) of the embodiment shown in the figure, such as, for UE 1, a CSI module (610-1), a beamforming module (640-1), a positioning module (620-1), a power control module (650-1), a MIMO module (630-1), a channel coding and / or modulation module (670-1) and / or a carrier aggregation module (660-1). The MIMO module (630-1), the channel coding and / or modulation module (670-1) and the carrier aggregation module (660-1) are not shown in the figure, but according to an embodiment, they can be used to utilize AI / ML technology for corresponding use cases. Similarly, for other UEs, such AI / ML modules can be configured. The AI / ML module can correspond to an AI / ML series, which includes one or more AI / ML models within the series. Further, the BS (311) may have one or more of such AI / ML modules associated with one or more UEs (e.g., those UEs supporting the AI / ML modules in the coverage area of ​​the BS), which may include separate configurations for UEs supporting one or more AI / ML models in the coverage area of ​​the BS (311). Figure 6bDetails of the functionality of the modules can be found in Figure 5a or 5b, which descriptions apply equally here.

[0135] In the scenarios of Figures 5 and 6, the AI / ML model configuration may include one or more of the following: an AI / ML model version number, an identifier or indicator (or index) of the AI / ML model, an identifier or indicator (or index) of the configuration of the AI / ML model, details of the performance metric to be used, AI / ML model parameters (such as one or more of the following: coefficients, weights, biases, the number of AI / ML tasks, one or more AI / ML task identifiers (or one or more indexes), classifiers (such as, for example, regression, KNN, vector machines, decision trees, or principal components), and / or cluster centroids in clustering), or hyperparameters (such as the number of layers, the number of hidden layers, the type of neural network (e.g., DNN, CNN, RNN, or LSTM), the number of nodes per layer, the number of activation units per layer, choice of loss function, batch size, pooling size, choice of activation function (e.g., Sigmoid, ReLU, or Tanh), choice of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer)).

[0136] In the embodiments under the scenarios of Figures 5 and 6, different AI / ML model configurations may correspond to different situations, such as, for example, high or low mobility / speed UE operation, low or high power UE operation, good or poor coverage UE operation, and / or high or low interference UE operation.

[0137] According to an embodiment, in the scenarios of FIG. 5 and FIG. 6, any subset of the AI / ML modules may be used based on UE capabilities and operator configuration. According to other embodiments, the modules may be combined or split as needed. For example, the CSI module (510, 610) may be split into a CSI prediction module and a CSI compression module. Similarly, in Figure 6b In the embodiment, the non-AI / ML signal processing module UE 1 (680-1) can be combined with the non-AI / ML signal processing module UE 2 (680-2). The AI / ML module or the non-AI-ML processing module can be implemented using pure software, pure hardware, or a combination of hardware and software.

[0138] Table 1 provides a description of general terms used in this application related to the deployment of AI / ML technologies in a radio access network (RAN) involving a UE (200) and / or a BS (300). Table 1 AI / ML Implementation in RAN

[0139] According to an embodiment, the AI / ML model may be implemented in the following configurations: One-sided AI / ML functionality only at BS / NW, One-sided AI / ML functionality at the UE only, • Dual-side joint AI / ML functionality at both UE and BS / NW (joint operation).

[0140] A one-sided AI / ML implementation refers to an implementation in which the AI / ML model training process does not need to include the details of the air interface. For example, consider an AI / ML model for enhancing beam management operating in a UE. If the output of the AI / ML model is signaled over the air interface as part of a CSI report or beam management report, the BS can interpret the content of the CSI report without using the AI / ML model. The BS does not need the AI / ML model to be jointly trained with the UE's AI / ML model to decode the CSI report.

[0141] according to Figure 7 In the embodiment shown, the AI / ML model (the AI / ML model on the BS side) can be (300) (e.g., the above combined Figure 3 and / or the BS (300) described in FIG. 6). According to this embodiment, the UE (200) may send one or more measurement reports (710) based on the received reference signal (700). The measurement report (700) may include, for example, one or more of the following: CSI report (including CQI (channel quality information), PMI (precoding matrix indicator), CRI (CSI-RS resource indicator), SSBRI (SS / PBCH resource block indicator), LI (layer indicator), Rl (rank indicator) and / or L1-RSRP), beam measurement reports (including one or more of beam pairs, beam IDs, CSI resources, measured RSRPs and / or measured SINRs), UE locations, UE speeds and / or neighbor cell reports (including one or more of neighbor cell IDs, neighbor cell frequencies, neighbor cell RSRPs, neighbor cell RSRQs, and / or neighbor SINRs). The BS (300) may provide the information contained in the one or more measurement reports (710) to appropriate one or more AI / ML models (720), which in turn may generate one or more BS inferences. These one or more BS inferences may be used to predict one or more BS decisions, including, for example, scheduling, modulation and coding rates, number of MIMO layers, number of antennas, beam pairs, UE locations, channel conditions, optimal UE transmission power, and potential switching conditions. Further, the BS (300) may transmit control information (730) indicating the BS decision to the UE. For example, the control information (730) may indicate one or more of the following: scheduling, modulation and coding rates, number of MIMO layers, number of antennas, beam pairs, optimal UE transmit power, switching commands and reference signal resource information, reference signal resource patterns, and measurement report requests.

[0142] according to Figure 7 In the embodiment shown, the AI / ML model (the AI / ML model on the BS side) can be (300) (e.g., the above combined Figure 3 According to this embodiment, the BS (300) may receive data (710) from multiple UEs within its coverage area to feed its AI / ML model. The BS (300) may use the AI / ML model to predict decisions for one or more new UEs that enter the BS coverage area or for UEs that move from one location in the coverage area to a new location. (300) can be fed with real-time information and / or past information of UEs within its coverage to predict one or more BS decisions for one or more new UEs or one or more UEs moving from one location to a new location within its coverage, the one or more BS decisions including, for example, resource scheduling information, MIMO parameters (such as, for example, the number of layers, the number of antenna ports, or the number of codewords), modulation and coding scheme, bandwidth part, UE transmit power, waveform type (CP-OFDM or DFST-s-OFDM), numerical scheme / subcarrier spacing, switching decision, and / or UE transmit / receive beam pair information. Further, the BS (300) can transmit UE control information (730) indicating the BS decision to one or more new UEs or one or more UEs moving from one location to a new location within its coverage.

[0143] according to Figure 8 In the embodiment shown, the AI / ML model (AI / ML model on the UE side) can be (200) (e.g., the above combined Figure 2 and / or implemented on the UE (200) described in FIG. 5. According to this embodiment, the UE (200) may receive a reference signal (800) from the BS (300) and use one or more AI / ML models to estimate, for example, channel characteristics and / or beam characteristics (820). Based on the output of the UE reasoning, the UE (200) may transmit one or more measurement reports (810). The measurement report (810) may include one or more of the following: a CSI report (including CQI (channel quality information), PMI (precoding matrix indicator), CRI (CSI-RS resource indicator), SSBRI (SS / PBCH resource block indicator), LI (layer indicator), Rl (rank indicator) and / or one or more of L1-RSRP), beam measurement reports (including one or more of beam pairs, beam IDs, CSI resources, measured RSRP and / or measured SINRs), UE location, UE speed and / or neighbor cell reports (including one or more of neighbor cell IDs, neighbor cell frequencies, neighbor cell RSRPs, neighbor cell RCRQs, and / or neighbor cell SINRs). One or more of the measurement reports (810) may include an indicator of the use of an AI / ML module or AI / ML model to generate the report, and may optionally include one or more of an AI / ML model identifier (or index), an AI / ML module identifier (or an AI / ML model family identifier (or index)), or an AI / ML model configuration identifier (or index). Based on the received measurement reports (810), the BS (300) may make decisions regarding, for example, scheduling, modulation and coding rates, number of MIMO layers, number of antennas, beam pairs, UE location, channel conditions, optimal UE transmission power, and potential switching conditions. The BS (300) may transmit control information (830) indicating the BS decision to the UE. For example, the control information (830) may indicate one or more of the following: scheduling, modulation and coding rate, number of MIMO layers, number of antennas, beam pairs, optimal UE transmission power, handover command, reference signal resource information, reference signal resource pattern, measurement report request.

[0144] A dual-sided or joint AI / ML implementation is one in which the AI / ML models are trained for joint use in both the UE and the BS, taking into account the air interface. For example, consider a CSI implementation in which an AI / ML-based encoder in the UE compresses downlink CSI-RS-based channel (feature) estimates, while an AI / ML-based decoder in the BS decompresses these estimates. In this case, the CSI reports signaled over the uplink may only be decodable by a properly trained AI / ML model in the BS.

[0145] To limit system complexity in two-sided or joint AI / ML implementations, it may be desirable to: BS-side AI / ML models can achieve good performance by leveraging many different UE-side AI / ML models developed by different vendors, and UE CSI AI / ML model can achieve good performance by leveraging many different BS-side AI / ML models developed by different vendors.

[0146] according to Fig. 9 In the embodiment shown in FIG. 1 , the AI / ML model or aspects thereof may be implemented in the UE (200) and the BS (300) (e.g., in combination with the above Figure 2 and / or the UE (200) described in FIG. 5, and the above combined Figure 3and / or the BS (300) described in FIG. 6). According to this embodiment, the UE (200) may receive a reference signal (900) from the BS (300) and use one or more AI / ML models to estimate channel characteristics and / or beam characteristics (920). Based on the output of the UE model inference engine, the UE (200) may transmit a measurement report (910). The measurement report (910) may include one or more of the following: a CSI report (including one or more of CQI (channel quality information), PMI (precoding matrix indicator), CRI (CSI-RS resource indicator), SSBRI (SS / PBCH resource block indicator), LI (layer indicator), Rl (rank indicator) and / or L1-RSRP), a beam measurement report (including one or more of beam pairs, beam IDs, CSI resources, measured RSRPs and / or measured SINRs), UE position, UE speed and / or neighbor cell reports (including one or more of neighbor cell IDs, neighbor cell frequencies, neighbor cell RSRPs, neighbor cell RRRQs and / or neighbor cell SINRs). One or more measurement reports (910) may include an indicator that the report was generated using an AI / ML module or AI / ML model, and may optionally include one or more of an AI / ML model identifier (or index), an AI / ML module identifier (or an AI / ML model family identifier (or index)), or an AI / ML model configuration identifier (or index). The BS (300) may provide information contained in the measurement report (910) to one or more AI / ML models (940), and based on the output of one or more BS model inference engines, the BS (300) may predict one or more BS decisions, including, for example, scheduling, modulation and coding rates, number of MIMO layers, number of antennas, beam pairs, UE location, channel conditions, optimal UE transmission power, potential switching scenarios, UE AI / ML model retraining, UE AI / ML model switching, UE AI / ML model updating, UE AI / MLC-model activation / deactivation, and / or UE AI / ML-M-model replacement. The BS (300) may transmit control information (930) indicating the BS decision to the UE.The control information (930) may indicate, for example, one or more of the following: scheduling, modulation and coding rate, number of MIMO layers, number of antennas, beam pairs, optimal UE transmit power, switching request, reference signal resource information, reference signal resource pattern, measurement report request, UE AI / ML model retraining request, UE AI / ML model or AI / ML model configuration switching request, AI / ML model or AI / ML engine switching to non-AI / ML signal processing module switching request, UE AI / ML model update request, UE AI / ML model activation / deactivation request, UE AI / ML model performance parameters (such as AI / ML model performance thresholds), UE location request, UE speed request, UE direction (or trajectory) vector and / or UE AI / ML model replacement request.

[0147] according to Fig. 9 In the embodiment shown in FIG. 1 , the AI / ML model or aspects thereof may be implemented in the UE ( 200 ) and the BS (300) (e.g., the above combined Figure 2 and / or the UE (200) described in FIG. 5, and the above combined Figure 3 6). According to this embodiment, the BS (300) can receive data from multiple UEs within its coverage area to feed its AI / ML model (940). The BS (300) can use the AI / ML model to predict decisions for one or more new UEs entering the BS coverage area or UEs moving from one location in the coverage area to a new location. The BS (300) can provide real-time information and / or past information of UEs within its coverage to predict one or more BS decisions for one or more new UEs or UEs moving from one location to a new location within its coverage, the one or more BS decisions including, for example, resource scheduling information, MIMO parameters (such as, for example, the number of layers, the number of antenna ports, or the number of codewords), modulation and coding schemes, bandwidth parts, UE transmit power, waveform type (CP-OFDM or DFST-s-OFDM), numerical scheme / subcarrier spacing, switching decisions and / or UE transmit / receive beam pair information, reference signal resource information, reference signal resource pattern, measurement report request, UE AI / ML model retraining request, UE AI / ML model or AI / ML model configuration switching request, AI / ML model or AI / ML engine switching to non-AI / ML signal processing module switching request, UE AI / ML model update request, UE AI / ML model activation / deactivation request, UE AI / ML model performance parameters (such as performance thresholds of AI / ML models), UE location request, UE speed request, UE direction (or trajectory) vector, and / or UE AI / ML model replacement request.

[0148] According to an embodiment, a dual-sided AI / ML model may be deployed in a RAN (including a BS and a UE) with the following collaborative configuration: Category 1: No collaboration between UE and BS Category 2: Signaling-based collaboration between UE and BS, without AI / ML model transmission. Category 3: Signaling-based collaboration between UE and BS with AI / ML model transmission. Category 4: Joint (UE and BS) AI / ML model training and / or inference The UE and BS described in the context of Category 2, Category 3, and Category 4 may refer to the UE and BS described above in conjunction with Figure 2 and / or the UE (200) described in FIG. 5, and the above combined Figure 3 and / or the BS described in FIG. 6 (300).

[0149] In Category 1, the application of AI / ML can be purely based on implementation. One or more AI / ML models can be trained and / or used at the UE or BS, but there is no communication between the UE and BS for AI / ML. Purpose of information exchange.

[0150] Category 1 deployment can be used in the following scenarios: CSI reporting with time prediction: Based on the UE's CSI report, the BS can predict the future channel. Beam prediction in the time domain: Like CSI prediction, the BS can predict future beam quality based on the UE’s beam reports. AI / ML-based OTDOA estimation at the UE, where the training and inference are transparent to the NW.

[0151] In Category 2, signaling information is exchanged over the air interface to facilitate AI / ML operations, e.g., training and / or inference, to enable the application of AI / ML on the BS and / or UE. However, there is no model transmission between them.

[0152] Signaling may include RRC layer signaling, physical layer signaling, or MAC layer signaling, or a combination thereof, to exchange information such as: UE-to-BS signaling may include one or more of the following, for example: o Measurement report, which includes CQI (channel quality information), PMI (precoding matrix indicator), CRI (CSI-RS resource indicator), SSBRI (SS / PBCH resource block indicator), LI (layer indicator), Rl (rank indicator) and / or L1-RSRP. o AI / ML reports, which include details of the AI / ML model used by the UE, such as weights, number of layers, number of nodes per layer or hidden nodes. o The UE may signal parameters such as, for example, an AI / ML model indicator (or index), an AI / ML model family indicator (or index), an AI / ML model configuration indicator (or index), an AI / ML model family, an AI / ML model or an AI / ML model configuration activation or deactivation request, and / or a request to switch an AI / ML model or an AI / ML model configuration. o The UE may also signal UE velocity and / or UE position and / or one or more direction (or trajectory) vectors. The signaling from the BS to the UE may include one or more of the following, for example: o AI / ML report, which includes details of the AI / ML model to be used by the UE, which may include, for example, weights, number of layers, number of nodes per layer, or hidden nodes. o The BS may signal parameters such as an AI / ML model indicator (or index), an AI / ML model family indicator (or index), an AI / ML model configuration indicator (or index), an AI / ML model family, an AI / ML model or an AI / ML model configuration activation or deactivation request, and / or a request for switching an AI / ML model or an AI / ML model configuration. o The BS may also request UE measurement reports, velocity and / or position and / or one or more direction (or trajectory) vectors.

[0153] The AI / ML model indicator (or index) may indicate the AI / ML model being used or to be used. It may indicate an AI / ML model within an AI / ML model family.

[0154] According to an embodiment, one or more bits of the AI / ML model indicator (or index) may indicate an AI / ML model family, and one or more bits may indicate a specific AI / ML model within the AI / ML model family. For example, in the case of a 4-bit AI / ML model indicator, 2 bits may indicate an AI / ML model family, and 2 bits may indicate a specific AI / ML model in the AI / ML model family. Other numbers of bits and encodings may be used for this purpose. According to another embodiment, one or more bits of the AI / ML model indicator (or index) may indicate an AI / ML model family, other one or more bits may indicate a specific AI / ML model in the AI / ML model family, and other one or more bits may indicate an AI / ML model configuration. For example, in the case of a 6-bit AI / ML model indicator (or index), 2 bits may indicate an AI / ML model family, 2 bits may indicate a specific AI / ML model within an AI / ML model family, and 2 bits may indicate a specific AI / ML model configuration within a specific AI / ML model. Other numbers of bits and encodings may also be used for this purpose.

[0155] The AI / ML model family, AI / ML model, or AI / ML model configuration activation / deactivation status may indicate the status of the AI / ML model family, AI / ML model, or AI / ML model configuration, for example, whether it is activated or deactivated.

[0156] According to an embodiment, the AI / ML model or AI / ML model configuration activation / deactivation state may indicate the activation and / or deactivation state of multiple AI / ML models or AI / ML model configurations at the same time. For example, the UE may indicate the status of all AI / ML models or AI / ML model configurations in a bit pattern, where a single bit may correspond to an AI / ML model or an AI / ML model configuration.

[0157] According to an embodiment, the AI / ML model activation / deactivation state may indicate the activation and / or deactivation state of multiple AI / ML model series at the same time. For example, the UE may use a bit pattern to indicate which AI / ML model series are activated and / or deactivated, where a single bit may correspond to an AI / ML model series.

[0158] The AI / ML model activation / deactivation request may indicate the activation and / or deactivation of a specific AI / ML model or family of AI / ML models.

[0159] According to an embodiment, the AI / ML model or AI / ML model configuration activation / deactivation request may indicate activation and / or deactivation of multiple AI / ML models at the same time. For example, the BS may request the UE to activate and / or deactivate multiple AI / ML models or AI / ML model configurations using a bit pattern, where a single bit may correspond to an AI / ML model or an AI / ML model configuration.

[0160] According to an embodiment, the AI / ML model activation / deactivation request may indicate activation and / or deactivation of multiple AI / ML model series at the same time. For example, the BS may request the UE to activate and / or deactivate multiple AI / ML model series using a bit pattern, where each bit may correspond to one AI / ML model series.

[0161] Other combinations or subsets of information may be used for such signaling.

[0162] Category 2 deployment can be used in the following scenarios: • Reference signal overhead reduction for channel estimation using CSI compression. • Beam prediction in space / time domain: The UE can measure the quality of a smaller number of beam pairs and estimate the quality of more (including all, in embodiments) beam pairs or the best beam pair. By signaling BS antenna information or calibration information to the UE, positioning accuracy can be improved.

[0163] In Category 2 deployments for AI / ML model training, the UE may send UE capabilities and / or the BS may send reference signal patterns.

[0164] In Category 3 deployment: On top of Category 2, the air interface can be further enhanced to allow the transmission of AI / ML models between the UE and BS. In this category, AI / ML models can be trained on one side of the network and deployed to the other side for inference / execution. However, there is no joint AI / ML operation between the two sides.

[0165] The size of AI / ML models can range from a few KB to hundreds of MB. Considering the overhead of AI / ML model transmission, it may be useful to define the format of AI / ML model exchange and the corresponding signaling. AI / ML model transfer:

[0166] According to an embodiment, the BS may transmit RRC signaling and / or physical layer signaling including AI / ML model configuration information to the UE, and the UE may upload and / or download the AI / ML model based on the received AI / ML model configuration information. The AI / ML configuration information may include, for example, one or more of the following: detailed information of the performance metric to be used, the AI / ML model format (or file type), AI / ML model parameters (such as one or more of the following: coefficients, weights, biases, the number of AI / ML tasks, one or more AI / ML task identifiers, classifiers (such as, for example, regression, KNN, vector machine, decision tree, principal component), cluster centroids in clustering), and / or hyperparameters (such as the number of layers, the number of hidden layers, the type of neural network (e.g., DNN, CNN, RNN, or LSTM), the number of nodes per layer, the number of activation units per layer, the choice of loss function, batch size, pooling size, the choice of activation function (e.g., Sigmoid, ReLU, Tanh), the choice of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer)). The AI / ML configuration information may also include details of the AI / ML model download and / or upload location, such as, for example, a server address, a URL for uploading and / or downloading, details of an API for downloading and / or uploading, or details of an application for uploading and / or downloading.

[0167] According to an embodiment, the UE may download the AI / ML model based on the AI / ML configuration information received at the application layer. The download may be done using an API.

[0168] According to an embodiment, the UE may download the AI / ML model based on the received AI / ML configuration information using a predefined URL created by the UE based on the received configuration information and parameters.

[0169] According to an embodiment, the UE may download an AI / ML model from a core network entity or an edge server based on the received AI / ML configuration information.

[0170] According to an embodiment, the UE can download the AI / ML model from the server of the UE manufacturer. The UE can reach the server of the UE manufacturer using pre-stored information.

[0171] According to an embodiment, the UE may upload the AI / ML model based on the AI / ML configuration information received at the application layer. The upload may be done using an API.

[0172] According to an embodiment, the UE may upload the AI / ML model to a core network entity or an edge server based on the received AI / ML configuration information.

[0173] According to an embodiment, the UE can upload the AI / ML model to the network operator's server. The UE can reach the network operator's server using pre-stored information. AI / ML model transmission format:

[0174] A common format for exchanging / transferring AI / ML models may be defined to make proprietary AI / ML models specific to UE manufacturers or network operators compatible with each other. An example of a common format may include Open Neural Network Exchange (ONNX), an open source AI / ML format, more details of which are available at https: / / onnx.ai / . The trained AI / ML model file may contain information about one or more of the parameters of the deep neural network (such as, for example, the number of layers, weights / biases, details of the quantization used, and / or loss function).

[0175] Depending on the machine learning framework used, the AI / ML model can be saved in a file format. Table 2 lists example frameworks and file formats for storing AI / ML models. New AI / ML model file formats may be developed for use in wireless environments, however, those skilled in the art will appreciate that the AI / ML model parameters may remain similar. Table 2 frame File Format File extensions PyTorch ONNX .onnx PyTorch Torch Script .pt Keras H5 .h5 iOS Core ML Apple ML Model .mlmodel TensorFlow Protobuf .pb Scikit-learn Featured Python .pkl Scikit-learn PMML .pmml

[0176] According to an embodiment, the BS may implement a format conversion module (312) for converting an AI / ML model uploaded by a UE and / or for converting an AI / ML model to be downloaded to a UE. Figure 2 and 3 The model format conversion module shown can operate to convert between the formats presented in Table 2 (or variations thereof developed as AI / ML technology advances). For example, the BS can convert the AI / ML model into a .mlmodel file before downloading it to an Apple smartphone.

[0177] Category 3 deployment can be used in the following scenarios: Utilizing AI / ML model from BS for CSI prediction: BS can transmit AI / ML model to UE, and UE can use this AI / ML model to predict future channels. Utilize AI / ML model from BS for beam prediction in space / time domain: BS can send AI / ML model that matches its beam pattern and wireless environment to UE. UE can use this AI / ML model to find the best beam pair. AI / ML models for positioning accuracy can be trained / aggregated at the BS for the environment and distributed to the UE to speed up training at the UE.

[0178] In a category 3 deployment for AI / ML model training, information may be exchanged between the UE and the BS. The information may include hyperparameters for training. A hyperparameter is a parameter whose value may control the learning process and determine the value of the AI / ML model parameter that the learning algorithm ultimately learns. When an AI / ML model is trained, hyperparameters may be used during AI / ML model training, but they are not part of the resulting AI / ML model. For example, the number of hidden layers, the number of activation units in each layer, the dropout rate (dropout probability), the number of iterations (epochs), the number of clusters in a clustering task, the kernel or filter size in a convolutional layer, the pooling size, the batch size, the learning rate in an optimization algorithm (e.g., gradient descent), the optimization algorithm (e.g., stochastic gradient descent, gradient descent, or Adam optimizer), the activation function in a neural network (nn) layer (e.g., Sigmoid, ReLU, Tanh), the choice of the cost or loss function of the AI / ML model, and / or the training test split ratio.

[0179] In Category 4 deployment: Beyond Category 3, joint AI / ML operations between UE and BS can be used, e.g., AI / ML model training and / or inference (e.g., a federated learning algorithm or an autoencoder type AI / ML model).

[0180] In Category 4 deployment, joint AI / ML operation The AI / ML model can be split into multiple parts, where both BS and UE can participate in training the AI / ML model. For example, in the CSI feedback enhancement use case, in order to reduce the CSI feedback overhead, autoencoder-like or transformer-like AI / ML model-based compression and recovery can be applied, where the UE is the encoder and the BS is the decoder, and joint AI / ML model training and joint AI / ML model inference can be expected. This type of AI / ML operation may require close collaboration between the UE and BS, as intermediate data (e.g., compressed CSI / PMI) may need to be exchanged.

[0181] Category 4 deployment can be used in the following scenarios: CSI payload overhead reduction: i.e., the UE can encode the channel information through the AI / ML model, To generate PMI. Then, the BS can use the matched AI / ML model to decode the PMI. Utilizing AI / ML based encoders and decoders for beam information compression in spatial / temporal domains: The UE can use AI / ML models to compress the beam information, which can be decoded by the BS. The UE-side positioning AI / ML model can extract features from the UE’s measurements and report them to the BS to feed the BS-side AI / ML model for determining the UE’s location.

[0182] In Category 4 deployments for AI / ML model training, regular training information can be exchanged between the UE and the BS. The information can include gradients or loss function results. In Category 4 deployments, joint AI / ML operations AI / ML models can be split into multiple parts and tasks can be performed on the UE (e.g., in conjunction with Figure 2 and / or the UE described in FIG. 5) and a base station (eg, as combined with Figure 3 and / or 6). For example, if the current battery power of the UE is low, the overall processing load of the UE is high, a large number of AI / ML tasks are currently being executed or scheduled, or the AI / ML tasks currently being executed or scheduled are complex, the UE may request the base station to split the tasks in a manner that reduces the UE processing load. The UE may send a request to the base station indicating one or more of the current task split ratio, the desired task split ratio, an AI / ML model indicator (or index), an AI / ML configuration indicator (or index), an AI / ML task indicator (or index), and the reason for the change in the current task split ratio (e.g., battery status, memory status, current processing load indicator, or task complexity indicator). Based on the received request, the base station may change the current task split ratio between the UE and the base station to a new task split ratio, and send control information indicating the new task split ratio to the UE. The new task split ratio may be the same as the desired task split ratio, or may be different from it. The control information may also include one or more of the following: an AI / ML model indicator (or index), an AI / ML configuration indicator (or index), the number of tasks, an AI / ML task indicator (or index) for a currently executed or scheduled AI / ML task to be stopped by the UE, an AI / ML task indicator (or index) for a task to be moved to the base station for execution, or a request to send data for a task to be performed by the base station. After receiving the control information from the base station, the UE performs one or more of the following: stopping the AI / ML task indicated by the base station, sharing data for the task to be performed at the base station. AI / ML model deployment scope

[0183] According to an embodiment, the scope of AI / ML model deployment (e.g., as combined with Figure 7 or 9) can be configured as follows: AI / ML models for each UE AI / ML models for each UE group AI / ML models applicable to each BS (or Cell) The complexity of implementing an AI / ML model applicable to each UE may be higher than the complexity of implementing an AI / ML model applicable to each UE group, and the complexity of implementing an AI / ML model applicable to each UE group is higher than the complexity of an AI / ML model applicable to each BS (or cell). However, at the expense of implementation complexity, improved gains can be achieved when implementing an AI / ML model applicable to each UE or each UE group.

[0184] According to an embodiment, the BS (300) (e.g., as combined with Figure 3 When implementing the AI / ML model for each UE (as described in the embodiments of the present invention and / or FIG. 6 ), unicast signaling specific to the AI / ML model may be generated. The signaling may include a UE identifier and an AI / ML model configuration specific to the UE. Further, the details of the AI / ML model configuration may be identified from other parts of this patent specification.

[0185] According to an embodiment, the BS (300) (e.g., as combined with Figure 3 When implementing the AI / ML model of each UE, signaling specific to the AI / ML model may be transmitted to multiple UEs simultaneously (as described in the embodiments of the present invention and / or FIG. 6). The signaling may include multiple UE identifiers and AI / ML model configurations corresponding to different UEs. Further, the details of the AI / ML model configuration may be identified from other parts of this patent specification.

[0186] According to an embodiment, the BS (300) (e.g., as combined with Figure 3 When implementing the AI / ML model for each UE group (as described in FIG. 6 and / or FIG. 7 ), multicast signaling specific to the AI / ML model may be generated. The signaling may include a group identifier common to a group of UEs and an AI / ML model configuration. Further, the details of the AI / ML model configuration may be identified from other parts of this patent specification.

[0187] According to an embodiment, the BS (300) (e.g., as combined with Figure 3 and / or as described in FIG. 6) in implementing each BS When an AI / ML model is configured for a UE (or cell), broadcast signaling specific to the AI / ML model may be generated. The signaling may include a cell identifier and / or a BS identifier that is common to UEs implementing the AI / ML model in the cell and the AI / ML model configuration. Further, the details of the AI / ML model configuration may be identified from other parts of this patent specification. AI / ML Data Collection Framework

[0188] According to an embodiment, the BS (300) (e.g., as combined with Figure 36) can collect and maintain historical data of the UE and create a cell map based on one or more of the following example parameters: ·time ·frequency ·Location Spatial orientation

[0189] The time parameter of the cell map may refer to the time of day. The time parameter may be associated with the number of UEs at each location within the BS coverage (or cell). For example, there may be more UEs at a commercial street market during the daytime than at night. Similarly, a given location may have a greater number of UEs during certain hours of the day. This information may be helpful for the BS to schedule resources for the UEs and estimate channel conditions.

[0190] In an embodiment, the time parameter may also refer to a timestamp of data collection. The timestamp may be used to predict future values ​​of the parameter. For example, data collected between a duration of T1 to T2 may be used to predict characteristics of a wireless channel within a duration of T3 to T4, where T1 < T2 < T3 < T4. The parameter may refer to CSI prediction, beam pair prediction, position prediction, transmit power prediction, MIMO-related parameter prediction, channel coding and / or modulation prediction, and / or carrier aggregation-related parameter prediction.

[0191] The frequency parameter of the cell map may refer to any one of a frequency band, a configured channel bandwidth, a bandwidth part (BWP), a number of subcarriers, or a subcarrier spacing.

[0192] The location parameter of the cell map may refer to a 2D UE location in the cell or a 3D UE location in the cell. The 3D UE location may also include the height of the UE from the ground or sea level. For example, if the UE is present in a high-rise building, the 3D location may be more useful in estimating channel conditions, while if the UE is standing or moving on a road, the 2D location may be more useful in estimating channel conditions.

[0193] The spatial orientation parameter of the cell map may refer to the direction of UE transmission or reception from the BS. The spatial orientation may refer to a 2D orientation or a 3D orientation. For example, if the UE is in dual connectivity mode, or if it is communicating with multiple TRPs, there may be two or more spatial orientations for the UE, each pointing towards the BS / TRP. The spatial orientation may be expressed in angles, such as, for example, angle of arrival (AoA) or angle of departure (AoD).

[0194] The collected data may be associated with a geographical map of the cell. The geographical map may help identify non-line-of-sight (NLOS) conditions for UEs and predict various parameters such as channel conditions, location and / or beam-related parameters of mobile UEs. The geographical map may be a 3D map or a 2D map.

[0195] The data collection framework may be used to collect information such as, for example, RSRP / RSRQ / SINR, UE / BS beam related information, UE trajectory and / or UE speed. Table 3 below provides an exemplary table that may be maintained by the BS for collecting data to feed the AI / ML model. Table 3

[0196] Without departing from the broader spirit of the present invention, Table 3 may be modified by adding and / or deleting parameters to accommodate different scenarios.

[0197] Furthermore, the UE may also maintain a table similar to Table 3 to provide information to the AI / ML model on the UE side to generate inferences.

[0198] Data collection and storage at the BS: According to an embodiment, the BS may collect and store data received from the UE according to a data collection framework. Data acquisition and storage based on reference signals: According to an embodiment, the BS may collect data according to a data collection framework based on reference signals (e.g., SRS and DMRS) received from the UE. The BS may estimate a value based on the received reference signal and store the estimated value together with the associated time, frequency, position and / or spatial orientation information. Depending on its configuration, data collection may be performed on demand, periodically or continuously. Further, the collected data may be stored at the BS, a node in the core network and / or a server of a network operator. The estimated values ​​may include one or more of the following: RSRP / RSRQ / RSSI Channel quality Beam related parameters MI MO related parameters Interference and noise related parameters Data collection and storage based on measurement reports or UE feedback: According to an embodiment, the BS may collect data according to a data collection framework based on measurement reports and / or feedback received from the UE. The BS may extract values ​​from the received measurement reports and / or feedback and store the values ​​together with associated time, frequency, location and / or spatial orientation information. The data collection step may be performed on demand, periodically or continuously according to the configuration. Further, the collected data may be stored at the BS, a node in the core network and / or a server of the network operator. In an embodiment, the UE (200) may provide the following to the BS (200): UE location information, including coordinates (2D or 3D), UE speed, UE direction or trajectory vector, or serving cell ID. UE measurement reports, including beam-level or cell-level measurements of RSRP, RSRQ, SINR or SNR measured by the UE.

[0199] Data collection and storage at UE: According to an embodiment, the UE (200) can collect and store data received from the base station (300) according to the data collection framework. AI / ML model configuration:

[0200] According to an embodiment, the UE (200) and / or the base station (300) may be deployed with a single AI / ML model, an AI / ML model series, a group of AI / ML models and / or a group of AI / ML model series. As shown in FIG. 5 for the UE (200) and FIG. 6 for the BS (300), a group of AI / ML model series corresponds to multiple AI / ML model series. For example, the CSI module may represent AI / ML model series 1, the beamforming module may represent AI / ML model series 2, the positioning module may represent AI / ML model series 3, and the power control module may represent AI / ML model series 4.

[0201] The AI / ML model family may include different configurations of AI / ML models for a specific function or different AI / ML models for a specific function. For example, the function may correspond to any one of CSI prediction, CSI compression, beam management, positioning, or power control.

[0202] A set of AI / ML models may include two or more AI / ML models for a specific function. For example, the function may correspond to any one of CSI prediction, CSI compression, beam management, positioning, or power control.

[0203] A set of AI / ML model series may include two or more AI / ML model series for a specific function. For example, the function may correspond to any one of CSI prediction, CSI compression, beam management, positioning, or power control. Each series may correspond to a different configuration of the AI / ML model, and there may be two or more AI / ML models for a specific function, or each series may correspond to a different type of input. The input may correspond to a reference signal input such as CSI-RS, DMRS, SSB, SRS, or PT-RS. For example, a set of AI / ML model series corresponds to a downlink, where a first series corresponds to two or more AI / ML models that take CSI-RS as input, and a set of second series may correspond to two or more AI / ML models that take SSB signals as input.

[0204] According to an embodiment, an example of different configurations of the AI / ML model may include a first configuration of the AI / ML model corresponding to a high SNR situation and another configuration of the AI / ML model corresponding to a low SNR situation. Adapting the same AI / ML model to different situations by adjusting various AI / ML model parameters corresponds to different configurations of the AI / ML model. Another example of different configurations of the AI / ML model may include a first configuration of the AI / ML model corresponding to a high mobility situation and another configuration of the AI / ML model corresponding to a low mobility situation.

[0205] According to an embodiment, the AI / ML models in the AI / ML model series can be trained for specific propagation environments of wireless signals (e.g., LOS and NLOS scenarios, indoor and outdoor scenarios, slow and fast moving scenarios). In different scenarios, the characteristics of the wireless channel can be diverse, for example, the multipath distribution characteristics and channel sparsity can be very different. AI / ML is good at learning channel characteristics in a data-driven manner and performing various signal processing tasks more efficiently. The AI / ML models in the AI / ML model series can be trained using scenario-specific training data sets, and suitable AI / ML models can be selected from the AI / ML model series for inference to adapt to various scenarios.

[0206] The broader concepts are not limited to the set of AI / ML modules and / or models disclosed in the description, as there may be additional AI / ML modules and / or models depending on the needs and requirements of network operators and UE manufacturers. Signals for AI / ML model training

[0207] according to Fig.10 In the presented embodiments, the AI / ML model can be used in a BS (such as a combination of Figure 3 and / or 6 described in the BS (300)) training (BS training), UE (200) (for example, in conjunction with Figure 2The BS (300) may receive UE capability information (1000), which may include one or more of the following: UE processing capability (AI / ML processor configuration, such as, for example, type of processor, type of processor configuration, or number of operations per second), memory configuration (size or available space). UE processing capability may include UE training processing capability or UE inference processing capability. Indications for supported AI / ML model formats, such as the AI / ML model formats shown in Table 2. Indication of support for collaboration categories: o Category 1: No cooperation between UE and BS. o Category 2: Signaling-based collaboration between UE and BS, without AI / ML model transmission. o Category 3: Signaling-based collaboration between UE and BS with AI / ML model transmission. o Category 4: Joint (UE and BS) AI / ML model training and / or inference. UE category indicating supported UE features as defined by 3GPP. UE category may include UE types such as reduced capability UE, URLLC UE or eMBB UE. Configured AI / ML model families, such as, for example, CSI compression, CSI prediction, beam prediction, positioning, power control. The AI / ML model family may be represented by a unique identifier. The configured AI / ML models of each family, such as one or more identifiers of one or more AI / ML models within an AI / ML model family or one or more identifiers of one or more AI / ML model configurations. One or more configured AI / ML model configurations, including information of one or more supported AI / ML features, such as, for example, the number of layers, the number of hidden layers, the type of neural network (e.g., DNN, CNN, RNN, or LSTM), the number of nodes per layer, the number of activation units per layer, loss function selection, batch size, pooling size, selection of activation function (e.g., Sigmoid, ReLU, or Tanh), selection of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer).

[0208] Based on the received UE capability information, the BS (300) may determine an AI / ML model and a reference signal configuration (1010), and transmit RRC signaling and / or physical layer signaling (1020) indicating the reference signal configuration, which may include, for example, one or more of the following: a reference signal pattern, a reference signal resource, a periodicity, a resource offset, and / or an antenna port. The BS (300) may also transmit trigger signaling as part of the reference signal configuration (1020) or as a separate signaling. The trigger signaling includes trigger information for indicating to the UE (200) to start collecting data for AI / ML model training at the BS (300). After transmitting the reference signal configuration information (1020) and / or the trigger signaling, the BS (300) may transmit a reference signal (1030) to the UE (200) for, for example, one or more of channel estimation, CSI reporting, beam measurement, determining positioning, and / or measuring power. Based on the received reference signal (1030), the UE (200) may transmit a trigger signal to the BS (300) transmits a measurement report (1040) indicating such measurements, including, for example, one or more of: a CSI report, a channel characteristic vector, a beam measurement report, a positioning report, a power measurement report, a UE location, a UE direction (or trajectory) vector, and / or a UE velocity. The BS (300) can use the received measurement reports to train and validate the AI / ML model (1050). Once the AI / ML model (1050) is trained and validated, the BS (300) can transmit a message (1060) to the UE (200) containing information for downloading the AI / ML model, or transmit a message (1060) to the UE (200) containing AI / ML model configuration parameters, which include trained AI / ML model parameters such as, for example, weights, biases, the number of AI / ML tasks, one or more AI / ML task identifiers, and / or cluster centroids in the case of clustering. UE (200) can download the AI / ML model (1060) using the AI / ML model download information, or receive the AI / ML model configuration parameters (1060) contained in the message. The AI / ML model download information (1060) can include a URL to be used for downloading. The BS (300) can send the AI / ML model to the UE using physical layer signaling and / or RRC signaling. (200) transmits a message (1060) containing information for downloading an AI / ML model, or transmits a message (1060) containing AI / ML model configuration parameters to a UE (200).

[0209] AI / ML model training can include supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning (RL) The advantage of training at the BS is that the BS can collect training data from multiple UEs, or the BS can simulate the training data.

[0210] according to Fig.11 In the presented embodiment, the AI / ML model may be trained on the UE (200) (training on the UE), and the UE (200) may upload the AI / ML model or transmit the trained AI / ML model parameters to the BS (300). The BS (300) may receive the UE capability information and / or the AI / ML model information (1100). The UE capability information or the AI / ML model information (1100) may include, for example, one or more of the following: UE processing capability (AI / ML processor configuration, such as, for example, type of processor, type of processor configuration, or number of operations per second), memory configuration (size or available space). UE processing capability may include UE training processing capability or UE inference processing capability. Indications for supported AI / ML model formats, such as the AI / ML model formats shown in Table 2. Indication of support for collaboration categories: o Category 1: No cooperation between UE and BS. o Category 2: Signaling-based collaboration between UE and BS, without AI / ML model transmission o Category 3: Signaling-based collaboration between UE and BS with AI / ML model transmission o Category 4: Joint (UE and BS) AI / ML model training and / or inference. UE category indicating supported UE features as defined by 3GPP. UE category may include UE types such as reduced capability UE, URLLC UE or eMBB UE. Configured AI / ML model families, such as, for example, CSI compression, CSI prediction, beam prediction, positioning, power control. The AI / ML model family may be represented by a unique identifier. The configured AI / ML models of each family, such as one or more identifiers of one or more AI / ML models within an AI / ML model family or one or more identifiers of one or more AI / ML model configurations. One or more configured AI / ML model configurations, including information about one or more supported AI / ML features, such as, for example, the number of layers, the number of hidden layers, the type of neural network (e.g., DNN, CNN, RNN, or LSTM), the number of nodes per layer, the number of activation units per layer, loss function selection, batch size, pooling size, selection of activation function (e.g., Sigmoid, ReLU, or Tanh), selection of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer).

[0211] Based on the received UE capability information and / or AI / ML model information (1100), the BS (300) may determine the AI / ML model and reference signal configuration (1110), and transmit RRC signaling or physical layer signaling (1120) indicating the reference signal configuration, which may include one or more of the following: reference signal pattern, reference signal resource, periodicity, resource offset and / or antenna port. The BS (300) may also transmit trigger signaling as part of the reference signal configuration (1120) or as a separate signaling. The trigger signaling includes trigger information for indicating to the UE (200) to start collecting data for AI / ML model training at the UE (200). After transmitting the reference signal configuration (1120), the BS (300) may transmit a reference signal (1130) to the UE (200) for training and verifying the AI / ML model (1140). Once the AI / ML model (1140) is trained and validated, the UE (200) may transmit a message (1150) containing AI / ML model upload information to the BS (300), or transmit a message (1150) containing AI / ML model configuration parameters to the BS (300), including, for example, an AI / ML model family identifier, an AI / ML model identifier, an AI / ML model configuration identifier, trained AI / ML model parameters (such as weights, biases, the number of AI / ML tasks, one or more AI / ML task identifiers, and / or cluster centroids in the case of clustering), and / or hyperparameters used during training (such as, for example, the number of layers, the number of hidden layers, the type of neural network (e.g., DNN, CNN, RNN, or LSTM), the number of nodes per layer, the number of activation units per layer, the choice of loss function, batch size, pooling size, the choice of activation function (e.g., Sigmoid, ReLU, or Tanh), the choice of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer)). The UE (200) may upload the AI / ML model to a predefined location or transmit the AI / ML model configuration parameters (1150) contained in the message. The AI / ML model upload information may include an indicator indicating a successful AI / ML model upload by the UE (200). The UE (200) may transmit the message (1150) containing the AI / ML model upload information to the BS (300) and / or transmit the message (1150) containing the AI / ML model configuration parameters to the BS (300) using physical layer signaling or RRC signaling.

[0212] AI / ML model training can include supervised learning, unsupervised learning, semi-supervised learning, and / or reinforcement learning (RL). Signaling for AI / ML model updates

[0213] according to Fig.12 In the presented embodiment, BS (300) (e.g., as combined with Figure 3 6) may be sent to the UE (200) in the UE AI / ML request (1200) (e.g., as described in conjunction with Figure 2 5 ) to request information about its AI / ML capabilities and / or supported / configured AI / ML models. The UE (200) may respond with AI / ML information (1210) including one or more of the following: UE processing capability (AI / ML processor configuration, such as, for example, type of processor, type of processor configuration, or number of operations per second), memory configuration (size or available space). UE processing capability may include UE training processing capability or UE inference processing capability. Indications for supported AI / ML model formats, such as the AI / ML model formats shown in Table 2. Indication of support for collaboration categories: o Category 1: No cooperation between UE and BS. o Category 2: Signaling-based collaboration between UE and BS, without AI / ML model transmission o Category 3: Signaling-based collaboration between UE and BS with AI / ML model transmission o Category 4: Joint (UE and BS) AI / ML model training and / or inference. UE category indicating supported UE features as defined by 3GPP. UE category may include UE types such as reduced capability UE, URLLC UE or eMBB UE. Configured AI / ML model families, such as, for example, CSI compression, CSI prediction, beam prediction, positioning, power control. The AI / ML model family may be represented by a unique identifier. The configured AI / ML models of each family, such as, for example, one or more identifiers of one or more AI / ML models within a family of AI / ML models or one or more identifiers of one or more AI / ML model configurations. One or more configured AI / ML model configurations, including one or more supported AI / ML features, such as, for example, the number of layers, the number of hidden layers, the type of neural network (e.g., DNN, CNN, RNN, or LSTM), the number of nodes per layer, the number of activation units per layer, loss function selection, batch size, pooling size, selection of activation function (e.g., Sigmoid, ReLU, or Tanh), selection of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer).

[0214] Based on the AI / ML information (1210) received from the UE (200), the BS (300) may determine whether to update the AI / ML models supported / configured by the UE. The update may include the installation of one or more new AI / ML models or one or more AI / ML model families, and / or the reconfiguration of one or more existing AI / ML models or one or more AI / ML model families based on the UE's capabilities. The BS (300) may send a request to the UE to update the AI / ML models supported / configured by the UE. (200) Transmitting a request for an AI / ML model update (1220), the request including whether to install a new AI / ML model and / or reconfigure an existing model. The AI / ML model update request (1220) may include identifiers of one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families.

[0215] The UE (200) may download (1230) and install one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families indicated in the request (1220) based on the received request (1220) for AI / ML model updates. The UE (200) may download (1230) one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families from any one of the BS (1220), the core network entity responsible for the AI / ML model in the RAN network, an edge server, or a server of a network operator. The location (1220) from which the one or more AI / ML models and / or one or more AI / ML model families are to be downloaded may be indicated in the request.

[0216] After downloading and installing (1230) one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families, the UE (200) may transmit to the BS (300) a confirmation indicating whether the one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families were downloaded and whether the installation was successful or failed (1240). In the event of a failure, the UE (200) may indicate which AI / ML models, AI / ML model configurations, and / or AI / ML model families were not successfully downloaded or installed (1230). The UE (200) may indicate success or failure by using a bit pattern indicating one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families.

[0217] After successfully installing one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series (1230), if more than one AI / ML model series is installed, the UE (200) may start monitoring the performance of one or more AI / ML models or one or more AI / ML model configurations on the AI / ML model series (1250), otherwise, the UE (200) may start monitoring the performance of the installed one or more AI / ML models or one or more AI / ML model configurations (1250). Based on the monitored performance (1250), the UE (200) may share AI / ML model performance feedback (1260) with the BS (300). The AI / ML model performance feedback (1260) may include one or more identifiers of one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series. The AI / ML model performance feedback (1260) may also include performance parameters corresponding to the indicated one or more AI / ML models or one or more AI / ML model configurations, such as classification metrics (e.g., accuracy, precision, recall, F1 score, or confusion matrix), regression metrics (e.g., mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), normalized mean square error (NMSE), decision coefficient (commonly referred to as R-squared) and / or adjusted R-squared), and / or metrics for online iterative optimization (such as optimization performance across iterations / operations).

[0218] The BS (300) may determine one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families performance (1270) based on the performance feedback (1260) received from the UE (200). Based on the performance determination (1270) and comparing with the required performance threshold of the received performance metric, the BS (300) may decide whether to update one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families, update the reasoning of one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families, retrain one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families, replace one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families, switch one or more AI / ML models, one or more AI / ML model configurations, and / or activate / deactivate one or more AI / ML models, one or more AI / ML model configurations, and / or AI / ML model families.

[0219] The BS (300) may transmit a request (1280) to the UE (200) for updating one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families, updating inference of one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families, retraining one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families, replacing one or more AI / ML models or one or more AI / ML model configurations, switching one or more AI / ML models or one or more AI / ML model configurations, or activating / deactivating one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families. The request (1280) may include an identifier of the indicated one or more AI / ML models or one or more AI / ML model configurations and a request type indicating whether it is a request for one or more AI / ML models or one or more AI / ML model configurations to update, update inference, retrain, replace, switch, and / or activate / deactivate. AI / ML model update / reconfiguration in case of UE mobility

[0220] According to an embodiment, one or more AI / ML models may be deployed in a cell-centric or BS-centric arrangement, i.e., a cell or BS may be configured with one or more different AI / ML models, one or more different AI / ML model series, and / or one or more different AI / ML model configurations.

[0221] like Fig.13a and 14a As shown, the first BS (BS1) (300-1) (for example, BS1 may refer to the above-mentioned Figure 3 and / or the BS described in FIG. 6) may be configured with a first AI / ML model (AI / ML model 1), and a second BS (BS2) (300-2) (for example, BS2 may refer to the BS described above in conjunction with Figure 3 And / or the BS described in Figure 6) can be configured with a second AI / ML model. Fig.13a and 14aAI / ML models 1 and 2 may represent any one of a single AI / ML model, a configuration of an AI / ML model, an AI / ML model series, a group of AI / ML models, and / or a group of AI / ML model series. For example, AI / ML model 1 may represent a CSI AI / ML model, a configuration of an AI / ML model, a CSI AI / ML model series, a group of AI / ML models including two or more AI / ML models (including CSI, beamforming, power control, and / or positioning), a group of AI / ML model series including two or more AI / ML model series (including CSI, beamforming, power control, and / or positioning). Similarly, AI / ML model 2 can represent a CSI AI / ML model, a configuration of an AI / ML model, a CSI AI / ML model family, a group of AI / ML models including two or more AI / ML models (including CSI, beamforming, power control and / or positioning), and a group of AI / ML model families including an AI / ML model family of two or more AI / ML model families (including CSI, beamforming, power control and / or positioning).

[0222] According to Fig.13a In the embodiment shown, when the UE (200) (for example, as described above in conjunction with Figure 2 and / or the UE described in FIG. 5) switches from BS1 (300-1) to BS2 (300-2) (for example, BS1 or BS2 may refer to the above combined Figure 36 ), information about one or more AI / ML models used by BS2 (300) may be provided to the UE (200). The AI / ML model 2 information may be provided in the handover command (1300). The AI / ML model 2 information may include one or more of the following: an AI / ML model version number, one or more identifiers or one or more indicators (or one or more indexes) of one or more AI / ML models, one or more identifiers or one or more indicators of configurations of one or more AI / ML models, one or more identifiers or one or more indicators of series of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a group of AI / ML models, and / or one or more identifiers or one or more indicators (or one or more indexes) of a group of AI / ML models, details of performance metrics to be used, AI / ML model parameters (such as For example, one or more of the following: coefficients, weights, biases, number of AI / ML tasks, one or more AI / ML task identifiers (or one or more indexes), classifiers (such as, for example, regression, KNN, vector machine, decision tree, or principal component), cluster centroids in clustering), and / or hyperparameters (such as, for example, number of layers, number of hidden layers, type of neural network (e.g., DNN, CNN, RNN, or LSTM), number of nodes per layer, number of activation units per layer, choice of loss function, batch size, pooling size, choice of activation function (e.g., Sigmoid, ReLU, or Tanh), choice of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer)). The UE (200) may receive a switching command (1300) including AI / ML model 2 information and configure its AI / ML model according to the information in AI / ML model 2.

[0223] according to Fig.13a In the illustrated embodiment, the AI / ML model information may be provided via RRC, MAC or physical layer signaling. It should be understood by those skilled in the art that, in the case of transmission via MAC or physical layer, the AI / ML model information may be transmitted in a message other than the handover command (1300).

[0224] According to an embodiment, when a UE (eg, as described above in conjunction with Figure 2 5) is about to switch from the coverage of BS1 to BS2 (for example, BS1 or BS2 may refer to the above combined Figure 36 ), or when a handover procedure is initiated, BS1 may transmit and the UE may receive signaling containing information about one or more AI / ML models and / or one or more AI / ML model configurations to be used within the coverage of BS2. The AI / ML engine of the UE within the coverage of BS1 (e.g., Figure 2 and / or the AI / ML engine (211) in FIG. 5) may operate or use a first AI / ML model having a first AI / ML model configuration. The signaling may be performed by the UE in a handover command (such as Fig.13aThe signaling may be received in a handover command (1300) in , or the signaling may be received by the UE in other RRC, MAC or physical layer messages. The signaling may be provided before the handover process is initiated, or may be provided in a handover trigger message. Information on one or more AI / ML models and / or one or more AI / ML model configurations to be used within the coverage of BS2 includes information on a second AI / ML model and / or a second AI / ML model configuration. The information on the second AI / ML model may include one or more of the following: an AI / ML model version number, an identifier or index of the second AI / ML model, an identifier or index of a series of the second AI / ML model, download information (e.g., URL information) of the second AI / ML model. The information of the second AI / ML model configuration may include one or more of the following: an identifier or index of the second AI / ML model configuration, download information of the second AI / ML model configuration (e.g., URL information), second AI / ML model parameters (such as, for example, one or more of the following: coefficients, weights, biases, number of AI / ML tasks, one or more AI / ML task identifiers (or one or more indexes), classifiers (such as, for example, regression, KNN, vector machine, decision tree, or principal component), cluster centroids in clustering), and / or hyperparameters (such as, for example, the number of layers, the number of hidden layers, the type of neural network (e.g., DNN, CNN, RNN, or LSTM), the number of nodes per layer, the number of activation units per layer, selection of loss function, batch size, pooling size, selection of activation function (e.g., Sigmoid, ReLU, or Tanh), selection of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer)). In the case where the download information is included in the signaling, the UE may download the indicated second AI / ML model or second AI / ML model configuration from the indicated download location, or if no download location is indicated, download the indicated second AI / ML model or second AI / ML model configuration from a predefined download location. The predefined download location may include the location of a server storing the AI / ML model and the corresponding AI / ML model configuration information. Based on the received information of the second AI / ML model and / or the second AI / ML model configuration, the UE switches the operation of the AI / ML engine from the first AI / ML model having the first AI / ML model configuration to the second AI / ML model and the second AI / ML model configuration, or the UE switches the operation of the AI / ML engine from the first AI / ML model having the first AI / ML model configuration to the first AI / ML model and the second AI / ML model configuration (if only the second AI / ML model configuration is received in the signaling received by the UE from BS1).After switching to the second AI / ML model and the second AI / ML model configuration or switching to the first AI / ML model and the second AI / ML model configuration, the UE sends an uplink signaling to BS2 to indicate the updated configuration. The uplink signaling includes information of the switched configuration.

[0225] According to an embodiment, when a UE (eg, as described above in conjunction with Figure 2 and / or UE described in 5) is about to switch from the coverage of BS1 to BS2 (for example, BS1 or BS2 may refer to the above combined Figure 3 6), or when a handover procedure is initiated, BS1 may transmit and the UE may receive signaling containing information about one or more AI / ML models to be used within the coverage of BS2. The AI / ML engine of the UE within the coverage of BS1 (e.g., Figure 2 and / or the AI / ML engine (211) in FIG. 5) may operate or use a first AI / ML model or a first set of AI / ML models. The signaling may be performed by the UE in a handover command (such as Fig.13aThe signaling may be received in a handover command (1300) in , or the signaling may be received by the UE in other RRC, MAC or physical layer messages. The signaling may be provided before the handover process is initiated, or may be provided in a handover trigger message. The information of one or more AI / ML models to be used within the coverage of BS2 includes information of a second AI / ML model or a second group of AI / ML models. The information of the second AI / ML model may include one or more of the following: an AI / ML model version number of the second AI / ML model, an identifier or index of the second AI / ML model, an identifier or index of a series of the second AI / ML model, download information of the second AI / ML model (e.g., URL information). The information of the second group of AI / ML models may include one or more of the following: an AI / ML model version number of the second group of AI / ML models, an identifier or index of the second group of AI / ML models, an identifier or index of a series of the second group of AI / ML models, download information of the second group of AI / ML models. In the case where the download information is included in the signaling, the UE may download the indicated second AI / ML model or the second group of AI / ML models from the indicated download location, or if the download location is not indicated, download the indicated second AI / ML model or the second group of AI / ML models from a predefined download location. The predefined download location may include the location of the server storing the AI / ML model and the corresponding information. The identifier or index of the second group of AI / ML models and the identifier or index of the series of the second group of AI / ML models may be indicated by a bit pattern in which each bit corresponds to the AI / ML model or the AI / ML model series in a predefined order. The predefined order may be an increasing order of the AI / ML model index or the AI / ML model series index starting from the smallest index. Based on the received information of the second AI / ML model or the second group of AI / ML models, the UE switches the operation of the AI / ML engine from the first AI / ML model to the second AI / ML model, or from the first group of AI / ML models to the second group of AI / ML models. After switching to the second AI / ML model or the second group of AI / ML models, the UE sends uplink signaling to BS2 to indicate the updated configuration. The uplink signaling includes information of a second AI / ML model or a second group of AI / ML models.

[0226] according to Fig.13b In the embodiment shown, BS (300-1) (for example, as described above in conjunction with Figure 3 6) can provide a UE (200) (e.g., as described above in conjunction with the BS described in conjunction with FIG. 6) with a predicted UE position determined based on the UE speed and / or trajectory information (using a positioning module (620) at the BS (300-1)). Figure 2The handover command (1310) may include a time offset parameter indicating a time after which the UE may handover to another BS (300-2). The handover command may implicitly or explicitly indicate to the UE (200) that the UE may stop / not perform neighbor cell measurements.

[0227] according to Fig.13c In the embodiment shown, the UE (for example, as described above in conjunction with Figure 2 5) operates a first AI / ML model (1300c) having a first AI / ML model configuration while being within the coverage of a first base station (BS1). The UE operates an AI / ML engine (e.g., AI / ML engine (211)) using the first AI / ML model having the first AI / ML model configuration. The UE transmits one or more UE parameters to BS1, the one or more UE parameters including UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of BS2 (1310c). BS1 or BS2 may refer to the above in combination with Figure 3and / or the BS described in FIG. 6 . After sending the UE parameters, the UE receives a first signaling from BS1, the first signaling including information (1320c) configured by the second AI / ML model. The first signaling is transmitted by BS1 based on a determination of a potential handover of the UE from the coverage of BS1 to the coverage of the second base station (BS2). The determination of the handover is based on the UE parameters. BS1 uses an AI / ML engine and a positioning module to predict the future location of the UE by using the received UE parameters and / or a reference signal transmitted by the UE in the uplink. If the future location is identified as being within the coverage of a base station other than BS1 (e.g., BS2), BS1 may initiate a handover process, or predict the initiation of the handover process after a time offset in BS1, and transmit the first signaling to the UE. The first signaling may be transmitted after the initiation of the handover process in BS1, or the first signaling may be transmitted before a predefined time offset from the initiation of the handover process in BS1. The first signaling may be transmitted by BS1 to initiate a handover process in the UE. When the first signaling is received, the UE may still be within the coverage of BS1, and the first signaling may include a time offset parameter for initiating a handover process in the UE after a predefined time. The UE may adjust the received time offset parameter for initiating the handover process according to changes in one or more aspects of the UE speed, trajectory, signal quality of BS1, and signal quality of BS2. The UE may use the AI / ML model for handover prediction to adjust the time offset parameter for initiating the handover process. The first signaling may also include information about the AI / ML model to be used for handover prediction, or include information about the AI / ML model configuration to be used for handover prediction. The information about the AI / ML model to be used for handover prediction includes an identifier or index of the AI / ML model. The information about the AI / ML model configuration to be used for handover prediction includes an identifier or index of the AI / ML model configuration. The UE may configure a first AI / ML model with a second AI / ML model configuration indicated in the received first signaling. After the first AI / ML model with the second AI / ML model configuration is configured by the UE, the UE begins to use the first AI / ML model with the second AI / ML model configuration to process information transmitted or received on a signal or channel, transmitted to or received from BS2. The first signaling may be received in a handover command message received in the RRC layer.

[0228] according to Fig.13d In the embodiment shown, the UE (for example, as described above in conjunction with Figure 25) while operating a first AI / ML model or a first set of AI / ML models (1300d) within the coverage of a first base station (BS1). The UE operates an AI / ML engine (e.g., AI / ML engine (211)) using the first AI / ML model or the first set of AI / ML models. The UE transmits one or more UE parameters to BS1, the one or more UE parameters including UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of BS2 (1310d). BS1 or BS2 may refer to the above in combination Figure 3and / or the BS described in FIG. 6 . After sending the UE parameters, the UE receives a first signaling from BS1, the first signaling including information of a second AI / ML model or a second group of AI / ML models. The first signaling is transmitted by BS1 based on a determination of a potential handover of the UE from the coverage of BS1 to the coverage of a second base station (BS2). The determination of the handover is made based on the UE parameters (1320d). BS1 uses an AI / ML engine and a positioning module to predict the future location of the UE by using the received UE parameters and / or a reference signal transmitted by the UE in the uplink. If the future location is identified as being within the coverage of a base station other than BS1 (e.g., BS2), BS1 may initiate a handover process, or predict the initiation of the handover process after a time offset in BS1, and transmit the first signaling to the UE. The first signaling may be transmitted after the handover process in BS1 is initiated, or the first signaling may be transmitted before a predefined time offset from the initiation of the handover process in BS1. The first signaling may be transmitted by BS1 to initiate a handover process in the UE. When the first signaling is received, the UE may still be within the coverage of BS1, and the first signaling may include a time offset parameter for initiating a handover process in the UE after a predefined time. The UE may adjust the received time offset parameter to initiate a handover process according to changes in one or more aspects of the UE speed, trajectory, signal quality of BS1, and signal quality of BS2. The UE may use an AI / ML model for handover prediction to adjust the time offset parameter for initiating a handover process. The first signaling may also include information about an AI / ML model to be used for handover prediction, or information about an AI / ML model configuration to be used for handover prediction. The information about the AI / ML model to be used for handover prediction includes an identifier or index of the AI / ML model. The information about the AI / ML model configuration to be used for handover prediction includes an identifier or index of the AI / ML model configuration. The UE may configure a second AI / ML model, or a second group of AI / ML models indicated in the received first signaling. After the second AI / ML model or the second group of AI / ML models is configured by the UE, the UE begins to use the second AI / ML model or the second group of AI / ML models to process information transmitted or received on a signal or channel, transmitted to or received from BS2. The first signaling may be received in a handover command message received in the RRC layer.

[0229] according to Fig.13e In the embodiment shown, the first base station (BS1 (for example, BS1 may refer to the Figure 3and / or the BS described in FIG. 6)) receives one or more UE parameters from the UE, the one or more UE parameters including UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of BS2. The UE operates a first AI / ML model (1310e) having a first AI / ML model configuration. BS1 uses an AI / ML engine and a positioning module to predict the future location of the UE by using the received UE parameters and / or a reference signal transmitted by the UE in an uplink. If the future location is identified as being within the coverage of a base station other than BS1 (e.g., a second base station (BS2)), BS1 may initiate a handover process, or predict the initiation of a handover process after a time offset in BS1, and transmit a first signaling to the UE. BS1 transmits a first signaling to the UE, the first signaling including information configured by a second AI / ML model. The first signaling is transmitted by BS1 based on a determination of a potential handover of the UE from the coverage of BS1 to the coverage of the second base station (BS2). The determination of the potential handover is performed based on the UE parameters and / or reference signals transmitted by the UE in the uplink (1320e). BS2 may refer to the above combination Figure 3 And / or the BS described in Figure 6. The first signaling may be transmitted after the initiation of the handover process in BS1, or the first signaling may be transmitted before a predefined time offset from the initiation of the handover process in BS1. The first signaling may be transmitted by BS1 to initiate the handover process in the UE. The first signaling may include a time offset parameter for initiating the handover process in the UE after a predefined time. The first signaling may also include information about an AI / ML model to be used for handover prediction, or include information about an AI / ML model configuration to be used for handover prediction. The information about the AI / ML model to be used for handover prediction includes an identifier or index of the AI / ML model. The information about the AI / ML model configuration to be used for handover prediction includes an identifier or index of the AI / ML model configuration. The time offset parameter for initiating the handover process in the UE may be adjusted by the UE using the AI / ML model to be used for handover prediction or the AI / ML model configuration to be used for handover prediction.

[0230] according to Fig.13f In the embodiment shown, the first base station (BS1 (for example, BS1 may refer to the Figure 3and / or the BS described in Figure 6)) receives one or more UE parameters from the UE, the one or more UE parameters including UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of BS2. The UE operates a first AI / ML model or a first group of AI / ML models (1310f). BS1 uses the AI / ML engine and the positioning module to predict the future position of the UE by using the received UE parameters and / or the reference signal transmitted by the UE in the uplink. If the future position is identified as being within the coverage of a base station other than BS1 (e.g., a second base station (BS2)), BS1 may initiate a handover process, or predict the initiation of a handover process after a time offset in BS1, and transmit a first signaling to the UE. BS1 transmits a first signaling to the UE, the first signaling including information of a second AI / ML model or a second group of AI / ML models. The first signaling is transmitted by BS1 based on a determination of a potential handover of the UE from the coverage of BS1 to the coverage of the second base station (BS2). The determination of potential handover is performed based on UE parameters and / or reference signals transmitted by the UE in the uplink (1320f). BS2 may refer to the above-mentioned Figure 3 And / or the BS described in Figure 6. The first signaling may be transmitted after the initiation of the handover process in BS1, or the first signaling may be transmitted before a predefined time offset from the initiation of the handover process in BS1. The first signaling may be transmitted by BS1 to initiate the handover process in the UE. The first signaling may include a time offset parameter for initiating the handover process in the UE after a predefined time. The first signaling may also include information about an AI / ML model to be used for handover prediction, or include information about an AI / ML model configuration to be used for handover prediction. The information about the AI / ML model to be used for handover prediction includes an identifier or index of the AI / ML model. The information about the AI / ML model configuration to be used for handover prediction includes an identifier or index of the AI / ML model configuration. The time offset parameter for initiating the handover process in the UE may be adjusted by the UE using the AI / ML model to be used for handover prediction or the AI / ML model configuration to be used for handover prediction.

[0231] According to an embodiment, a UE (eg, as described above in conjunction with Figure 2 5) operates a first AI / ML model with a first AI / ML model configuration while being within the coverage of a first base station (BS1). The UE operates an AI / ML engine (e.g., AI / ML engine (211)) using the first AI / ML model with the first AI / ML model configuration. The UE transmits one or more UE parameters to BS1, the one or more UE parameters including UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of BS2 (1310c). BS1 or BS2 may refer to the above in combination with Figure 3 and / or the BS described in FIG. 6 . After sending the UE parameters, the UE receives a first signaling from BS1, the first signaling including information for switching the AI / ML engine to the non-AI / ML signal processing module. The first signaling is transmitted by BS1 based on a determination of a potential handover of the UE from the coverage of BS1 to the coverage of the second base station (BS2). The determination of the handover is based on the UE parameters. BS1 uses the AI / ML engine and the positioning module to predict the future location of the UE by using the received UE parameters and / or the reference signal transmitted by the UE in the uplink. If the future location is identified as being within the coverage of a base station other than BS1 (e.g., BS2), BS1 may initiate a handover process or predict the initiation of the handover process after a time offset in BS1, and transmit the first signaling to the UE. The first signaling may be transmitted after the handover process in BS1 is initiated, or the first signaling may be transmitted before a predefined time offset from the initiation of the handover process in BS1. The first signaling may be transmitted by BS1 to initiate a handover process in the UE. When the first signaling is received, the UE may still be within the coverage of BS1, and the first signaling may include a time offset parameter for initiating a handover process in the UE after a predefined time. The UE may adjust the received time offset parameter to initiate a handover process according to changes in one or more aspects of the UE speed, trajectory, signal quality of BS1, and signal quality of BS2. The UE may use an AI / ML model for handover prediction to adjust the time offset parameter for initiating a handover process. The first signaling may also include information about an AI / ML model to be used for handover prediction, or information about an AI / ML model configuration to be used for handover prediction. The information about the AI / ML model to be used for handover prediction includes an identifier or index of the AI / ML model. The information about the AI / ML model configuration to be used for handover prediction includes an identifier or index of the AI / ML model configuration. The UE may switch the AI / ML engine to a non-AI / ML signal processing module. After switching to the non-AI / ML signal processing module, the UE begins to use the non-AI / ML signal processing module to process information transmitted or received on a signal or channel, transmitted to BS2, or received from BS2. The first signaling may be received in a handover command message received in the RRC layer. The first signaling may include implicit signaling for switching to a non-AI / ML signal processing module. BS2 may have a different RAT (e.g., LTE) from BS1, or BS2 may not support the AI / ML model for processing signals, or BS2 may not have the ability to transmit / receive the signal configuration required to support the UE's AI / ML engine. Therefore, it is very important for the UE to switch to a non-AI / ML signal processing module.

[0232] according to Fig.14aIn the embodiment shown, when the UE (200) (for example, as described above in conjunction with Figure 2 5) from the first BS (BS1) (300-1) to the second BS (BS2) (300-2) (for example, BS1 or BS2 may refer to the above combined Figure 3 6), AI / ML model 2 information of the AI / ML model used by BS2 (300-2) may be provided to the UE (200). The UE (200) may receive a handover command (1400a) for initiating a handover from BS1 (300-1) to BS2 (300-2). In response to a PRACH (1410a) transmitted by the UE (200), the AI / ML model 2 information may be provided in a RACH response (1420a) transmitted by BS2 (300-2). The AI / ML model 2 information may include one or more of the following: an AI / ML model version number, one or more identifiers or one or more indicators (or one or more indexes) of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of configurations of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a series of one or more AI / ML models, and / or one or more identifiers or one or more indicators (or one or more indexes) of a group of AI / ML models, and / or one or more identifiers or one or more indicators (or one or more indexes) of a group of AI / ML models, details of performance metrics to be used, and A. I / ML model parameters (such as, for example, one or more of the following: coefficients, weights, biases, number of AI / ML tasks, one or more AI / ML task identifiers (or one or more indexes), classifiers (such as, for example, regression, KNN, vector machine, decision tree, or principal component), and / or cluster centroids in clustering), and / or hyperparameters (such as, the number of layers, the number of hidden layers, the type of neural network (e.g., DNN, CNN, RNN, or LSTM), the number of nodes per layer, the number of activation units per layer, the choice of loss function, batch size, pooling size, the choice of activation function (e.g., Sigmoid, ReLU, or Tanh), the choice of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer)). The UE (200) may receive a RACH response (1420a) including AI / ML model 2 information and configure its AI / ML model according to the information in AI / ML model 2.

[0233] according to Fig.14a In the embodiment shown, the UE (200) (e.g., as described above in conjunction with Figure 2and / or the UE described in Figure 5) provides update and / or reconfiguration information for a single AI / ML model, configuration of one or more AI / ML models, an AI / ML model series, a subset of an AI / ML model series, or a subset of an AI / ML model.

[0234] According to an embodiment, when a UE (eg, as described above in conjunction with Figure 2 5) from the first BS (BS1) to the second BS (BS2) (eg, BS1 or BS2 may refer to the above combined Figure 3 6), the UE may receive information from BS2 for switching the AI / ML engine to a non-AI / ML signal processing module. The information for switching the AI / ML engine to a non-AI / ML signal processing module depends on the BS2 capability for supporting the AI / ML model. In an embodiment, BS2 may broadcast its ability to support the AI / ML model within its coverage. For example, BS2 may transmit an indicator in one of the MIB or SIB to indicate its support for the AI / ML model. The UE may use the received capability information as implicit signaling for switching from the AI / ML engine to the non-AI / ML signal processing module. In another embodiment, BS2 may transmit explicit signaling for switching to the non-AI / ML signal processing module to the UE. The explicit signaling may be a flag bit in physical or RRC signaling. In yet another embodiment, the UE may determine to switch from the AI / ML engine to the non-AI / ML signal processing module based on parameters of BS2, such as the RAT type of BS2 (e.g., LTE) or the type of BS2 (relay node or reduced capability base station) of BS2.

[0235] according to Fig.14b In the embodiment shown, the UE (for example, as described above in conjunction with Figure 2 5) operates a first AI / ML model (1400b) having a first AI / ML model configuration while being within the coverage of a first base station (BS1). The UE operates an AI / ML engine (e.g., AI / ML engine (211)) using the first AI / ML model having the first AI / ML model configuration. The UE transmits one or more UE parameters to BS1, the one or more UE parameters including UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of BS2 (1410b). BS1 or BS2 may refer to the above in combination with Figure 3and / or the BS described in FIG. 6 . After sending the UE parameters, the UE receives a first signaling from BS1, the first signaling including handover information to the second base station (BS2) (1420b). The first signaling is transmitted by BS1 based on a determination of a potential handover of the UE from the coverage of BS1 to the coverage of the second base station (BS2). The determination of the handover is based on the UE parameters. BS1 uses an AI / ML engine and a positioning module to predict the future location of the UE by using the received UE parameters and / or a reference signal transmitted by the UE in the uplink. If the future location is identified as being within the coverage of a base station other than BS1 (e.g., BS2), BS1 may initiate a handover process or predict the initiation of a handover process after a time offset in BS1, and transmit the first signaling to the UE. The first signaling may be transmitted after the handover process in BS1 is initiated, or the first signaling may be transmitted before a predefined time offset from the initiation of the handover process in BS1. The first signaling may be transmitted by BS1 to initiate a handover process in the UE. When the first signaling is received, the UE may still be within the coverage of BS1, and the first signaling may include a time offset parameter for initiating a handover process in the UE after a predefined time. The UE may adjust the received time offset parameter for initiating the handover process according to changes in one or more aspects of the UE speed, trajectory, signal quality of BS1, and signal quality of BS2. The UE may use the AI / ML model for handover prediction to adjust the time offset parameter for initiating the handover process. The first signaling may also include information of the AI / ML model to be used for handover prediction, or include information of the AI / ML model configuration to be used for handover prediction. The information of the AI / ML model to be used for handover prediction includes an identifier or index of the AI / ML model. The information of the AI / ML model configuration to be used for handover prediction includes an identifier or index of the AI / ML model configuration. The UE transmits a PRACH signal to BS2 (1430b). The PRACH signal is transmitted based on the received first signaling. The UE receives a second signaling from BS2, which includes information of a second AI / ML model configuration (1440b). The UE configures the first AI / ML model with a second AI / ML model configuration based on the received second signaling. The UE processes information transmitted or received on a signal or channel, transmitted to or received from BS2, using the first AI / ML model with the second AI / ML model configuration. The first signaling may be received in a handover command message received in the RRC layer. The second signaling may be received in response to a PRACH transmitted by the UE.

[0236] according to Fig.14c In the embodiment shown, the UE (for example, as described above in conjunction with Figure 25 ) operates a first AI / ML model with a first AI / ML model configuration while being within the coverage of a first base station (BS1) (1400c). The UE operates an AI / ML engine (e.g., AI / ML engine (211)) using the first AI / ML model with the first AI / ML model configuration. The UE transmits one or more UE parameters to BS1, the one or more UE parameters including UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of BS2 (1410c). BS1 or BS2 may refer to the above in combination Figure 3and / or the BS described in FIG. 6 . After sending the UE parameters, the UE receives a first signaling from BS1, the first signaling including handover information to the second base station (BS2) (1420c). The first signaling is transmitted by BS1 based on a determination of a potential handover of the UE from the coverage of BS1 to the coverage of the second base station (BS2). The determination of the handover is based on the UE parameters. BS1 uses an AI / ML engine and a positioning module to predict the future location of the UE by using the received UE parameters and / or a reference signal transmitted by the UE in the uplink. If the future location is identified as being within the coverage of a base station other than BS1 (e.g., BS2), BS1 may initiate a handover process or predict the initiation of a handover process after a time offset in BS1, and transmit the first signaling to the UE. The first signaling may be transmitted after the handover process in BS1 is initiated, or the first signaling may be transmitted before a predefined time offset from the initiation of the handover process in BS1. The first signaling may be transmitted by BS1 to initiate a handover process in the UE. When the first signaling is received, the UE may still be within the coverage of BS1, and the first signaling may include a time offset parameter for initiating a handover process in the UE after a predefined time. The UE may adjust the received time offset parameter for initiating the handover process according to changes in one or more aspects of the UE speed, trajectory, signal quality of BS1, and signal quality of BS2. The UE may use the AI / ML model for handover prediction to adjust the time offset parameter for initiating the handover process. The first signaling may also include information about the AI / ML model to be used for handover prediction, or include information about the AI / ML model configuration to be used for handover prediction. The information about the AI / ML model used for handover prediction includes an identifier or index of the AI / ML model. The information about the AI / ML model configuration to be used for handover prediction includes an identifier or index of the AI / ML model configuration. The UE transmits a PRACH signal to BS2 (1430c). The PRACH signal is transmitted based on the received first signaling. The UE receives a second signaling from BS2, which includes information about a second AI / ML model or a second group of AI / ML models (1440c). The UE configures the AI / ML engine using a second AI / ML model or a second set of AI / ML models based on the received second signaling. The UE uses the second AI / ML model or a second set of AI / ML models to process information transmitted or received on a signal or channel, transmitted to or received from BS2. The first signaling may be received in a handover command message received in the RRC layer. The second signaling may be received in response to a PRACH transmitted by the UE.

[0237] according to Fig.14d In the embodiment shown, the second base station (BS2 (for example, BS1 may refer to the Figure 3And / or the BS described in Figure 6)) receives PRACH (1400d) from the UE. BS2 transmits a first signaling to the UE, and the first signaling includes information of a second AI / ML model configuration (1410d). The first signaling is transmitted by BS2 based on a comparison of the UE's AI / ML model information and the BS2 AI / ML model information. The UE's AI / ML model information includes information of a first AI / ML model with a first AI / ML model configuration, and the BS2 AI / ML model information includes information of a first AI / ML model with a second AI / ML model configuration. The UE's AI / ML model information is received by BS2 from the first base station (BS1). When the UE switches from the coverage of BS1 to BS2, the channel characteristics and beam direction of the UE change, and it is necessary to update the AI / ML model configuration.

[0238] according to Fig.14e In the embodiment shown, the second base station (BS2 (for example, BS1 may refer to the Figure 3 And / or the BS described in Figure 6)) receives PRACH (1400e) from the UE. BS2 transmits a first signaling to the UE, which includes information about a second AI / ML model or a second group of AI / ML models (1410e). The first signaling is transmitted by BS2 based on a comparison of the UE's AI / ML model information and the BS2 AI / ML model information. The UE's AI / ML model information includes information about a first AI / ML model or a first group of AI / ML models, and the BS2 AI / ML model information includes information about a second AI / ML model or a second group of AI / ML models. BS2 receives the UE's AI / ML model information from the first base station (BS1). When the UE switches from the coverage of BS1 to BS2, the UE's channel characteristics and beam direction change, and it is necessary to update the AI / ML model configuration.

[0239] according to Fig.15a In the illustrated embodiment, BS (300) (e.g., as described above in conjunction with Figure 36) can be configured to support two or more sectors, each sector corresponding to a different AI / ML model configuration (within the same AI / ML model family). For example, the BS cell coverage can be divided into 8 sectors (sectors 1-8), and each sector can be assigned a different AI / ML model configuration (1500a, 1510a, 1520a, 1530a, 1540a, 1550a, 1560a, 1570a), as shown in the figure. The AI / ML model configuration can be selected based on the characteristics of the sector (such as the presence of buildings, mountains, forests, the number of UEs, average UE speed, time of day / week, past interference data and / or NLOS conditions). According to an embodiment, UEs with similar UE capabilities in a sector can be configured with the same AI / ML model configuration (within the same AI / ML model family). According to another embodiment, UEs in a sector may be configured with an AI / ML model configuration (within the same AI / ML model family) based on UE-specific parameters such as UE speed, UE frequency band, time of day, current interference faced on the UE, and / or NLOS conditions. According to yet another embodiment, different AI / ML model configurations may be selected by the BS (300) and / or the UE (200) based on UE mobility. When the UE (200) moves from one sector to another, the BS (300) may indicate a new AI / ML model configuration in physical layer, MAC layer, or RRC layer signaling. The UE (200) may use the received signaling to select another AI / ML model configuration. The signaling may include one or more of the following: an AI / ML model identifier or indicator (or index), an AI / ML model configuration identifier or indicator (or index), a sector identifier or indicator (or index), and / or an AI / ML configuration parameter. Details of the configuration parameters may be identified from other parts of the present application.

[0240] according to Fig.15a In the embodiment shown, when the UE (200) (for example, as described above in conjunction with Figure 25 ) moves from one sector to another (e.g., the UE moves from sector 2 to sector 3, as shown in the figure), the UE (200) may select a new AI / ML model configuration (e.g., AI / ML model configuration 3 (1530)). The UE (200) may select another AI / ML model configuration (1530) based on AI / ML model performance or based on a predefined relationship between the sector and the AI / ML model configuration. In the event that the UE (200) selects a new AI / ML model configuration (1530) based on a predefined relationship between the sector and the AI / ML model configuration, the UE (200) may monitor changes in the sector and change the AI / ML model configuration accordingly. The BS (300) may indicate a sector identifier or indicator (or index) to help the UE (300) select the AI / ML model configuration.

[0241] according to Fig.15b In the illustrated embodiment, BS (300) (e.g., as described above in conjunction with Figure 3 6) can be configured to support two or more sectors, where each sector corresponds to a different and / or the same AI / ML model and / or AI / ML model configuration. For example, the BS cell coverage can be divided into 8 sectors, and each sector can be assigned a different and / or the same AI / ML model or AI / ML model configuration (1500b, 1510b, 1520b, 1530b, 1540b, 1550b, 1560b, 1570b), as shown. Some sectors can have the same AI / ML model and AI / ML model configuration (e.g., sectors 5 and 6) (1540b, 1550b), some other sectors can have the same AI / ML model and different AI / ML model configurations (e.g., sectors 1 and 2) (1500b, 1510b), and some other sectors can have different AI / ML models (e.g., sectors 2 and 3) (1510b, 1520b). The AI / ML model or AI / ML model configuration may be selected based on characteristics of the sector, such as the presence of buildings, mountains, forests, the number of UEs, average UE speed, time of day / week, past interference data, and / or NLOS conditions. According to an embodiment, UEs with similar UE capabilities in a sector may be configured with the same AI / ML model or AI / ML model configuration. According to another embodiment, the UEs in a sector may be configured with an AI / ML model or AI / ML model configuration based on UE-specific parameters, such as UE speed, UE frequency band, time of day, current interference faced by the UE, and / or NLOS conditions. According to yet another embodiment, when UE1 (200-1) (e.g., as described above in conjunction with Figure 25 ) moves from one sector to another sector (e.g., sector 2 to 3), UE1 (200-1) may select a new AI / ML model (1520b). UE1 (200-1) may select a new AI / ML model (1520b) based on the performance of the currently used AI / ML model (1510b) or based on a predefined relationship between the sector and the AI / ML model. In the case where UE1 (200-1) selects a new AI / ML model (1520b) based on a predefined relationship between the sector and the AI / ML model, UE1 (200-1) may monitor changes in the sector and change the AI / ML model accordingly. The BS (300) may indicate a sector identifier or indicator (or index) to help UE1 (200-1) select the AI / ML model. When UE2 (200-2) (e.g., as described above in conjunction with Figure 2 5) moves from one sector to another sector, UE2 (200-2) maintains the same AI / ML model configuration and / or AI / ML model (1540b). UE2 (200-2) may maintain the AI / ML model configuration and / or AI / ML model based on the performance of the currently used AI / ML model configuration and / or AI / ML model (1550b) or based on a predefined relationship between the sector and the AI / ML model configuration and / or AI / ML model. In the case where UE2 (200-2) maintains the AI / ML model configuration and / or AI / ML model based on the predefined relationship between the sector and the AI / ML model configuration and / or AI / ML model, UE2 (200-2) may monitor changes in the sector and maintain / change the AI / ML model configuration and / or AI / ML model accordingly. The BS (300) may indicate a sector identifier or indicator (or index) to help UE2 (200-2) select the AI / ML model configuration and / or AI / ML model. The BS (300) may have within its coverage some UEs with the same AI / ML model and AI / ML model configuration (e.g., UE2 (200-2) and UE4 (200-4)), some other UEs with the same AI / ML model and different AI / ML model configurations (e.g., UE1 (200-1) and UE3 (200-3)), and some other UEs with different AI / ML models (e.g., UE1 (200-1) and UE2 (200-3).

[0242] according to Fig.15a or the embodiment shown in 15b, when the UE (200) (for example, as described above in conjunction with Figure 25) from one sector to another, the UE (200) may select another AI / ML model configuration and / or another AI / ML model. The UE (200) may select another AI / ML model configuration and / or another AI / ML model based on information from the BS (300) (e.g., as described above in conjunction with Figure 3 6) to select another AI / ML model configuration and / or another AI / ML model. The BS (300) may indicate another AI / ML model configuration and / or another AI / ML model to help the UE (200) select the AI / ML model configuration and / or the AI / ML model. The BS (300) may indicate another AI / ML model configuration and / or another AI / ML model by using an indicator or identifier (or index) of another AI / ML model configuration and / or another AI / ML model. The BS (300) may select another new AI / ML model configuration and / or another AI / ML model based on the movement of the UE from one sector to another within the coverage of the BS. The BS (300) may use the AI / ML model to detect the movement of the UE from one sector to another, for predicting the UE position (using a positioning module on the BS) based on the speed and / or trajectory information of the UE.

[0243] according to Fig.16 In the embodiment shown, when the UE (for example, as described above in conjunction with Figure 2 5) switches from a first BS (BS1 (300-1)) to a second BS (BS2 (300-2)) (for example, BS1 or BS2 may refer to the above combined Figure 36), BS1 (300-1) may share the UE's AI / ML model information (1600) with BS2 (300-2) so that the UE does not synchronize one or more AI / ML models when switching. The UE AI / ML model information (1600) may include one or more of the following: an AI / ML model identifier or indicator (or index), an AI / ML model configuration identifier or indicator (or index), an AI / ML series identifier or indicator (or index), an AI / ML model configuration, UE capabilities, an activation / deactivation status of one or more AI / ML models and / or one or more AI / ML model configurations, and / or current version information of one or more AI / ML models. Based on the received AI / ML model information (1600), BS2 (300-2) may compare it with its stored AI / ML model information (1610) and determine whether it has one or more AI / ML models available (1620) so that it can serve the UE without any undue delay. If BS2 (300-2) determines that one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series supported by the UE are available and the version is the latest (1620), BS2 (300-2) may send a confirmation message to BS1 (300-1), indicating that the available one or more AI / ML models, one or more AI / ML model configurations, and one or more AI / ML model series are and the version is the latest. If BS2 (300-2) determines that one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series supported by the UE are not available and / or the version is not the latest (1620), BS2 (300-2) may send a message containing information (1630) identifying one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series to be downloaded and / or updated by BS2 (300-2) from BS1 (300-1). Based on the received information identifying the one or more AI / ML models to be downloaded and / or updated by BS2 (300-2), BS1 (300-1) may send a message containing details of the one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families download and / or update information (1640). The details may include a URL for downloading and / or updating the one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model families.Based on the received details of the one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series download and / or update information, BS2 (300-2) downloads / updates the one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series. If BS2 (300-2) successfully downloads or updates the one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series, BS2 (300-2) may send a confirmation message (1650) to BS1 (300-1) indicating that the one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series are available and / or the version is the latest. In the event that BS2 (300-2) fails to successfully download or update one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series, BS2 (300-2) may send a failure message (1650) to BS1 (300-1) indicating that one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series have not been successfully downloaded and / or updated. In the event that BS2 (300-2) fails to successfully download and / or update a subset of one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series, BS2 (300-2) may send a failure message (1650) to BS1 (300-1) indicating that one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series have not been successfully downloaded and / or updated.

[0244] according to Fig.16a In the alternative embodiment shown, when the UE (for example, as described above in conjunction with Figure 2 5) switches from a first BS (BS1 (300-1)) to a second BS (BS2 (300-2)) (for example, BS1 or BS2 may refer to the above combined Figure 36), BS1 (300-1) can share the UE's AI / ML model information (1600a) with BS2 (300-2) so that the UE does not have to synchronize one or more AI / ML models when switching. The UE AI / ML model information (1600) may include one or more of the following: an AI / ML model identifier or indicator (or index), an AI / ML model configuration identifier or indicator (or index), an AI / ML series identifier or indicator (or index), an AI / ML model configuration, UE capabilities, one or more AI / ML models and / or one or more AI / ML model configurations activation / deactivation status, and / or one or more AI / ML model current version information. Based on the received AI / ML model information (1600), BS2 (300-2) can compare it with its stored AI / ML model information (1610) and determine whether it has one or more AI / ML models and one or more AI / ML model configurations (1620a) available with it so that it can serve the UE without any delay. If BS2 (300-2) determines that one or more AI / ML models, one or more AI / ML model configurations and / or one or more AI / ML model series supported by the UE are available and / or the version is the latest, BS2 (300-2) may send a confirmation message to BS1 (300-1) indicating that one or more AI / ML models, one or more AI / ML model configurations and / or one or more AI / ML model series are available and the version is the latest. If BS2 (300-2) determines that one or more AI / ML models, one or more AI / ML model configurations and / or one or more AI / ML model series supported by the UE are unavailable and / or the version is not the latest (1620a), BS2 (300-2) may download / update the identified one or more AI / ML models, one or more AI / ML model configurations and / or one or more AI / ML model series from a predefined location (e.g., BS3 (300-3), a core network entity (1660a), or an AI / ML model server (1670a) (the AI / ML model server may refer to a central server for storing AI / ML models used in the network)) without interacting with BS1 (300-1) for downloading / updating. BS2 (300-2) may send information (1630a) identifying the AI / ML model to be downloaded or updated to a predefined location storing the AI / ML model or AI / ML model configuration, such as BS3 (300-3), a core network entity (1660a), or an AI / ML model server (1670a).In response to the information (1630a) identifying the AI / ML model to be downloaded or updated, BS3 (300-3), the core network entity (1660a), or the AI / ML model server (1670a) sends AI / ML model download or update information (1640a) to BS2 (300-2). If BS2 (300-2) successfully downloads or updates one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series, BS2 (300-2) may send a confirmation message (1650a) to BS1 (300-1) indicating that one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series are available and / or the version is the latest. In the event that BS2 (300-2) fails to successfully download or update one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series, BS2 (300-2) may send a failure message (1650a) to BS1 (300-1) indicating that one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series have not been successfully downloaded or updated. In an embodiment, BS2 (300-2) may generate a URL based on the one or more AI / ML model information (1600a) received from BS1 (300-1) for downloading / updating one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series. If BS2 (300-2) successfully downloads or updates one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series, BS2 (300-2) may send a confirmation message (1650a) to BS1 (300-1) indicating that the one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series are available and / or the version is the latest. In the event that BS2 (300-2) cannot successfully download or update one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series, BS2 (300-2) may send a failure message (1650a) to BS1 (300-1) indicating that the one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series were not successfully downloaded or updated.In the event that BS2 (300-2) fails to successfully download and / or update a subset of one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series, BS2 (300-2) may send a failure message (1650a) to BS1 indicating that the AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series were not successfully downloaded and / or updated. Thereafter, based on the received information identifying the one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series that BS2 (300-2) failed to successfully download and / or update, BS1 (300-1) may send a message containing details of the download and / or update information of the one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series. The details may include a URL for downloading and / or updating the one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series. Based on the received details of the received one or more AI / ML models, one or more AI / ML model configurations and / or one or more AI / ML model series download and / or update information, BS2 (300-2) can download / update one or more AI / ML models, one or more AI / ML model configurations and / or one or more AI / ML model series.

[0245] according to Fig.17 In the embodiment shown, when the UE (for example, as described above in conjunction with Figure 2 5) switches from a first BS (BS1 (300-1)) to a second BS (BS2 (300-2)) (for example, BS1 or BS2 may refer to the above combined Figure 36), BS1 (300-1) may send BS2 (300-2) the AI / ML model information of the UE and details of downloading and / or updating information of one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series, so that the UE does not have to synchronize one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series when switching. The UE AI / ML model information may include one or more of the following: an AI / ML model identifier or indicator (or index), an AI / ML model configuration identifier or indicator (or index), an AI / ML series identifier or indicator (or index), an AI / ML model configuration, UE capabilities, activation / deactivation status of one or more AI / ML models and / or one or more AI / ML model configurations, and / or current version information of one or more AI / ML models. The details of the one or more AI / ML models, one or more AI / ML model configurations and / or one or more AI / ML model series download and / or update information may include a URL for downloading and / or updating the one or more AI / ML models, one or more AI / ML model configurations and / or one or more AI / ML model series. Based on the received AI / ML model information (1700), BS2 (300-2) may compare (1710) with the stored one or more AI / ML model information and determine whether it has available one or more AI / ML models, one or more AI / ML model configurations and / or one or more AI / ML model series so that it can serve the UE without any delay. If BS2 (300-2) determines that one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series supported by the UE are available and / or the version is the latest (1720), BS2 (300-2) may transmit a confirmation message (1730) indicating that one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series are available and / or the version is the latest to BS1 (300-1). If BS2 (300-2) determines that one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series supported by the UE are not available and / or the version is not the latest, BS2 (300-2) may download / update the AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series that are not available and / or the latest to BS2 (300-2).After BS2 (300-2) successfully downloads and / or updates one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series, BS2 (300-2) may send a confirmation message (1730) to BS1 (300-1) indicating that the one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series are available and / or the version is the latest. In the event that BS2 (300-2) cannot successfully download or update one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series, BS2 (300-2) may send a failure message (1730) to BS1 (300-1) indicating that the one or more AI / ML models, one or more AI / ML model configurations, and / or one or more AI / ML model series were not successfully downloaded and / or updated. In the event that BS2 (300-2) fails to successfully download and / or update one or more AI / ML models and a subset of one or more AI / ML model series, BS2 (300-2) may send a failure message (1730) to BS1 (300-1) indicating that the AI / ML models and / or a subset of one or more AI / ML model series were not successfully downloaded and / or updated. Switching based on AI / ML model performance monitoring

[0246] According to an embodiment, a UE (eg, as described above in conjunction with Figure 2 5) can be configured with multiple AI / ML models for the same functionality to cater to various types of UE operation situations, such as, for example, high or low mobility (speed) operation, low or high power operation, good or poor coverage operation, high or low interference operation. Figure 3 6) provides signaling for switching AI / ML models or AI / ML model configurations within an AI / ML model series to the UE. The BS may provide signaling via the RRC layer, the MAC layer, or the physical layer. Another criterion for switching may be the performance of the AI / ML model and / or the AI / ML model configuration. If the performance of the AI / ML model and / or the AI / ML model configuration is lower than a predetermined threshold, the BS may signal a switch to another AI / ML model and / or AI / ML model configuration in the same series. An identifier known to both the UE and the BS may be assigned to the AI / ML model and / or the AI / ML model configuration in the series. The signaling for switching the AI / ML model and / or the AI / ML model configuration may include an identifier for the AI / ML model and / or the AI / ML model configuration to be switched.

[0247] According to an embodiment, a UE (eg, as described above in conjunction with Figure 2 5) may be based on various types of UE operating conditions (such as, for example, high or low mobility (speed) operation, low or high power operation, good or poor coverage operation, high or low interference operation, line-of-sight (LOS) / non-line-of-sight (NLOS)) determination in the absence of a BS (e.g., as described above in conjunction with Figure 3 and / or the BS described in 6), switch one or more AI / ML models and / or AI / ML model configurations by itself. The UE can determine high or low mobility (speed) operation based on a comparison of the UE speed with a threshold speed. The UE can determine high or low power operation based on a comparison of the UE transmit power with a threshold transmit power or based on signaling received from the BS indicating the UE transmit power. The UE can determine good or poor coverage operation based on a comparison of the UE RSRP / RSRQ / SINR with a threshold RSRP / RRSQ / SINR. The UE can determine high or low interference operation based on a comparison of the UE ACK / NACK ratio or packet error rate with a threshold ACK / NACK ratio or packet error rate. The UE can determine line of sight (LOS) / non-line of sight (NLOS) based on a comparison of the power deviation of the strongest signal path and this first-arrival signal path with a threshold. In the case of joint AI / ML operation between the UE and the BS, after switching the AI / ML model and / or AI / ML model configuration, the UE can provide the BS with an identifier of the selected AI / ML model and / or AI / ML model configuration via uplink signaling. The associated uplink signaling may be provided via the RRC layer or the physical layer.

[0248] According to an embodiment, a BS (e.g., as described above in conjunction with Figure 3 and / or the BS described in FIG. 6) may be based on a UE (eg, as described above in conjunction with Figure 2 5) speed and / or trajectory information to predict situations such as high / low UE mobility. The BS can predict high / low UE mobility based on the traffic conditions in the BS cell coverage at the predicted future location determined based on the speed and / or trajectory information of the UE (using the positioning module at the BS). The BS can determine the traffic conditions using real-time traffic maps such as Google Maps and the average speed of other UEs at the predicted location. Based on the prediction, the BS can transmit an AI / ML model and / or AI / ML model configuration switching signaling to the UE. The UE can receive the AI / ML model and / or AI / ML model configuration switching signaling and switch to the indicated AI / ML model and / or AI / ML model configuration. The AI / ML model and / or AI / ML model configuration switching signaling may include an AI / ML model and / or AI / ML model configuration identifier or a bit indicating an AI / ML model or AI / ML model configuration.

[0249] According to an embodiment, a BS (e.g., as described above in conjunction with Figure 3 and / or the BS described in FIG. 6) may be based on a UE (eg, as described above in conjunction with Figure 2 5) speed and / or trajectory information to predict situations such as line of sight (LOS) / non-line of sight (NLOS). The BS may predict non-line of sight (NLOS) situations based on a geographic map and / or structures such as, for example, buildings, forests, or mountains. In the cell coverage of the BS at a predicted future location determined based on the speed and / or trajectory information of the UE (using a positioning module at the BS), the BS may transmit an AI / ML model and / or AI / ML model configuration switching signaling to the UE. The UE may receive the AI / ML model and / or AI / ML model configuration switching signaling and may switch to the indicated AI / ML model and / or AI / ML model configuration accordingly. The AI / ML model switching signaling may include an AI / ML model and / or AI / ML model configuration identifier or a bit indicating an AI / ML model and / or AI / ML model configuration.

[0250] According to an embodiment, a BS (e.g., as described above in conjunction with Figure 3 and / or the BS described in FIG. 6) may be based on a UE (eg, as described above in conjunction with Figure 2 5) speed and / or trajectory information to predict situations such as high / low interference. The BS can estimate the interference situation based on the number of other UEs at the predicted future location determined based on the speed and / or trajectory information of the UE (using the positioning module at the BS). Based on the high / low interference prediction, the BS can transmit AI / ML model and / or AI / ML model configuration switching signaling to the UE. The UE can receive the AI / ML model and / or AI / ML model configuration switching signaling, and can switch to the indicated AI / ML model and / or AI / ML model configuration accordingly. The AI / ML model switching signaling may include an AI / ML model and / or AI / ML model configuration identifier or a bit indicating the AI / ML model and / or AI / ML model configuration.

[0251] According to an embodiment, a BS (e.g., as described above in conjunction with Figure 3 and / or the BS described in FIG. 6) may be based on a UE (eg, as described above in conjunction with Figure 25) speed and / or trajectory information to predict situations such as good / bad coverage. The BS can predict good / bad coverage based on the historical signal strength received at other UEs in the cell coverage of the BS at a predicted future location determined based on the speed and / or trajectory information of the UE (using a positioning module at the BS), and the BS can transmit an AI / ML model or AI / ML model configuration switching signaling to the UE. The UE can receive the AI / ML model and / or AI / ML model configuration switching signaling, and can switch to the indicated AI / ML model and / or AI / ML model configuration accordingly. The AI / ML model switching signaling may include an AI / ML model and / or AI / ML model configuration identifier or one or more bits indicating an AI / ML model and / or AI / ML model configuration.

[0252] The UE trajectory information may include a direction vector, an indicator of linear or angular movement, a direction of movement such as, for example, north, east, south, west (or a deflection such as, for example, northeast, southwest, northwest, southeast), a direction of movement at an angle to a reference direction such as north, or trajectory information known from a map (e.g., Google Maps or Apple Maps). The UE trajectory information may be associated with the current position of the UE expressed in longitude / latitude or a reference position relative to the BS location. AI / ML model activation / deactivation

[0253] According to an embodiment, a UE (eg, as described above in conjunction with Figure 2 5) may be configured with multiple AI / ML models for the same functionality to cater to various types of UE operation situations, such as, for example, high or low mobility (speed) operation, low or high power operation, good or poor coverage operation, and / or high or low interference operation. Figure 3 6) provides signaling for activating / deactivating the AI / ML model and / or the AI / ML model configuration to the UE. The BS may provide the signaling via the RRC, MAC, or physical layer. Another criterion for activation / deactivation may be the performance of the AI / ML model and / or the AI / ML model configuration. If the performance of the AI / ML model and / or the AI / ML model configuration is higher / lower than a predetermined threshold, the BS may signal the activation / deactivation of the AI / ML model. An identifier known to both the UE and the BS may be assigned to the AI / ML model and / or the AI / ML model configuration in the AI / ML model family. The signaling for activating / deactivating the AI / ML model or the AI / ML model configuration may include an identifier of the AI / ML model and / or the AI / ML model configuration.

[0254] According to an embodiment, by BS (e.g., as described above in conjunction with Figure 3and / or the BS described in FIG. 6) and transmitted by the UE (eg, as described above in conjunction with Figure 2 The activation / deactivation signaling for the AI / ML model and / or AI / ML model configuration received by the UE described in and / or Figure 5 may include a bit pattern corresponding to multiple AI / ML models and / or AI / ML model configurations, in which a single bit may correspond to the AI / ML model and / or AI / ML model configuration. Retraining based on AI / ML model performance monitoring.

[0255] According to an embodiment, retraining for one or more AI / ML models on the UE side may be used to solve performance issues, for periodic retraining, for AI / ML model updates, and / or for synchronization of UE / BS and / or AI / ML models. Figure 3 and / or the BS described in FIG. 6) may be a UE (eg, the above combined Figure 2 5) configures one or more predefined resources (time-frequency) for transmitting reference signals (e.g., CSI-RS, DMRS, SSB, or positioning reference signals) for retraining of the AI / ML model at the UE. The UE may receive the reference signal for retraining. The BS may configure the one or more predefined resources for the UE via RRC signaling and / or physical layer signaling. The one or more predefined resources may have a staggered configuration to capture the entire frequency range and / or a configured frequency range according to the UE capability.

[0256] According to an embodiment, a BS (e.g., as described above in conjunction with Figure 3 6) can download retraining related data to the UE (for example, as described above in conjunction with Figure 2 5) configures retraining data. The retraining data may include one or more of the following: simulated training data, compressed training data, AI / ML model parameters (such as, for example, one or more of the following: coefficients, weights, biases, the number of AI / ML tasks, one or more AI / ML task identifiers, classifiers (such as, for example, regression, KNN, vector machines, decision trees, and / or principal components), and / or cluster centroids in clustering), and / or may include hyperparameters (such as, for example, the number of layers, the number of hidden layers, the type of neural network (e.g., the number of nodes per layer of DNN, CNN, the number of activation units per layer, the choice of loss function, batch size, pooling size, the choice of activation function (e.g., Sigmoid, ReLU, or Tanh), the choice of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer)).

[0257] According to an embodiment, a BS (e.g., as described above in conjunction with Figure 3 6) side) can be used to solve performance issues, for regular retraining, for AI / ML model updates and / or UE / BS AI / ML model synchronization. The BS can send a message to the UE (for example, the above combined Figure 2 5) is configured to receive one or more predefined resources (time / frequency) for reference signals (SRS, DMRS, SSB or positioning reference signals) for retraining of the AI / ML model at the BS. The UE may transmit a reference signal (e.g., SRS) for retraining. The BS may configure the one or more predefined resources to the UE via RRC signaling and / or physical layer signaling. The one or more predefined resources may have an interlaced configuration to capture the entire frequency range and / or a configured frequency range according to the UE capability.

[0258] According to an embodiment, a BS (e.g., as described above in conjunction with Figure 3 and / or the BS described in Figure 6) can simulate retraining data for retraining one or more AI / ML models on the BS. Signaling for carrier aggregation

[0259] According to an embodiment, a UE (eg, as described above in conjunction with Figure 2 5) may be configured with carrier aggregation, and the UE may maintain separate configurations of the AI / ML models for separate carriers because the propagation environments of the UE on different carriers may have different characteristics. When the UE is configured for carrier aggregation, it may be configured by the BS (e.g., the above combined Figure 3and / or the BS described in 6) provides RRC signaling of the AI / ML model configuration of the indicated carrier to the UE according to the UE capability. For example, the RRC signaling transmitted by the BS and received by the UE may indicate to the UE one or more of the AI / ML model series 1 (AI / ML model 1 configuration 1) and / or the AI / ML model series 2 (AI / ML model 1 configuration 1) in the RRC configuration for component carrier 1 (CC1), and one or more of the AI / ML model series 1 (AI / ML model 1 configuration 2) and / or the AI / ML model series 2 (AI / ML model 1 configuration 2) in the RRC configuration for component carrier 2 (CC2). The UE may maintain the AI / ML model configuration for CC1 and / or CC2, and use the corresponding configuration while communicating through CC1 and / or CC2. The AI / ML model configuration may include one or more of the following: an AI / ML model version number, one or more identifiers or one or more indicators (or one or more indexes) of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of configurations of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a family of AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a set of AI / ML models, and / or one or more identifiers or one or more indicators (or one or more indexes) of a set of AI / ML models, details of the performance metrics to be used, details of the AI / ML Model parameters (such as, for example, one or more of the following: coefficients, weights, biases, number of AI / ML tasks, one or more AI / ML task identifiers (or one or more indexes), classifiers (such as, for example, regression, KNN, vector machine, decision tree, or principal components), cluster centroids in clustering), and / or hyperparameters (such as, for example, number of layers, number of hidden layers, type of neural network (e.g., DNN, CNN, RNN, or LSTM), number of nodes per layer, number of activation units per layer, choice of loss function, batch size, pooling size, choice of activation function (e.g., Sigmoid, ReLU, or Tanh), choice of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer)).

[0260] According to an embodiment, a UE configured with carrier aggregation (eg, as described above in conjunction with Figure 2 and / or the UE described in FIG. 5) may be provided by a BS (eg, as described above in conjunction with Figure 36) provides RRC signaling indicating the AI / ML model configuration. The RRC signaling transmitted by the BS and received by the UE may include one or more of the following: an AI / ML model version number, one or more identifiers or one or more indicators (or one or more indexes) of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of the configuration of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a series of AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a group of AI / ML models, and / or one or more identifiers or one or more indicators (or one or more indexes) of a group of AI / ML models, details of the performance metrics to be used, and A I / ML model parameters (such as, for example, one or more of the following: coefficients, weights, biases, number of AI / ML tasks, one or more AI / ML task identifiers (or one or more indexes), classifiers (such as, for example, regression, KNN, vector machine, decision tree, or principal components), cluster centroids in clustering), and / or hyperparameters (such as, for example, number of layers, number of hidden layers, type of neural network (e.g., DNN, CNN, RNN, or LSTM), number of nodes per layer, number of activation units per layer, choice of loss function, batch size, pooling size, choice of activation function (e.g., Sigmoid, ReLU, or Tanh), choice of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer)).

[0261] According to an embodiment, a UE configured with carrier aggregation (eg, as described above in conjunction with Figure 2 and / or the UE described in FIG. 5) may be provided by a BS (eg, as described above in conjunction with Figure 3and / or the BS described in Figure 6) provides signaling indicating the AI / ML model configuration. The signaling transmitted by the BS and received by the UE may include information of multiple carriers (or cell indexes) configured with a common AI / ML model configuration. For example: in a heterogeneous network deployment, a carrier associated with a microcell may be provided with a common first AI / ML model configuration, and a carrier associated with a macrocell may be provided with a common second AI / ML model configuration. Another example may include a dual-connection deployment, in which a carrier associated with a first base station may be provided with a common first AI / ML model configuration, and a carrier associated with a second base station may be provided with a common second AI / ML model configuration. The signaling may be provided via the RRC, physical layer, or MAC layer. The signaling may include one or more of the following: a cell index or an indicator of a cell index, a cell group indicator (or index) indicating a cell group configured for a common AI / ML model configuration, one or more AI / ML model version numbers, one or more identifiers or one or more indicators (or one or more indexes) of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of configurations of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a series of AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a group of AI / ML models, and / or one or more identifiers or one or more indicators (or one or more indexes) of a group of AI / ML models. multiple indices), details of the performance metric to be used, AI / ML model parameters (such as, for example, one or more of the following: coefficients, weights, biases, number of AI / ML tasks, one or more AI / ML task identifiers (or one or more indices), classifiers (such as, for example, regression, KNN, vector machine, decision tree, or principal components), cluster centroids in clustering), and / or hyperparameters (such as, for example, the number of layers, the number of hidden layers, the type of neural network (e.g., DNN, CNN, RNN, or LSTM), the number of nodes per layer, the number of activation units per layer, choice of loss function, batch size, pooling size, choice of activation function (e.g., Sigmoid, ReLU, or Tanh), choice of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer)).

[0262] According to an embodiment, the BS (for example, as described above in conjunction with Figure 3 and / or the BS described in FIG. 6) to a UE configured with carrier aggregation (eg, as described above in conjunction with Figure 25) provides a first signaling indicating an AI / ML model configuration. The first signaling transmitted by the BS and received by the UE may include information of multiple carriers (or cell indexes) configured with a common AI / ML model configuration. For example, the first signaling may refer to RRC signaling. Subsequently, the BS transmits and the UE receives a second signaling indicating multiple carriers (or cell indexes) to activate the use of the common AI / ML model configuration. Based on the second signaling, the UE activates the common AI / ML model configuration and configures the AI / ML engine (211) to process the transmission / reception of signals / channels corresponding to the multiple carriers. The second signaling may refer to physical layer or MAC layer signaling.

[0263] According to an embodiment, the BS (for example, as described above in conjunction with Figure 3 and / or the BS described in FIG. 6) to a UE configured with carrier aggregation (eg, as described above in conjunction with Figure 2 5) provides a first signaling indicating multiple AI / ML model configurations. The first signaling transmitted by the BS and received by the UE may include information of multiple carriers (or cell indexes) configured with multiple common AI / ML model configurations. For example, the first signaling may refer to RRC signaling. Subsequently, the BS transmits and the UE receives a second signaling indicating a first common AI / ML model configuration for activating the use of the first common AI / ML model configuration among multiple carriers (or cell indexes) and multiple common AI / ML model configurations. Based on the second signaling, the UE activates the first common AI / ML model configuration and configures the AI / ML engine (211) to process the transmission / reception of signals / channels corresponding to multiple carriers. The second signaling may refer to physical layer or MAC layer signaling.

[0264] according to Fig.24 In the embodiment shown, a UE configured with carrier aggregation (eg, as described above in conjunction with Figure 2 and / or the UE described in FIG. 5) from a base station (eg, as described above in conjunction with Figure 36) receives a first signaling including information of multiple carriers. The information of multiple carriers includes a carrier index and an association between the carrier index and a first AI / ML model or a first group of AI / ML models (2400). The first signaling may refer to RRC signaling. After receiving the first signaling, the UE configures itself with carrier aggregation and configures the AI / ML engine using a first AI / ML model or a first group of AI / ML models. The UE receives multiple downlink channels from a base station on multiple carriers. The first downlink channel among the multiple downlink channels includes a reference signal. The first downlink channel is received on a first carrier among the multiple carriers (2410). The UE uses a first AI / ML model or a first group of AI / ML models to predict channel state information of multiple carriers. Channel state information is predicted based on the received reference signal (2420). The UE uses a first AI / ML model or a first group of AI / ML models to predict the channel states of other carriers using a reference signal received on the first carrier. Since multiple carriers are transmitted from the same base station and received by the same UE, a first AI / ML model or a first group of AI / ML models can be used to exploit redundancy in channel characteristics across carriers. A reference signal is received only on the first carrier, or the reference signal received on the first carrier has a denser reference signal configuration than the reference signal configuration received on other carriers. The reference signal configuration refers to a reference signal pattern. Example reference signals may include CSI-RS, DMRS, or SSB signals. The first carrier may be configured as a carrier with a minimum cell index, or it may be a carrier predefined by a base station for predicting channel state information of multiple carriers. The UE decodes the downlink channels received on multiple carriers by using the predicted channel state information (2430). The UE may decode the downlink channels received on the first carrier using the estimated channel state information, and decode the downlink channels received on other carriers using the predicted channel state message, or the UE may decode the downlink channels received on all carriers using the predicted channel state information.

[0265] according to Fig.25 In the embodiment shown, a UE configured with carrier aggregation (eg, as described above in conjunction with Figure 2 and / or the UE described in FIG. 5) from a base station (eg, as described above in conjunction with Figure 3and / or the BS described in FIG. 6) receives a first signaling including information of multiple carriers. The information of multiple carriers includes a carrier index and an association between the carrier index and a corresponding AI / ML model configuration (2500). The corresponding AI / ML model configuration for each carrier can be defined based on the carrier frequency. For example, a carrier within the FR1 range may have an AI / ML model configuration different from that of a carrier within the FR2 range. Some carriers may have the same AI / ML model configuration. The description of the AI / ML model configuration can be identified from other parts of the present disclosure. The first signaling may refer to RRC signaling. After receiving the first signaling, the UE configures itself with carrier aggregation and configures the AI / ML engine for each carrier using the corresponding AI / ML model configuration. The UE receives multiple downlink channels from the base station on multiple carriers. The first downlink channel among the multiple downlink channels includes a reference signal. The first downlink channel is received on the first carrier among the multiple carriers (2510). The UE uses the corresponding AI / ML model configuration to predict the channel state information of the multiple carriers. The channel state information is predicted based on the received reference signal (2520). The UE uses a reference signal received on a first carrier to predict the channel state of other carriers using a corresponding AI / ML model configuration. Since multiple carriers are transmitted from the same base station and received by the same UE, the corresponding AI / ML model configuration can be used to exploit redundancy in channel characteristics across carriers. The reference signal is received only on the first carrier, or the reference signal received on the first carrier has a denser reference signal configuration than the reference signal configuration received on other carriers. The reference signal configuration refers to a reference signal pattern. Example reference signals may include CSI-RS, DMRS, or SSB signals. The first carrier may be configured as a carrier with a minimum cell index, or it may be a carrier predefined by a base station for predicting channel state information of multiple carriers. The UE decodes the downlink channels received on multiple carriers by using the predicted channel state information (2530). The UE may decode the downlink channel received on the first carrier using the estimated channel state information, and decode the downlink channels received on other carriers using the predicted channel state information, or the UE may decode the downlink channels received on all carriers using the predicted channel state information.

[0266] according to Fig.26 In the embodiment shown, the base station (for example, as described above in conjunction with Figure 3 6) by transmitting a first signaling including information of a plurality of carriers to a UE (eg, as described above in conjunction with Figure 2and / or the UE described in FIG. 5) configures carrier aggregation. The information of multiple carriers includes a carrier index and an association of the carrier index with a first AI / ML model or a first group of AI / ML models (2600). The first signaling may refer to RRC signaling. The first signaling is used by the UE to configure itself with carrier aggregation and configure its AI / ML engine using a first AI / ML model or a first group of AI / ML models. The base station transmits multiple downlink channels to the UE on multiple carriers. The first downlink channel among the multiple downlink channels includes a reference signal. The first downlink channel is transmitted on a first carrier among the multiple carriers (2610). The UE uses a first AI / ML model or a first group of AI / ML models to predict channel state information of multiple carriers. The channel state information is predicted based on the transmitted reference signal. The UE uses a first AI / ML model or a first group of AI / ML models to predict the channel states of other carriers using a reference signal transmitted on the first carrier. Since multiple carriers are transmitted from the same base station and received by the same UE, the first AI / ML model or the first group of AI / ML models can be used to exploit redundancy in channel characteristics across carriers. The reference signal is transmitted only on the first carrier, or the reference signal transmitted on the first carrier has a denser reference signal configuration than the reference signal configuration received on other carriers. The reference signal configuration refers to a reference signal pattern. Example reference signals may include CSI-RS, DMRS, or SSB signals. The first carrier may be configured as a carrier with a minimum cell index, or it may be a carrier predefined by the base station for predicting channel state information of multiple carriers.

[0267] according to Fig. 27 In the embodiment shown, the base station (for example, as described above in conjunction with Figure 3 6) by transmitting a first signaling including information of a plurality of carriers to a UE (eg, as described above in conjunction with Figure 2and / or the UE described in FIG. 5) configures carrier aggregation. The information of multiple carriers includes a carrier index and an association between the carrier index and the corresponding AI / ML model configuration (2700). The corresponding AI / ML model configuration for each carrier can be defined based on the carrier frequency. For example, a carrier within the FR1 range may have an AI / ML model configuration different from a carrier within the FR2 range. Some carriers may have the same AI / ML model configuration. The description of the AI / ML model configuration can be identified from other parts of the present disclosure. The first signaling may refer to RRC signaling. The UE uses the first signaling to configure itself with carrier aggregation and configures its AI / ML engine using the corresponding AI / ML model configuration. The base station transmits multiple downlink channels to the UE on multiple carriers. The first downlink channel among the multiple downlink channels includes a reference signal. The first downlink channel is transmitted on the first carrier among the multiple carriers (2710). The UE uses the corresponding AI / ML model configuration to predict the channel state information of each carrier in the multiple carriers. The channel state information is predicted based on the transmitted reference signal. The UE uses the reference signal transmitted on the first carrier to predict the channel state of other carriers using the corresponding AI / ML model configuration. Since multiple carriers are transmitted from the same base station and received by the same UE, the corresponding AI / ML model configuration can be used to exploit redundancy in channel characteristics across carriers. The reference signal is transmitted only on the first carrier, or the reference signal transmitted on the first carrier has a denser reference signal configuration than the reference signal configuration received on other carriers. The reference signal configuration refers to a reference signal pattern. Example reference signals may include CSI-RS, DMRS, or SSB signals. The first carrier may be configured as a carrier with a minimum cell index, or it may be a carrier predefined by the base station for predicting channel state information of multiple carriers. Signaling for Bandwidth Part (BWP)

[0268] According to an embodiment, a UE configured with multiple BWPs (eg, as described above in conjunction with Figure 2 5) can maintain separate configurations of the AI / ML models for the BWPs, because the propagation environments of the UEs on different BWPs may have different characteristics. Figure 36) provides RRC signaling indicating the AI / ML model configuration of one or more BWPs to the UE configured for the BWP according to the capabilities of the UE. For example, the RRC signaling may indicate to the UE one or more of the AI / ML model series 1 (AI / ML model 1 configuration 1) and / or the AI / ML model series 2 (AI / ML model 1 configuration 1) in the RRC configuration for BWP1, and one or more of the AI / ML model series 1 (AI / ML model 1 configuration 2) and / or the AI / ML model series 2 (AI / ML model 1 configuration 2) in the RRC configuration for BWP2. The UE may maintain the AI / ML model configuration for both BWP1 and BWP2, and use the corresponding configuration while communicating through BWP1 and / or BWP2. The AI / ML model configuration may include one or more of the following: an AI / ML model version number, one or more identifiers or one or more indicators (or one or more indexes) of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of configurations of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a family of AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a set of AI / ML models, and / or one or more identifiers or one or more indicators (or one or more indexes) of a set of AI / ML models, details of the performance metrics to be used, details of the AI / ML Model parameters (such as, for example, one or more of the following: coefficients, weights, biases, number of AI / ML tasks, one or more AI / ML task identifiers (or one or more indexes), classifiers (such as, for example, regression, KNN, vector machine, decision tree, or principal components), cluster centroids in clustering), and / or hyperparameters (such as, for example, number of layers, number of hidden layers, type of neural network (e.g., DNN, CNN, RNN, or LSTM), number of nodes per layer, number of activation units per layer, choice of loss function, batch size, pooling size, choice of activation function (e.g., Sigmoid, ReLU, or Tanh), choice of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer)).

[0269] According to an embodiment, a UE configured with multiple BWPs (eg, as described above in conjunction with Figure 2 and / or the UE described in 5) can be provided by a BS (for example, Figure 3and / or the BS described in 6) provides RRC signaling indicating the AI / ML model configuration. The RRC signaling may include one or more of the following: an AI / ML model version number, one or more identifiers or one or more indicators (or one or more indexes) of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of the configuration of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a series of AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a group of AI / ML models, and / or one or more identifiers or one or more indicators (or one or more indexes) of a group of AI / ML models, details of the performance metrics to be used, and the AI / ML model Parameters (such as, for example, one or more of the following: coefficients, weights, biases, number of AI / ML tasks, one or more AI / ML task identifiers (or one or more indexes), classifiers (such as, for example, regression, KNN, vector machine, decision tree, or principal components), cluster centroids in clustering), and / or hyperparameters (such as, for example, number of layers, number of hidden layers, type of neural network (e.g., DNN, CNN, RNN, or LSTM), number of nodes per layer, number of activation units per layer, choice of loss function, batch size, pooling size, choice of activation function (e.g., Sigmoid, ReLU, or Tanh), choice of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer)).

[0270] According to an embodiment, a UE configured with multiple BWPs (eg, as described above in conjunction with Figure 2 and / or the UE described in FIG. 5) may be provided by a BS (eg, as described above in conjunction with Figure 3 and / or the BS described in 6) provides a first signaling indicating multiple AI / ML model configurations each corresponding to a BWP. The first signaling transmitted by the BS and received by the UE may include information of multiple BWPs (or BWP IDs), each BWP being configured with an AI / ML model configuration. For example, the first signaling may refer to RRC signaling. Subsequently, the BS transmits and the UE receives a second signaling indicating a BWP (or BWP ID) among the BWPs and / or a first AI / ML model configuration among multiple AI / ML model configurations to activate the use of the first AI / ML model configuration. Based on the second signaling, the UE activates the first / ML model configuration and configures the AI / ML engine (211) to process the transmission / reception of the signal / channel corresponding to the indicated BWP. The second signaling may refer to physical layer or MAC layer signaling. Signaling for dual connectivity

[0271] according to Fig.18In the embodiment shown, the UE (for example, as described above in conjunction with Figure 2 5) can be configured in a dual connectivity mode, whereby it can simultaneously transmit to / receive from a first base station (BS1 (300-1)) and transmit to / receive from a second BS (BS2 (300-2)). (BS2 (300-2)) receives (for example, BS1 or BS2 may refer to the above combined Figure 3and / or the BS described in FIG. 6 ). In dual connectivity mode, given that the propagation environment of the UE (200) and BS1 (300-1) may be different from the propagation environment of the UE (200) and BS2 (300-2), the UE (200) may maintain separate configurations of the AI / ML model while communicating with the two base stations. Depending on the UE capability, an RRC message (or signaling) indicating the AI / ML model configurations of BS1 (300-1) and BS2 (300-2) may be provided to the UE (200) configured for dual connectivity. For example, the RRC message (or signaling) may indicate to the UE (200) one or more of the AI / ML model series 1 (AI / ML model 1 configuration 1) and / or the AI / ML model series 2 (AI / ML model 1 configuration 1) in the RRC configuration for BS1 (300-1), and / or one or more of the AI / ML model series 1 (AI / ML model 1 configuration 2) and / or the AI / ML model series 3 (AI / ML model 1 configuration 1) in the RRC configuration for BS2 (300-1). The UE (200) may maintain the AI / ML model configurations for both BS1 (300-1) and BS2 (300-2), and use the corresponding configurations while communicating with BS1 (300-1) and / or BS2 (300-2). The AI / ML model configuration may include one or more of the following: an AI / ML model version number, one or more identifiers or one or more indicators (or one or more indexes) of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of configurations of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a family of AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a set of AI / ML models, and / or one or more identifiers or one or more indicators (or one or more indexes) of a set of AI / ML models, details of the performance metrics to be used, details of the AI / ML Model parameters (such as, for example, one or more of the following: coefficients, weights, biases, number of AI / ML tasks, one or more AI / ML task identifiers (or one or more indexes), classifiers (such as, for example, regression, KNN, vector machine, decision tree, or principal components), cluster centroids in clustering), and / or hyperparameters (such as, for example, number of layers, number of hidden layers, type of neural network (e.g., DNN, CNN, RNN, or LSTM), number of nodes per layer, number of activation units per layer, choice of loss function, batch size, pooling size, choice of activation function (e.g., Sigmoid, ReLU, or Tanh), choice of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer)).

[0272] According to an embodiment, a dual connectivity mode is configured so that it can simultaneously transmit to / receive from a first base station (BS1 (300-1)) and transmit to / receive from a second BS (BS2 (300-2)) (for example, BS1 or BS2 may refer to the above in conjunction with Figure 3 and / or the BS described in FIG. 6) of a UE (eg, as described above in conjunction with Figure 2 and / or the UE described in FIG. 5 ) receives a first signaling from BS1 (300-1) indicating a first plurality of AI / ML model configurations corresponding to BS1 (300-1) and a second plurality of AI / ML model configurations corresponding to BS2 (300-2). For example, the first signaling may refer to RRC signaling. Subsequently, BS1 (300-1) transmits and UE (200) receives a second signaling indicating a first AI / ML model configuration among the first plurality of AI / ML model configurations to activate the use of the first AI / ML model configuration. Based on the second signaling, the UE activates the first / ML model configuration and configures the AI / ML engine (211) to process the transmission / reception of signals / channels related to BS1 (300-1). The second signaling may refer to physical layer or MAC layer signaling. BS2 (300-2) transmits and UE (200) receives a third signaling indicating a second AI / ML model configuration among the second plurality of AI / ML model configurations to activate the use of the second AI / ML model configuration. Based on the third signaling, the UE activates the second AI / ML model configuration and configures the AI / ML engine (211) to process the transmission / reception of signals / channels related to BS2 (300-2). The third signaling may refer to physical layer or MAC layer signaling.

[0273] According to an embodiment, a dual connectivity mode is configured so that it can simultaneously transmit to / receive from a first base station (BS1 (300-1)) and transmit to / receive from a second BS (BS2 (300-2)) (for example, BS1 or BS2 may refer to the above in conjunction with Figure 3 and / or the BS described in FIG. 6) of a UE (eg, as described above in conjunction with Figure 2and / or the UE described in Figure 5) receives a first signaling from BS1 (300-1) indicating a first plurality of AI / ML model configurations corresponding to BS1 (300-1), and receives a second signaling from BS2 (300-2) indicating a second plurality of AI / ML model configurations corresponding to BS2 (300-2). For example, the first signaling and the second signaling may refer to RRC signaling. Subsequently, BS1 (300-1) transmits and UE (200) receives a third signaling indicating a first AI / ML model configuration among the first plurality of AI / ML model configurations to activate the use of the first AI / ML model configuration. Based on the third signaling, the UE activates the first / ML model configuration and configures the AI / ML engine (211) to process the transmission / reception of signals / channels related to BS1 (300-1). The third signaling may refer to physical layer or MAC layer signaling. BS2 (300-2) transmits and UE (200) receives fourth signaling indicating a second AI / ML model configuration among the second plurality of AI / ML model configurations to activate use of the second AI / ML model configuration. Based on the fourth signaling, the UE activates the second AI / ML model configuration and configures the AI / ML engine (211) to process transmission / reception of signals / channels related to BS2 (300-2). The fourth signaling may refer to physical layer or MAC layer signaling.

[0274] according to Fig.18a In the embodiment shown, BS1 (300-1) can provide a UE (for example, as described above in conjunction with Figure 2 5) provides an indication BS1 (300-1) and BS2 (300-2) (for example, BS1 or BS2 may refer to the above combined Figure 3 6) and / or one or more AI / ML models and / or one or more AI / ML model configurations of the BS. For example, the RRC message (1800a) (or signaling) may be provided by BS1 (300-1) to send a signal to the UE. (200) indicates one or more of AI / ML model series 1 (AI / ML model 1 configuration 1) and / or AI / ML model series 2 (AI / ML model 1 configuration 1) in the RRC configuration for BS1 (300-1), and one or more of AI / ML model series 1 (AI / ML model 1 configuration 2) and / or AI / ML model series 3 (AI / ML model 1 configuration 1) in the RRC configuration for BS2 (300-2). UE (200) can maintain AI / ML model configurations for both BS1 (300-1) and BS2 (300-2) and use the corresponding configurations while communicating with BS1 (300-1) and / or BS2 (300-2). The RRC message (1800a) (or signaling) may include one or more messages sent by BS1 (300-1) to the UE (200).

[0275] according to Fig.18b In the embodiment shown, BS2 (300-2) can provide the UE with a signal according to the UE capability (for example, as described above in conjunction with Figure 2 5) provides an indication BS1 (300-1) and / or BS2 (300-2) (for example, BS1 or BS2 may refer to the above combined Figure 3 6) and / or one or more AI / ML models and / or one or more AI / ML model configurations of the BS. For example, the RRC message (1800b) (or signaling) may be provided by BS2 (300-2) to provide the UE with the information. (200) indicates one or more of AI / ML model series 1 (AI / ML model 1 configuration 1) and / or AI / ML model series 2 (AI / ML model 1 configuration 1) in the RRC configuration for BS1 (300-1), and one or more of AI / ML model series 1 (AI / ML model 1 configuration 2) and / or AI / ML model series 3 (AI / ML model 1 configuration 1) in the RRC configuration for BS2 (300-2). UE (200) can maintain AI / ML model configurations for both BS1 (300-1) and BS2 (300-2) and use the corresponding configurations while communicating with BS1 (300-1) and / or BS2 (300-2). The RRC message (1800b) (or signaling) may include one or more messages sent by BS2 (300-2) to the UE (200).

[0276] according to Fig.18b In the embodiment shown, BS1 (300-1) and / or BS2 (300-2) (for example, BS1 or BS2 may refer to the above-mentioned Figure 3 and / or the BS described in FIG. 6) to the UE (for example, as described above in conjunction with the BS described in FIG. 6) according to the UE capability. Figure 25) provides an RRC message (1800c) (or signaling) indicating the AI / ML model configuration of BS1 (300-1) and / or BS2 (300-2). For example, BS1 (300-1) may provide an RRC message (1800c) (or signaling) to indicate to the UE (200) one or more of AI / ML model series 1 (AI / ML model 1 configuration 1) and / or AI / ML model series 2 (AI / ML model 1 configuration 1) in the RRC configuration of BS1 (300-1), and / or BS2 may provide an RRC message (1800c) (or signaling) to indicate to the UE (200) one or more of AI / ML model series 1 (AI / ML model 1 configuration 1) and / or AI / ML model series 2 (AI / ML model 1 configuration 1) in the RRC configuration of BS1 (300-1). (300-2) provides an RRC message (1810c) (or signaling) to indicate to the UE (200) one or more of the UE AI / ML model series 1 (AI / ML model 1 configuration 2) and / or the AI / ML model series 3 (AI / ML model 1 configuration 1) in the RRC configuration for BS2 (300-2). The UE (200) can maintain the AI / ML model configuration for BS1 (300-1) and / or BS2 (300-2) and use the corresponding configuration while communicating with BS1 (300-1) and / or BS2 (300-2). The RRC message (1800c, 1810c) (or signaling) may include one or more messages sent by BS1 (300-1) and / or BS2 (300-2) to the UE (200).

[0277] According to an embodiment, a dual connection (such as Fig.18 , 18a , 18b or 18c) (For example, as described above in conjunction with Figure 2and / or the UE described in Figure 5) provides an RRC message (1800a, 1800b, 1800c, 1810c) (or signaling) indicating the AI / ML model configuration. The RRC message (1800a, 1800b, 1800c, 1810c) (or signaling) may include one or more of the following: an AI / ML model version number, one or more identifiers or one or more indicators (or one or more indexes) of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of the configuration of one or more AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a series of AI / ML models, one or more identifiers or one or more indicators (or one or more indexes) of a group of AI / ML models, and / or one or more identifiers or one or more indicators (or one or more indexes) of a group of AI / ML models, to be used Details of performance metrics, AI / ML model parameters (such as, for example, one or more of the following: coefficients, weights, biases, number of AI / ML tasks, one or more AI / ML task identifiers (or one or more indexes), classifiers (such as, for example, regression, KNN, vector machine, decision tree, or principal components), cluster centroids in clustering), and / or hyperparameters (such as, for example, number of layers, number of hidden layers, type of neural network (e.g., DNN, CNN, RNN, or LSTM), number of nodes per layer, number of activation units per layer, choice of loss function, batch size, pooling size, choice of activation function (e.g., Sigmoid, ReLU, or Tanh), choice of optimization algorithm (e.g., gradient descent, stochastic gradient descent, or Adam optimizer)). AI / ML-based early switching commands

[0278] according to Fig.19 In the illustrated embodiment, one or more AI / ML models may be implemented to predict early switching decisions. Fig.19 As shown, according to UE 1 (200-1) (for example, UE1, UE2 or UE3 may refer to the above combined Figure 2 and / or the speed and / or trajectory of the UE described in 5), the first base station (BS1 (300-1)) (for example, BS1, BS2 or BS3 may refer to the above combined Figure 36) can predict (using the positioning module at the BS) that UE1 (200-2) will switch to cell 3, and in advance, BS1 (300-1) can send a switching command to UE1 (200-1) in an RRC reconfiguration message, which includes one or more of the following: the target cell ID, the new C-RNTI and the security algorithm identifier of the third base station (BS3 (300-3)) for the selected security algorithm, and other fields of the switching command. BS1 (300-1) can implement an AI / ML model that can use UE speed and / or trajectory information to predict the switching decision. BS1 (300-1) may receive speed and / or trajectory information from a UE (including UE 1 (200-1)), or BS1 (300-1) itself may estimate speed and / or trajectory information based on information received from a UE (e.g., UE1 (200-1), UE2 (200-2), or UE3 (200-3)), such as, for example, a reference signal (e.g., DMRS or SRS) or arrival / departure angle. BS1 (300-1) may then provide the received / estimated speed and / or trajectory information to the AI / ML model of the positioning module for predicting future UE positions. If the predicted UE position indicates that the UE is in a cell different from cell 1 of BS1 (300-1), BS1 (300-1) may generate a handover command and send it to the UE. For example, BS1 (300-1) may predict that the position of UE3 (200-3) will be in cell 2 after time t1 seconds. BS1 (300-1) may generate and send a handover command before t1-x seconds, where x may be determined so that UE3 (200-3) does not need to perform neighbor cell measurements and send measurement reports to BS1 (300-1), thereby saving UE3 (200-3) resources and power as well as network resources for sending measurement reports. (200-3) receives a handover command while performing neighbor cell measurements, then UE3 (200-3) may ignore sending a measurement report. In the case of UE2 (200-2), BS1 (300-1) may predict that the location of UE2 (200-2) is in cell 1 even after t1 seconds. BS1 (300-1) may avoid sending a handover command and continue to monitor UE2 (200-2) speed and / or trajectory to predict UE2 (200-2) location in order to make a handover decision.

[0279] Further, according to Fig.19 In the embodiment shown, when sending a signal to a UE (for example, as described above in conjunction with Figure 2 5) sends a handover command (determined based on UE speed and / or trajectory information), BS1 (300-1) (e.g., BS1 (300-1), BS2 (300-2), or BS3 (300-3) may refer to the above combined Figure 3 and / or the BS described in 6) can prepare the second base station (BS2 (300-2)) and / or BS3 (300-2) for handover by sending a handover request (e.g., a handover request from UE1 (200-1) to BS3 (300-3) and / or a handover request from UE3 (300-3) to BS2 (300-2)) containing one or more of UE history information, UE context information, GUAMI, target cell ID and / or PDU session list. After receiving confirmation from BS2 (300-2) and / or BS3 (300-3), BS1 (300-1) can send a handover command to UE3 (200-3) and / or UE1 (200-1). BS1 (300-1) can also send a handover command to BS2 (300-2) and / or BS3 (300-3) sends an SN status transmission message to transmit the uplink and downlink PDCP SN and super frame number (HFN) status of UE 3 (300-3) and / or UE1 (200-1). BS1 (300-1) can start buffering DL data from UPF and forward it to BS2 (300-2) and / or BS3 (300-3). Fig.19 In one aspect shown, after a handover request from BS1 (300-1) is accepted by BS2 (300-2) and / or BS3 (300-3), BS1 (300-1) may send details of a contention-free random access (RACH) preamble to UE1 (200-1) and / or UE3 (200-3) to perform a RACH procedure with the corresponding BS3 (300-3) and / or BS2 (300-2). BS1 (300-1) may receive a RACH preamble from BS3 (300-3) and / or BS2 (300-2) receive RACH preambles, and / or it can generate these RACH preambles based on information received from BS3 (300-3) and / or BS2 (300-2). UE1 (200-1) and / or UE3 (300-3) can use the preambles received from BS1 (300-1) and perform corresponding RACH procedures.

[0280] according to Fig. 20 In the embodiment shown, one or more AI / ML models may be implemented to (200) (e.g., as described above in conjunction with Figure 2and / or the speed and / or trajectory of the UE described in FIG. 5) to predict the early handover decision. The first base station (BS1 (300-1)) (for example, BS1 (300-1), BS2 (300-2) or BS3 (300-3) may refer to the above combined Figure 3 and / or the BS described in Figure 6) can predict that the first user equipment (UE (200)) is about to switch to BS2 (300-2), and in advance, BS1 (300-1) can send a switching command in an RRC reconfiguration message, which RRC reconfiguration message includes one or more of the following: the target cell ID, the new C-RNTI and the security algorithm identifier of BS2 (300-2) for the selected security algorithm, and other fields of the switching command. BS1 (300-1) can implement an AI / ML model that can use UE speed and / or trajectory information to predict the switching decision. BS1 (300-1) can receive the speed and / or trajectory information from the UE (200), or BS1 (300-1) can receive the speed and / or trajectory information from the core network entity or the location server. Fig. 20 In the present invention, UE (200) may refer to a smartphone, a tablet, a mobile device with a cellular connection, or a car / drone / UVA / train / ship / vehicle with a cellular connection and a map application operating on the UE (200). The UE (200) may allow secure access to its map application to BS1 (300-1), to a core network entity connected to BS1 (300-1), or to a location server connected to a core network entity connected to BS1 (300-1). BS1 (300-1) may receive UE speed and / or trajectory information from the UE's map application, from a core network entity connected to BS1 (300-1), or from a location server connected to a core network entity connected to BS1 (300-2). Secure access to the map application may include encryption-based security, limited access to only relevant details such as UE speed and / or trajectory, and / or any security or access control mechanism for protecting user privacy. BS1 (300-1) may provide the received speed and / or trajectory information to the AI / ML model to predict the future UE location, and if the predicted UE location indicates that UE (200) is in a cell different from cell 1 of BS1 (300-1), BS1 (300-1) may generate a handover command and send it to UE (200). For example, BS1 (300-1) can predict that the location of UE (200) will be in cell 2 of BS2 (300-2) after time t1 seconds. BS1 (300-1) can generate a handover command and send the handover command before t1-x seconds, where x can be determined so that UE (200) does not need to perform neighbor cell measurements and send measurement reports to BS1 (300-1), thereby saving UE resources and power as well as network resources for sending measurement reports. If UE (200) If a handover command is received while performing neighbor cell measurements, the UE (200) may ignore sending a measurement report.

[0281] according to Fig.21 In the embodiment shown, the UE (200) (for example, as described above in conjunction with Figure 2 5) can be connected to BS1 (300-1) (e.g., BS1 (300-1), BS2 (300-2), or BS3 (300-3) at the beginning of the UE journey or after a time offset from the beginning of the UE journey) as described above in conjunction with Figure 3 UE (200) can share UE information (2110) details including the destination address / location of the UE, UE speed and / or trajectory information with BS1. (300-1) shares the UE information (2110), and BS1 (300-1) can forward the received UE information (or modified UE information) (2120) to BS2 (300-2). BS1 (300-1) can determine the BS2 to which to forward the UE information (2120) based on the identity of the next base station on the UE's route to its destination. (300-2). Similarly, BS2 (300-2) can forward the UE information (2130) to BS3 (300-3) so that it reaches all base stations on the route. BS1 (300-1) can forward the details to BS2 (300-2) during the handover preparation step. In a variation of this embodiment, BS1 (300-1) can receive the UE information (2110) from a map application server (e.g., a Google Maps server) of the UE's map application (2100) via an API.

[0282] according to Fig. 22 In the embodiment shown, the UE (200) (for example, as described above in conjunction with Figure 2 5) can be connected to and BS1 (300-1) (e.g., BS1 (300-1), BS2 (300-2), or BS3 at the beginning of the UE journey or after a time offset from the beginning of the UE journey. (300-3) may refer to the above combination Figure 3 The core network entities (2250) may share the received UE information (2210) including details of the UE's destination address / location, UE speed and / or trajectory information with the core network entities connected to the BS (300-1) and / or BS described in 6. One or more core network entities (2250) may share the received UE information (or modified UE information) (2210) with base stations located on the route of the UE to the destination such as BS1 (300-1), BS2 (300-2) and BS3 (300-3). One or more core network entities (2250) may share the UE information (2220, 2230, 2240) with the base stations before the UE reaches the coverage of the corresponding base station. One or more core network entities (2250) may share the UE information (2220, 2230, 2240) with one base station at a time. In a variation of this embodiment, one or more core network entities (2250) may receive the UE information (2210) from a map application server (e.g., a Google Maps server) of the UE's map application (2200) via an API.

[0283] according to Fig.23 In the embodiment shown, the UE (200) (for example, as described above in conjunction with Figure 2 5) at the beginning of the UE journey or after a time offset from the beginning of the UE journey and connected to BS1 (300-1) (e.g., BS1 (300-1), BS2 (300-2), or BS3 (300-3) may refer to the above combination Figure 3 The location server (or network operator's server) (2350) may share the details of the received UE information (or modified UE information) (2310) with base stations (such as BS1 (300-1), BS2 (300-2), and BS3 (300-3)) located on the route from the UE to the destination. The location server (or network operator's server) (2350) may share the UE information (2320, 2330, 2340) with the base stations before the UE reaches the coverage of the corresponding base station. The location server (or network operator's server) (2350) may share the UE information (2320, 2330, 2340) with one base station at a time. In a variation of this embodiment, the location server (or network operator's server) (2350) may receive UE information (2310) from a map application server of the UE's map application (2300) (eg, Google Maps Server) via an API. Base station decision based on UE speed and trajectory

[0284] According to an embodiment, a BS (e.g., as described above in conjunction with Figure 3 and / or the BS described in FIG. 6) may be based on a UE (eg, as described above in conjunction with Figure 2 5) to predict situations such as high / low UE mobility. The BS may predict high / low UE mobility situations based on the traffic situation in the cell coverage of the BS at the predicted future location determined based on the speed and / or trajectory information of the UE (using a positioning module at the BS). The BS may determine the traffic situation using a real-time traffic map such as Google Maps or Apple Maps and the average speed of other UEs at the predicted location. Based on the prediction of high / low UE mobility situations, the BS may transmit to the UE a signal indicating one or more parameters (such as, for example, resource scheduling information, MIMO parameters (such as, for example, the number of layers, the number of antenna ports, or the number of codewords), modulation and coding schemes, one or more bandwidth parts, UE transmit power, waveform type (CP-OFDM or DFST-s-OFDM), numerical scheme / subcarrier spacing, and / or information about UE transmit / receive beam pairs). The UE may receive the updated signaling, extract the received parameters, and use the extracted parameters in the transmission / reception operation of the UE with the BS. The updated signaling may be received at the physical layer, the MAC layer, or the RRC layer.

[0285] According to an embodiment, a BS (e.g., as described above in conjunction with Figure 3 and / or the BS described in FIG. 6) may be based on a UE (eg, as described above in conjunction with Figure 2 5) speed and / or trajectory information to predict situations such as line-of-sight (LOS) / non-line-of-sight (NLOS). The BS may predict non-line-of-sight (NLOS) situations based on the determination of structures (such as, for example, buildings, forests, tunnels, underground passages, or mountains) in the cell coverage of the BS at the predicted future location determined based on the speed and / or trajectory information of the UE (using a positioning module at the BS). Based on the prediction of the line-of-sight (LOS) / non-line-of-sight (NLOS) situation, the BS may send to the UE an update signaling indicating one or more parameters (such as, for example, resource scheduling information, MIMO parameters (such as, for example, the number of layers, the number of antenna ports, the number of codewords), modulation and coding schemes, one or more bandwidth parts, UE transmit power, waveform type (CP-OFDM or DFST-s-OFDM), value scheme / subcarrier spacing, and / or information about UE transmit / receive beam pairs). The UE may receive the updated signaling, extract the received parameters, and use the extracted parameters in the transmission / reception operation of the UE with the BS. The signaling may be received at the physical layer, the MAC layer, or the RRC layer.

[0286] According to an embodiment, a BS (e.g., as described above in conjunction with Figure 3 and / or the BS described in FIG. 6) may be based on a UE (eg, as described above in conjunction with Figure 2 5) speed and / or trajectory information to predict situations such as high / low interference. The BS can predict interference situations based on the number of other UEs at predicted future locations determined based on the speed and / or trajectory information of the UE (using a positioning module at the BS). Based on the prediction of high / low interference situations, the BS can transmit to the UE updated signaling indicating one or more parameters (such as, for example, resource scheduling information, MIMO parameters (such as, for example, the number of layers, the number of antenna ports, or the number of codewords), modulation and coding schemes, one or more bandwidth parts, UE transmit power, waveform type (CP-OFDM or DFST-s-OFDM), numerical scheme / subcarrier spacing, and / or UE transmit / receive beam pair information). The UE can receive the updated signaling, extract the received parameters, and use the extracted parameters in the transmission / reception operations of the UE and the BS. The signaling can be received at the physical layer, the MAC layer, or the RRC layer.

[0287] According to an embodiment, a BS (e.g., as described above in conjunction with Figure 3 and / or the BS described in FIG. 6) may be based on a UE (eg, as described above in conjunction with Figure 2 5) speed and / or trajectory information to predict situations such as good / bad coverage. The BS can predict good / bad coverage situations based on historical signal strengths received at other UEs in the cell coverage of the BS at a predicted future location determined based on the speed and / or trajectory information of the UE (using a positioning module at the BS). Based on the prediction of good / bad coverage situations, the BS can transmit to the UE updated signaling indicating one or more parameters (such as, for example, resource scheduling information, MIMO parameters (such as, for example, the number of layers, the number of antenna ports, or the number of codewords), modulation and coding schemes, one or more partial bandwidths, UE transmit power, waveform type (CP-OFDM or DFST-s-OFDM), numerical scheme / subcarrier spacing, and / or UE transmit / receive beam pair information). The UE can receive the updated signaling, extract the received parameters, and use the extracted parameters in the transmission / reception operation of the UE with the BS. The updated signaling can be received at the physical layer, MAC layer, or RRC layer.

Claims

1. A method performed by a user equipment (UE), comprising: operating a first AI / ML model having a first AI / ML model configuration within a coverage area of ​​a first base station (BS1); transmitting one or more UE parameters to the BS1; as well as Receiving, by the UE, first signaling from the BS1, where the first signaling includes information about a second AI / ML model configuration; wherein the first signaling is transmitted by the BS1 based on a determination of a potential handover of the UE from the coverage of the BS1 to the coverage of the BS2; wherein the determination of the switching is performed based on the UE parameter; and The one or more UE parameters include UE speed, UE location, UE trajectory information, signal quality of BS1 and signal quality of the second base station (BS2).

2. The method according to claim 1, wherein: The information about the second AI / ML model configuration includes a second AI / ML model configuration identifier.

3. The method according to claim 1, wherein: The first signaling is received in a handover command message.

4. A method performed by a user equipment (UE), comprising: operating a first AI / ML model having a first AI / ML model configuration within a coverage area of ​​a first base station (BS1); transmitting one or more UE parameters to the BS1; Receiving, by the UE from the BS1, first signaling including information about a handover to a second base station (BS2); wherein the first signaling is transmitted by the BS1 based on a determination of a potential handover of the UE from the coverage of the BS1 to the coverage of the second base station (BS2); wherein the determination of the handover is made based on the UE parameters; wherein the one or more UE parameters include UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of the second base station (BS2); The UE transmits a PRACH to the BS2, and A second signaling is received by the UE from the BS2, where the second signaling includes information about a second AI / ML model configuration.

5. The method according to claim 4, wherein: The information about the second AI / ML model configuration includes a second AI / ML model configuration identifier.

6. The method according to claim 4, wherein: The first signaling is received in a handover command message.

7. A device comprising: an AI / ML engine configured to operate a first AI / ML model having a first AI / ML model configuration within a coverage area of ​​a first base station (BS1); A transceiver configured to transmit one or more UE parameters to the BS1; wherein the one or more UE parameters include UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of a second base station (BS2); The transceiver is configured to receive first signaling from the BS1, wherein the first signaling includes information about a second AI / ML model configuration; wherein the first signaling is transmitted by the BS1 based on a determination of a potential handover of the UE from the coverage of the BS1 to the coverage of the BS2; Therein, the determination of the switching is performed based on the UE parameters.

8. The device according to claim 7, wherein: The information about the second AI / ML model configuration includes a second AI / ML model configuration identifier.

9. The device according to claim 7, wherein: The first signaling is received in a handover command message.

10. An apparatus comprising: an AI / ML engine configured to operate a first AI / ML model having a first AI / ML model configuration within a coverage area of ​​a first base station (BS1); A transceiver configured to transmit one or more UE parameters to the BS1; wherein the one or more UE parameters include UE speed, UE location, UE trajectory information, signal quality of BS1, and signal quality of a second base station (BS2); The transceiver is configured to receive first signaling from the BS1, the first signaling comprising handover information to the second base station (BS2); wherein the first signaling is transmitted by the BS1 based on a determination of a potential handover of the UE from the coverage of the BS1 to the coverage of the second base station (BS2); wherein the determination of the handover is performed based on the UE parameter; The transceiver is configured to transmit a PRACH to the BS2, and A second signaling is received from the BS2, where the second signaling includes information about a second AI / ML model configuration.

11. The device according to claim 10, wherein: The information about the second AI / ML model configuration includes a second AI / ML model configuration identifier.

12. The device according to claim 10, wherein: The first signaling is received in a handover command message.

Images