[go: up one dir, main page]

WO2025171677A1 - Procédé de communication sans fil et dispositif de communication sans fil conçus pour surveiller un modèle/une fonction d'ai/ml - Google Patents

Procédé de communication sans fil et dispositif de communication sans fil conçus pour surveiller un modèle/une fonction d'ai/ml

Info

Publication number
WO2025171677A1
WO2025171677A1 PCT/CN2024/077477 CN2024077477W WO2025171677A1 WO 2025171677 A1 WO2025171677 A1 WO 2025171677A1 CN 2024077477 W CN2024077477 W CN 2024077477W WO 2025171677 A1 WO2025171677 A1 WO 2025171677A1
Authority
WO
WIPO (PCT)
Prior art keywords
monitoring
model
function
condition
side device
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
PCT/CN2024/077477
Other languages
English (en)
Chinese (zh)
Inventor
张笛笛
张银成
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen TCL New Technology Co Ltd
Original Assignee
Shenzhen TCL New Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen TCL New Technology Co Ltd filed Critical Shenzhen TCL New Technology Co Ltd
Priority to PCT/CN2024/077477 priority Critical patent/WO2025171677A1/fr
Publication of WO2025171677A1 publication Critical patent/WO2025171677A1/fr
Pending legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/16—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks using machine learning or artificial intelligence
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W24/00—Supervisory, monitoring or testing arrangements
    • H04W24/02—Arrangements for optimising operational condition

Definitions

  • the embodiments of the present application relate to the field of mobile communication technology, and specifically to a wireless communication method and wireless communication device for monitoring artificial intelligence/machine learning (AI/ML) models/functions.
  • AI/ML artificial intelligence/machine learning
  • AI/ML artificial intelligence/machine learning
  • a wireless communication method and wireless communication device for monitoring AI/ML models/functions are needed to address these and other issues in the existing technology.
  • Embodiments of the present application provide a wireless communication method and wireless communication device for monitoring artificial intelligence/machine learning (AI/ML) models/functions.
  • AI/ML artificial intelligence/machine learning
  • An embodiment of the present application provides a wireless communication method for monitoring an AI/ML model/function, which is executed on a user equipment (UE), wherein the wireless communication method includes: receiving multiple monitoring time windows configured by a network-side device, wherein the multiple monitoring time windows are periodic; and before monitoring the AI/ML model/function during a first monitoring time window of the multiple monitoring time windows, receiving first signaling sent by the network-side device, wherein the first signaling is used to indicate whether to perform monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model during the first monitoring time window.
  • the wireless communication method includes: receiving multiple monitoring time windows configured by a network-side device, wherein the multiple monitoring time windows are periodic; and before monitoring the AI/ML model/function during a first monitoring time window of the multiple monitoring time windows, receiving first signaling sent by the network-side device, wherein the first signaling is used to indicate whether to perform monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the
  • the first signaling is used to indicate whether to perform monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on an AI/ML model during the first monitoring time window. In this way, unnecessary monitoring can be avoided, reducing the UE's processing complexity, power consumption, and reporting overhead.
  • An embodiment of the present application provides a wireless communication method for monitoring AI/ML models/functions, which is executed on a UE, wherein the wireless communication method includes: determining whether to perform monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model based on a first condition and/or a second condition, wherein the first condition is that a monitoring quantity related to the monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model is lower than, higher than or equal to a defined threshold, or an event related to the monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model occurs, and the first condition and the second condition are related.
  • the first condition and/or the second condition it is determined whether to perform monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model. This can avoid unnecessary monitoring and reduce the processing complexity, power consumption, and reporting overhead of the UE.
  • An embodiment of the present application provides a wireless communication method for monitoring AI/ML models/functions, which is executed on a UE, wherein the wireless communication method includes: receiving multiple first monitoring time windows and at least one second monitoring time window configured by a network-side device, wherein the multiple first monitoring time windows are periodic; and based on the behavior of the UE during the at least one second window, determining whether to perform monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model during the multiple first monitoring time windows.
  • a decision is made as to whether to perform monitoring of at least one AI/ML model under the same function, or monitoring of a function implemented based on an AI/ML model, during the multiple first monitoring time windows. This avoids unnecessary monitoring and reduces UE processing complexity, power consumption, and reporting overhead.
  • An embodiment of the present application provides a wireless communication method for monitoring AI/ML models/functions, which is executed on a network-side device.
  • the wireless communication method includes: configuring multiple monitoring time windows, wherein the multiple monitoring time windows are periodic; and before monitoring the AI/ML model/function during a first monitoring time window of the multiple monitoring time windows, determining whether to perform monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model during the first monitoring time window.
  • An embodiment of the present application provides a wireless communication method for monitoring AI/ML models/functions, which is executed on a network-side device, wherein the wireless communication method includes: determining whether to perform monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model based on a first condition and/or a second condition, wherein the first condition is that a monitoring quantity related to the monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model is lower than, higher than, or equal to a defined threshold, or an event related to the monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model occurs, and the first condition and the second condition are related.
  • a decision is made as to whether to monitor at least one AI/ML model under the same function or a function implemented by an AI/ML model. This avoids unnecessary monitoring and reduces processing complexity, power consumption, and overhead of network-side devices.
  • a decision is made as to whether to monitor at least one AI/ML model under the same function, or a function implemented by the AI/ML model, during the multiple first monitoring time windows. This avoids unnecessary monitoring and reduces processing complexity, power consumption, and overhead of the network-side device.
  • a wireless communication device provided in an embodiment of the present application includes: a processor and a memory, the memory being used to store a computer program, and the processor being used to call and run the computer program stored in the memory to execute the above-mentioned wireless communication method.
  • the chip provided in the embodiment of the present application is used to implement the above-mentioned wireless communication method.
  • FIG3A is a schematic diagram of a flow chart of a wireless communication method provided in an embodiment of the present application.
  • FIG3C is a schematic diagram of a flow chart of a wireless communication method provided in an embodiment of the present application.
  • FIG3D is a schematic diagram of a flow chart of a wireless communication method provided in an embodiment of the present application.
  • FIG3E is a schematic flow chart of a wireless communication method according to an embodiment of the present application.
  • FIG3F is a schematic flow chart of a wireless communication method according to an embodiment of the present application.
  • FIG4A is a schematic diagram of a flow chart of a wireless communication method provided in an embodiment of the present application.
  • FIG4B is a schematic flow chart of a wireless communication method according to an embodiment of the present application.
  • the wireless communication system 100 may include a network-side device 110, which may be a device that communicates with a user equipment 120 (User Equipment, UE).
  • the network-side device 110 may provide communication coverage for a specific geographical area and may communicate with user equipment located within the coverage area.
  • the network-side device 110 may be a base station or a location management function (LMF) for providing positioning services.
  • LMF location management function
  • the base station may be an evolved Node B (eNB or eNodeB) in an LTE system, or the base station may be a mobile switching center, a relay station, an access point, an in-vehicle device, a wearable device, a hub, a switch, a bridge, a router, a network-side device in a 5G network, or a base station in a future communication system.
  • eNB evolved Node B
  • eNodeB evolved Node B
  • the base station may be a mobile switching center, a relay station, an access point, an in-vehicle device, a wearable device, a hub, a switch, a bridge, a router, a network-side device in a 5G network, or a base station in a future communication system.
  • eNB evolved Node B
  • eNodeB evolved Node B
  • a user device configured to communicate via a wireless interface may be referred to as a "wireless communication user device 120," a “wireless user device 120,” or a “mobile user device 120.”
  • mobile user equipment 120 include, but are not limited to, satellite or cellular telephones; Personal Communications System (PCS) user equipment 120, which may combine a cellular radio telephone with data processing, fax, and data communications capabilities; a PDA, which may include a radiotelephone, a pager, Internet/Intranet access, a Web browser, a notepad, a calendar, and/or a Global Positioning System (GPS) receiver; and conventional laptop and/or handheld receivers or other electronic devices that include a radiotelephone transceiver.
  • PCS Personal Communications System
  • GPS Global Positioning System
  • User equipment may be referred to as access user equipment 120, a subscriber unit, a subscriber station, a mobile station, a mobile station, a remote station, a remote user device, a mobile device, a wireless communication device, or a user agent.
  • the access user device 120 can be a cellular phone, a cordless phone, a Session Initiation Protocol (SIP) phone, a Wireless Local Loop (WLL) station, a Personal Digital Assistant (PDA), a handheld device with wireless communication capabilities, a computing device or other processing device connected to a wireless modem, an in-vehicle device, a wearable device, a user device in a 5G network, or a user device in a future evolved PLMN, etc.
  • SIP Session Initiation Protocol
  • WLL Wireless Local Loop
  • PDA Personal Digital Assistant
  • Some embodiments of this application mainly use AI/ML to solve various problems in communication systems.
  • how to monitor the AI/ML model/function so that the AI/ML model/function can work better In order to better enable at least one AI/ML model under the same function or a function implemented based on the AI/ML model to work better, how to monitor the AI/ML model/function so that the AI/ML model/function can work better.
  • Some embodiments of this application mainly involve the setting of trigger conditions for AI/ML model/function monitoring, the design of triggering processes, and the control of terminal processing complexity, reducing terminal reporting overhead, ensuring system performance stability, etc.
  • the user equipment 120 decides whether to perform monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on an AI/ML model during the first monitoring time window. In this way, Unnecessary monitoring can be avoided, reducing the processing complexity, power consumption, and overhead of the user device 120.
  • the network device 110 determines whether to perform monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on an AI/ML model during the first monitoring time window. In this way, unnecessary monitoring can be avoided, reducing the processing complexity, power consumption, and overhead of the network device 110.
  • user devices 120 can perform direct user device communication (Device to Device, D2D) between each other.
  • D2D Device to Device
  • the 5G communication system or 5G network can also be referred to as a New Radio (NR) system or NR network.
  • NR New Radio
  • the wireless communication system 100 also includes a network 130.
  • Network 130 may be an IP mobile communication network operated by a mobile communication operator.
  • network 130 may be a core network used by a mobile communication operator that operates and manages the wireless communication system 100, or may be a core network used by a virtual mobile communication operator such as an MVNO (Mobile Virtual Network Operator).
  • MVNO Mobile Virtual Network Operator
  • the network 130 can be connected to the network side device 110 as a relay device for transmitting user data.
  • the user device 120 sends and receives user data via the network 130. It should be noted that the communication of user data is not limited to IP communication, and non-IP communication may also be used.
  • a device having wireless communication capabilities in a network/system may be referred to as a wireless communication device.
  • the wireless communication device may include a network-side device 110 having communication capabilities, a user device 120, and a network 130.
  • the network-side device 110 and the user device 120 may be the specific devices described above and will not be described in detail here.
  • the wireless communication device may also include other devices (network 130) in the wireless communication system 100.
  • the network 130 may include other network entities such as a network controller and a mobility management entity, which is not limited in the embodiments of the present application.
  • system and “network” are often used interchangeably in this document.
  • the term “and/or” in this document is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and/or B may represent three situations: A exists alone, A and B exist at the same time, and B exists alone.
  • the character “/” in this document generally indicates that the objects associated before and after are in an “or” relationship.
  • the term “configuration” may refer to "pre-configuration” and "network configuration”.
  • FIG2B is a flow chart of a wireless communication method provided in an embodiment of the present application.
  • the wireless communication method is executed on a network-side device and includes at least one of the following operations: Operation 201B: Configuring multiple monitoring time windows. The multiple monitoring time windows are periodic. Operation 202B: Before monitoring an AI/ML model/function during a first monitoring time window of the multiple monitoring time windows, determine whether to perform monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on an AI/ML model during the first monitoring time window.
  • the monitoring of AI/ML models/functions is performed in a periodic manner.
  • signaling is used to indicate whether the next time window is started/triggered, and at least one AI/ML model under the same function or a function implemented based on the AI/ML model is monitored.
  • the first signaling includes Media Access Control (MAC) control element (CE) signaling or downlink control information (DCI).
  • the UE also receives a second signaling sent by the network side device, and the second signaling is used to indicate or update the period and window length of the multiple monitoring time windows.
  • the second signaling includes Radio Resource Control (RRC) signaling, MAC CE signaling, or DCI.
  • RRC Radio Resource Control
  • the first signaling may also be RRC signaling.
  • the UE when the first signaling indicates deactivation of monitoring of the AI/ML model/function during the first monitoring time window, the UE immediately stops or stops performing monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model during the next monitoring time window.
  • the user device 120 activates or deactivates the monitoring of at least one AI/ML model under the same function or the monitoring of a function implemented based on the AI/ML model from the monitoring time window, or the user device 120 activates or deactivates the monitoring of at least one AI/ML model under the same function or the monitoring of a function implemented based on the AI/ML model from the next monitoring time window.
  • the user device 120 if the user device 120 receives an activation or deactivation indication of MAC CE signaling, DCI, or RRC signaling before the nth periodic monitoring time window, the user device 120 activates or deactivates the monitoring of at least one AI/ML model under the same function or the monitoring of a function implemented based on the AI/ML model during the nth monitoring time window.
  • the monitoring of the AI/ML model/function on the user device 120 side can immediately take effect after the period and monitoring window length are configured, and the monitoring of at least one AI/ML model under the same function or the monitoring of a function implemented based on the AI/ML model.
  • the network side device 110 sends an activation or deactivation indication through MAC CE signaling, DCI, or RRC signaling, so that the user device 120 activates or deactivates the monitoring of at least one AI/ML model under the same function or the monitoring of a function implemented based on the AI/ML model.
  • the wireless communication method further includes the UE sending a request message to the network side device to monitor the AI/ML model/function.
  • the user device 120 may have a better understanding of environmental changes, the environment in which the user device 120 is located, changes in the status of the AI/ML model/function, etc. Therefore, the user device 120 can send a request message to the network device 110 to monitor the AI/ML model/function, requesting that the network device 110 allocate time-frequency domain resources for reporting monitoring results to the user device 120. Based on the request of the user device 120, the network device 110 configures parameters such as the period of the monitoring time window, the monitoring window length, and the time-frequency domain resources for reporting monitoring results for the user device 120. In some embodiments, the network device 110 can also configure the period and length of the monitoring time window for the user device 120.
  • the multiple monitoring time windows have multiple candidate periods and window lengths.
  • the multiple candidate periods and window lengths of the multiple monitoring time windows are configured by a network-side device.
  • the period of the monitoring time window has multiple candidate periods and window lengths
  • the network side device 110 can configure different monitoring time window periods and window lengths for the user device 120 based on factors such as model type (model ID or model group ID), generalization, stability, system performance changes, changes in the external environment, etc.
  • the period and window length of the monitoring time window can be time slot level (how many time slots), frame level (how many time frames), second level (how many seconds), minute level (how many minutes), hour level (how many hours), day level (how many days), week level (how many weeks), month level (how many months), year level (how many years), etc.
  • the period of the configured monitoring time window is related to factors such as the generalization, stability, and changes in the external environment of the AI/ML model/function.
  • the network side device 110 can update the period of the monitoring time window currently configured for the user equipment 120, the window length of the monitoring window and other parameters by reconfiguring/reissuing MAC CE signaling, DCI, or RRC signaling, so that the configuration of the period of the monitoring time window and the window length of the monitoring window is more suitable for the currently activated or working AI/ML model/function.
  • the network-side device 110 is further configured to decide whether to activate or deactivate monitoring of the AI/ML model/function. In some embodiments of the present application, when the network-side device 110 decides to activate monitoring of the AI/ML model/function during the first monitoring time window, the network-side device 110 performs monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model during the next monitoring time window of the first monitoring time window and the monitoring time window after the next monitoring time window, until the network-side device 110 decides to deactivate monitoring of the AI/ML model/function.
  • the network-side device 110 when the network-side device 110 decides to deactivate monitoring of the AI/ML model/function during the first monitoring time window, the network-side device 110 immediately stops or stops performing monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model during the next monitoring time window.
  • a periodic monitoring method can also be adopted for the AI/ML model/function on the network side.
  • the monitoring period and window length can be agreed upon through standards, or it can be entirely an implementation behavior on the network side. That is, the period and window length of the monitoring time window are defined or configured.
  • whether to start/start/trigger the monitoring of at least one AI/ML model under the same function or the monitoring of a function implemented based on the AI/ML model in each monitoring time window can also be determined by the network side device 110 or based on the request of the user device 120.
  • the network side device 110 can determine which event in the lifecycle management (LCM) the network side device 110 triggers/starts based on the monitoring results of at least one AI/ML model under the same function or a function implemented based on the AI/ML model.
  • LCM events include, for example, activation, deactivation, selection, switching, fallback and other operations of the AI/ML model/function.
  • the monitoring of the AI/ML model/function is the monitoring of the currently activated or working AI/ML model/function, or the monitoring of all or part of the AI/ML models under the same function.
  • the monitoring of AI/ML models/functions herein may refer to the monitoring of currently activated or working AI/ML models/functions, or may refer to the monitoring of all or part of the AI/ML models under the same function, wherein if some AI/ML models are not currently in an activated or working state, in order to monitor them, they need to be activated and then monitored, or the activation of the AI/ML models to be monitored can be completed through monitoring message configuration.
  • the monitoring of activated or working AI/ML models/functions can be real-time, but the decision on LCM events of AI/ML models/functions may require monitoring of multiple AI/ML models under the same function, and these multiple AI/ML models may include those that are already activated or working, as well as those that are inactivated or not in working state.
  • the monitoring of AI/ML models/functions here may refer to the monitoring of at least one AI/ML model under the same function or the monitoring of a function implemented based on an AI/ML model, including those that have been activated or are working, and may also include those that are inactivated or not in working state.
  • AI/ML models/functions that are inactivated or not in working state they need to be activated or put into working state before monitoring.
  • the activation or working state means that the AI/ML model/function can perform normal reasoning output.
  • AI/ML model/function refers to, for example, at least one AI/ML model under the same function or a function implemented based on an AI/ML model.
  • AI/ML model refers to, for example, at least one AI/ML model under the same function.
  • AI/ML function refers to, for example, a function implemented based on an AI/ML model.
  • the network device 110 may issue an instruction to activate/start/trigger/instruct the user equipment 120 to monitor the at least one AI/ML model under the same function or a function implemented based on the AI/ML model for the AI/ML model/function on the user equipment 120 or the AI/ML model/function on the user equipment 120 in the dual-side AI/ML model/function.
  • the instruction may be issued using MAC CE signaling, DCI, or RRC signaling.
  • the user equipment 120 Based on the instruction issued by the network device 110, the user equipment 120 begins monitoring the at least one AI/ML model under the same function or a function implemented based on the AI/ML model during the next or current monitoring time window.
  • the user equipment 120 may monitor a single AI/ML model/function, multiple AI/ML models/functions simultaneously, or at least one AI/ML model supported by the same function.
  • the two sides refer to the network side device 110 and the user equipment 120.
  • monitoring AI/ML models/functions can refer to monitoring at least one AI/ML model under the same function or monitoring a function implemented based on an AI/ML model, including those that are already activated or working, and those that are inactivated or not working. Inactivated or not working AI/ML models/functions must first be activated or put into working state before monitoring. Activation or working state here means that the AI/ML model/function can perform normal inference output.
  • the monitoring of at least one AI/ML model under the same function, or the monitoring of a function implemented based on the AI/ML model further comprises: constraining the monitoring of at least one AI/ML model under the same function, or the monitoring of a function implemented based on the AI/ML model.
  • the constraint includes that the output of the monitored AI/ML model/function and the activated AI/ML model/function are based on the reference signal measurement results of the same or multiple adjacent time slots.
  • the inputs of the monitored AI/ML model/function and the activated AI/ML model/function are based on reference signal measurement results of the same or multiple adjacent time slots.
  • the monitoring process may also require constraining the behavior of the monitored AI/ML model/function, including measurement and reporting behavior, such as constraining the output of the monitored AI/ML model/function and the activated AI/ML model/function to be based on the same reference signal measurement result, or the activated AI/ML model/function and the monitoring AI/ML model/function to be based on the reference signal measurement results of multiple adjacent time slots, but the interval in the measurement signal time domain is not greater than the defined value, such as the input of the activated AI/ML model/function and the monitoring AI/ML model/function is based on the reference signal measurement results of the same or multiple adjacent time slots.
  • the monitored AI/ML model/function refers to all monitored AI/ML models/functions minus the AI/ML model/function currently working in the activated state.
  • the input of the AI/ML model/function needs to be based on the measurement results of the same time slot or multiple adjacent time slots.
  • the reporting of the AI/ML model/function monitoring results also needs to meet certain constraints and can be based on the same time slot or multiple adjacent time slots, or within a certain time window.
  • the sending of the reference signal for the activated AI/ML model/function may be periodic, semi-continuous, or non-periodic, and the relationship between the reference signal for the monitoring AI/ML model/function and the reference signal for the activated AI/ML model/function may be one or more of the following: the sending of the reference signal for the activated AI/ML model/function is periodic, and the sending of the reference signal for the monitoring AI/ML model/function is semi-continuous or non-periodic; the sending of the reference signal for the activated AI/ML model/function is semi-continuous, and the sending of the reference signal for the monitoring AI/ML model/function is semi-continuous or non-periodic; the sending of the reference signal for the activated AI/ML model/function is non-periodic, and the sending of the reference signal for the monitoring AI/ML model/function is also non-periodic.
  • the output result reporting of the activated AI/ML model/function is periodic
  • the output result reporting of the monitored AI/ML model/function is semi-continuous or non-periodic.
  • the period of reporting the output result of the monitored AI/ML model/function and the period of reporting the output result of the activated AI/ML model/function satisfy an integer multiple relationship.
  • the monitored AI/ML model/function refers to all monitored AI/ML models/functions minus the AI/ML model/function currently working in the activated state.
  • the starting/earliest reporting moment/time slot of the output result of the monitored AI/ML model/function coincides with a reporting moment/time slot of the periodic reporting of the output result of the activated AI/ML model/function.
  • the monitored AI/ML model/function refers to all monitored AI/ML models/functions excluding the AI/ML model/function currently working in the activated state.
  • the monitored AI/ML model/function refers to all monitored AI/ML models/functions (e.g., the first AI/ML model/function, the second AI/ML model/function, and the third AI/ML model/function) excluding the activated AI/ML model/function (e.g., the first AI/ML model/function).
  • the output result reporting of the activated AI/ML model/function e.g., the first AI/ML model/function
  • the output result reporting of the monitored AI/ML model/function e.g., the first AI/ML model/function and the second AI/ML model/function
  • the output result reporting of the activated AI/ML model/function is aperiodic
  • the output result reporting of the monitored AI/ML model/function is aperiodic
  • the reporting of the output results of the activated AI/ML model/function may also take various forms, such as periodic, aperiodic, and semi-continuous. If the output results of the activated AI/ML model/function and the monitoring AI/ML model/function are reported independently, the reporting of the output results of the activated AI/ML model/function and the monitoring AI/ML model/function must also meet certain constraints. For example, it can be one or more of the following:
  • the output result reporting of the currently activated AI/ML model/function is periodic
  • the output result reporting of other AI/ML models/functions monitored during the AI/ML model/function monitoring period may be semi-continuous or non-periodic, and the reporting period of the monitored AI/ML model/function or the AI/ML model/function output result is an integer multiple of the reporting period of the activated AI/ML model/function output result, or the reporting period of the activated AI/ML model/function output result is an integer multiple of the reporting period of the monitored AI/ML model/function output result.
  • the starting/earliest reporting time/time slot for monitoring the AI/ML model/function or the AI/ML model/function output result coincides with a reporting time/time slot for the periodic reporting of the output result of the activated AI/ML model/function; or there is no constraint on the period for semi-continuous reporting of the monitoring AI/ML model/function output result.
  • the reporting time/time slot of the output result of the monitoring AI/ML model/function coincides with a reporting time/time slot of the periodic reporting of the output result of the activated AI/ML model/function, or the interval is less than the defined value, for example, the two are at the same or adjacent time/time slot, or there is no constraint.
  • the output result reporting of the currently active AI/ML model/function is semi-continuous
  • the output result reporting of other AI/ML models/functions monitored during the monitoring period of the AI/ML model/function can be semi-continuous or aperiodic
  • the reporting period for the model/function's output results is an integer multiple of the reporting period for the activated AI/ML model/function's output results
  • the reporting period for the activated AI/ML model/function's output results is an integer multiple of the reporting period for the monitoring AI/ML model/function's output results
  • the starting/earliest reporting time/time slot for the monitoring AI/ML model/function's output results coincides with a reporting time/time slot for the periodic reporting of the activated AI/ML model/function's output results; or there are no constraints on the periodic reporting of the monitoring AI/ML model/function's output results.
  • the reporting time/time slot for the monitoring AI/ML model/function's output results coincides with a reporting time/time slot for the periodic reporting of the activated AI/ML model/function's output results, or the interval is less than a defined value, for example, they occur at the same or adjacent time/time slots, or there are no constraints.
  • the output result reporting of the currently active AI/ML model/function is non-periodic
  • the output result reporting of other AI/ML models/functions monitored during the monitoring period of the AI/ML model/function can be non-periodic, and the reporting time/time slot of the output result of the monitored AI/ML model/function coincides with the reporting time/time slot of the output result of the activated AI/ML model/function, or the interval is less than the defined value, for example, the two are at the same or adjacent time/time slot, or there is no constraint.
  • the activated AI/ML model/function may refer to at least one AI/ML model under the same function that is in an activated state or in a working state.
  • the activation or working state here means that the AI/ML model/function (at least one AI/ML model under the same function) can perform normal reasoning output.
  • the correspondence between the activated AI/ML model/function and the monitored AI/ML model/function may be that the monitored AI/ML model/function refers to all monitored AI/ML models/functions minus the AI/ML model/function currently working in the activated state.
  • the post-processing here refers to a series of processing/transformation/conversion based on the inferred results of the AI/ML model/function, and there is no restriction on the specific processing/transformation/conversion method. There is no restriction on the functions corresponding to the above-mentioned AI/ML model/function, nor is there any restriction on the reported amount.
  • Method 1 Multiple AI/ML models under different AI/ML models/functions use independent reporting.
  • the CSI corresponding to multiple AI/ML models conflicts, the CSI corresponding to the initial AI/ML model under the currently active AI/ML model/function has the highest reporting priority. Prioritizing the reporting of the CSI corresponding to the active AI/ML model/function can help maintain stable system performance.
  • the initial AI/ML model refers to the AI/ML model that is initially activated when the function takes effect.
  • the number of AI/ML models under the same function that the UE monitors simultaneously is used as a capability item of the UE. In some embodiments of the present application, based on the number of AI/ML models under the same function that the UE monitors simultaneously as a capability item of the UE, monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model is performed.
  • the number of AI/ML models under the same function that the network-side device monitors simultaneously is used as a capability item of the network-side device. In some embodiments of the present application, based on the number of AI/ML models under the same function that the network-side device monitors simultaneously as a capability item of the network-side device, monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model is performed.
  • the network-side device monitors different AI/ML models/functions in batches based on the configured monitoring time window, and the number of AI/ML models/functions monitored each time is the same or different.
  • Solution 1 Considering the processing capacity limitation of the user device 120, the number of AI/ML models under the same function that the user device 120 monitors simultaneously is regarded as a capability item of the user device 120.
  • the user device 120 reports the maximum number of AI/ML models under the same function that are monitored simultaneously or the maximum number of monitored models to the network device 110. For example, the user device 120 monitors 1 or 2 AI/ML models under the same function simultaneously as the basic capability of the user device 120, and the user device 120 monitors more than 1 or 2 AI/ML models under the same function simultaneously.
  • the AI/ML model is an optional capability of the user equipment 120.
  • the network-side device 110 can instruct the user equipment 120 to simultaneously monitor the number of AI/ML models under the same function, taking into account the constraints of uplink time-frequency domain resources.
  • the number of AI/ML models under the same function that the user equipment 120 is instructed by the network-side device 110 to monitor simultaneously cannot exceed the maximum number of AI/ML models under the same function that are reported by the user equipment 120 or the number of models to be monitored simultaneously.
  • Solution 2 Considering the processing power limitations of user device 120 and the effectiveness of AI/ML model/function monitoring, the standard stipulates or the network device 110 instructs the user device 120 on the length of the time window for AI/ML model/function monitoring. Furthermore, based on the capability item for simultaneously monitoring the number of AI/ML models under the same function reported by the user device 120, or based on the capability item for simultaneously monitoring the number of AI/ML models under the same function reported by the user device 120, the network device 110 instructs the user device 120 on the number of AI/ML models under the same function to monitor simultaneously, and the user device 120 monitors the AI/ML models under the same function.
  • the user device 120 may continuously monitor different AI/ML models under the same function based on the configured or agreed monitoring time window, and the number of AI/ML models under the same function monitored each time may be the same or different. Assuming that the total number of AI/ML models under the same function that need to be monitored is 5, and the number of AI/ML models under the same function that the user device 120 monitors simultaneously at a single time is 3, the user device 120 needs to complete the monitoring of the AI/ML models under the same function in two times. For example, the user device 120 monitors 3 AI/ML models under the same function for the first time.
  • the user device 120 uses the same monitoring time window to monitor the other 2 AI/ML models under the same function.
  • the window lengths of the multiple monitoring time windows can be different. It depends on the configuration of the network side device 110.
  • the network side device 110 or the user device 120 can also group multiple AI/ML models under the same function, and the window lengths of the monitoring windows corresponding to the AI/ML models in the same group are the same.
  • the standard may also specify multiple time windows, which the network-side device 110 configures to the user device 120 through RRC signaling.
  • Solution three Taking into account the effectiveness of AI/ML model/function monitoring, on the basis of the agreed monitoring time window or the monitoring time window configured by the network side device 110, the minimum effective monitoring time window length or the effective monitoring time window length of the user device 120 for a single AI/ML model/function can also be limited by agreement or the configuration of the network side device 110.
  • the user device 120 completes the monitoring of the AI/ML model/function based on the agreed monitoring time window length or the monitoring time window length configured by the network side device 110, as well as the minimum effective monitoring time window length or the effective monitoring time window length for a single AI/ML model/function.
  • event-triggered monitoring where monitoring is triggered based on a single result or performance, potentially leading to false triggers and unnecessary monitoring.
  • triggering based on statistical results requires continuous monitoring of the output results or performance, requiring long-term memory allocation and increasing processing complexity for user device 120/network device 110.
  • event-triggered monitoring relies heavily on the criteria for measuring the event. Setting a low threshold for the event metric can degrade system performance, while setting a high threshold for the metric can lead to unnecessary monitoring.
  • a short-term event + long-term event joint decision-making mechanism is adopted to avoid mismonitoring of the AI/ML model and leave a certain amount of operating space for processing on the network side or the terminal side. At the same time, this mechanism can also be used for decision-making of LCM events.
  • an event and periodic method is used to determine when to monitor at least one AI/ML model under the same function or monitoring of a function implemented based on an AI/ML model.
  • the above method can reduce the power consumption of the system to a certain extent while ensuring performance.
  • the solution of the fourth embodiment may be implemented in conjunction with the solutions of the first embodiment, the second embodiment, and/or the third embodiment, or may be implemented independently of the solutions of the first embodiment, the second embodiment, and/or the third embodiment.
  • the wireless communication method further includes triggering a fallback operation of the AI/ML model/function and notifying the network-side device of the fallback operation.
  • the event/condition triggering the fallback operation of the AI/ML model/function includes the failure of the AI/ML model/function to load, or the inability to implement the corresponding function based on the current AI/ML model/function, or the detection of an instantaneous difference or statistical information of the difference between the output result of the model and the true/approximate true label exceeding a predefined threshold, or the monitoring of system performance exceeding a defined threshold, thereby triggering the fallback operation of the AI/ML model/function.
  • the first signaling is further used to instruct the UE to perform the fallback operation, or the UE notifies the network-side device to perform the fallback operation.
  • the wireless communication method further includes triggering a rollback operation of the AI/ML model/function and notifying the user equipment (UE) of the rollback operation.
  • the event/condition triggering the rollback operation of the AI/ML model/function includes the failure of the AI/ML model/function to load, the inability to implement the corresponding function based on the current AI/ML model/function, the detection of an instantaneous difference or statistical information of the difference between the output result of the model and the true/approximate true label exceeding a predefined threshold, or the monitoring of system performance degradation exceeding a defined threshold, thereby triggering the rollback operation of the AI/ML model/function.
  • the network-side device performs the rollback operation, or the network-side device notifies the UE to perform the rollback operation.
  • the selection of the triggering moment is crucial. If the system performance is better than the performance of the traditional signal processing method when the AI/ML model/function monitoring is triggered, it may result in waste of network-side device 110 and/or user equipment 120 capabilities and waste of air interface resources. If the system performance is worse than the performance of the traditional signal processing method when the AI/ML model/function monitoring is triggered, the system performance cannot be guaranteed. In view of this, this embodiment believes that the following solution can be used to ensure the performance of the system:
  • the network-side device 110 and/or the user device 120 have the ability or authority to control the system to fallback to the traditional signal processing method, and the fallback operation may not depend on the monitoring results of the AI/ML model/function; for example, if the AI/ML model/function fails to load or if the system performance is significantly degraded, the network-side device 110 and/or the user device 120 can trigger the AI/ML model/function fallback operation and notify the other end of the fallback operation. For example, if the network-side device 110 detects that the system's performance has significantly degraded, the network-side device 110 can directly instruct the user device 120 through signaling to perform a fallback operation on the AI/ML model/function.
  • the user device 120 can directly fall back from the state of the AI/ML model/function to the traditional signal processing method and notify the network-side device 110 through signaling, so that the network-side device 110 can adopt corresponding processing.
  • the network-side device 110 after receiving the notification signaling from the user device 120, the network-side device 110 also falls back from the state of the AI/ML model/function to the traditional signal processing method.
  • a significant performance degradation for example, refers to a performance degradation exceeding a defined threshold.
  • Solution 2 Considering that the performance of AI/ML models/functions is affected by generalization, when the external environment or channel conditions change, if the generalization of the AI/ML model/function is poor, the system performance may be significantly affected. Monitoring of the AI/ML model/function is usually triggered only when the system performance is unstable or drops below a certain threshold. Therefore, it is difficult to guarantee the system performance during the monitoring process. To address the above issues, this solution proposes that during the monitoring of the AI/ML model/function, the network device 110 and/or the user device 120 directly fall back to the traditional signal processing method.
  • the network device 110 sends a signaling to notify the user device 120 to roll back the AI/ML model/function, or the user device 120 sends a signaling to notify the network device 110 to roll back the AI/ML model/function.
  • LCM life cycle management
  • the network side device 110 sends a signaling to notify the user device 120 to activate the AI/ML model/function, or the user device 120 sends a signaling to notify the network side device 110 to activate the AI/ML model/function; if the result after the AI/ML model/function is monitored is to switch to another AI/ML model under the same function, the network side device 110 sends a signaling to notify the user device 120 to switch the AI/ML model, or the user device 120 sends a signaling to notify the network side device 110 to switch the AI/ML model.
  • the indication of the event in the LCM corresponding to the above-mentioned AI/ML model/function can be indicated in the form of a bitmap, or it can be indicated by bits (bits) are used for indication, where the value of X is the type of event for which the network side device 110 or the user device 120 makes a decision after the AI/ML model/function is monitored.
  • the network device 110 or the user device 120 may instruct monitoring of a certain AI/ML model.
  • the monitoring may be real-time monitoring, that is, the system monitors the AI/ML model in real time and determines the corresponding LCM event based on the real-time monitoring results.
  • the scheme of the fifth embodiment may be implemented in conjunction with the schemes of the first embodiment, the second embodiment, the third embodiment and/or the fourth embodiment, or may be implemented independently of the schemes of the first embodiment, the second embodiment, the third embodiment and/or the fourth embodiment.
  • the AI/ML model can be It is indicated by the network side device to the UE, or it can be determined when the model is loaded, or it can be constrained by the standard, for example, it can be the AI/ML model with the smallest/largest AI/ML model ID, or it can be the AI/ML model that is loaded first.
  • the AI/ML model on the network side it can be determined by the network side device itself, or it can be agreed upon by the standard, for example, it can be the AI/ML model with the smallest/largest AI/ML model ID; and for the AI/ML model on both sides, it can be decided by the network side or the UE side and notified to the other side, or it can be agreed upon by the standard, for example, it can be the AI/ML model with the smallest/largest AI/ML model ID, or it can be the AI/ML model that is loaded first.
  • an AI/ML model that has been loaded under a function is upgraded or updated.
  • the UE receives the parameter/architecture information for the AI/ML model upgrade or update sent by the network side device, and the UE also receives the upgraded or updated AI/ML model indicated by the network side device.
  • the number of AI/ML models loaded by a single function of the network-side device is reported as a capability item of the network-side device.
  • an AI/ML model already loaded under a function is upgraded or updated.
  • the network-side device If the AI/ML model is deployed on the UE side and downloaded from the network-side device, the network-side device sends parameter/architecture information of the AI/ML model upgrade or update to the UE, and the network-side device also indicates the upgraded or updated AI/ML model to the UE.
  • this embodiment considers restricting the number of AI/ML models that can be loaded by a single function on the network device 110 and/or the user device 120.
  • the number of AI/ML models that can be downloaded by a single function on the user device 120 is reported as a capability item of the user device 120, or the standard directly restricts the number of AI/ML models that can be loaded by the network device 110 and/or the user device 120 for a single function.
  • the network device 110 or the user device 120 can upgrade or update the AI/ML model, or delete the AI/ML model, but the user device 120 or the network device 110 needs to be informed. For example, if the number of AI/ML models loaded by the network device 110 and/or the user device 120 under a certain function has reached the constraint of the maximum number of supported AI/ML models, if the network device 110 and/or the user device 120 needs to load a new AI/ML model for the function, if the AI/ML model is deployed on the user device 120 side and is downloaded from the network device 110 side, the network device 110 side needs to instruct the user device 120 to delete one of the AI/ML models and send the newly added AI/ML model to the user device 120.
  • the AI/ML model to be deleted can be indicated via a bitmap, with each bit corresponding to a specific AI/ML model. If the AI/ML model is deployed on the network device 110 and uploaded from the user device 120, the user device 120 needs to send a request to the network device 110, requesting that the network device 110 delete the specific AI/ML model and allocate resources for uploading the AI/ML model to the user device 120.
  • the AI/ML model to be deleted can also be indicated via a bitmap, with each bit corresponding to a specific AI/ML model.
  • the network device 110 For dual-sided AI/ML models, if the AI/ML model is sent from the network device 110 to the user device 120, the network device 110 needs to instruct the user device 120 to delete one of the AI/ML models and send the newly added AI/ML model to the user device 120.
  • the AI/ML model to be deleted can be indicated via a bitmap, with each bit corresponding to a specific AI/ML model. If an AI/ML model is uploaded from user device 120 to network device 110, user device 120 needs to send a request to network device 110 to delete the AI/ML model and allocate the resources used to upload the AI/ML model to user device 120.
  • the AI/ML model to be deleted can also be indicated using a bitmap, with each bit corresponding to an AI/ML model.
  • the network device 110 and/or user device 120 load multiple AI/ML models for a certain function
  • an AI/ML model already loaded for a certain function needs to be upgraded or updated
  • the network device 110 in addition to sending the parameters/architecture information for the AI/ML model upgrade or update, needs to indicate to the user device 120 which of the multiple AI/ML models is being upgraded or updated.
  • This specific indication can be provided via a bitmap, with each bit corresponding to a specific AI/ML model.
  • the user device 120 needs to send an AI/ML model upgrade or update request to the network device 110, notifying the network device 110 of which AI/ML model needs to be upgraded or updated.
  • This AI/ML model indication can be provided via a bitmap, with each bit corresponding to a specific AI/ML model.
  • network device 110 After receiving the request from user device 120, network device 110 sends a response message to user device 120, which is used to configure the parameters/architecture information for the AI/ML model upgrade or update.
  • the process is the same as that for single-side AI/ML models.
  • the scheme of the sixth embodiment may be implemented in conjunction with the scheme of the first embodiment, the second embodiment, the third embodiment, the fourth embodiment, and/or the fifth embodiment, or may be implemented independently of the scheme of the first embodiment, the second embodiment, the third embodiment, the fourth embodiment, and/or the fifth embodiment.
  • the first condition is that a monitoring quantity related to the monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model is lower than, higher than, or equal to a defined threshold, or an event related to the monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model occurs, and the first condition is related to the second condition.
  • FIG3B is a flow chart of a wireless communication method provided in an embodiment of the present application.
  • the wireless communication method is executed on a network-side device and includes at least one of the following operations: Operation 301B: Based on the first condition and/or the second condition, the network-side device determines whether to perform monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model.
  • the first condition is that the monitoring quantity related to the monitoring of at least one AI/ML model under the same function or the monitoring of a function implemented based on the AI/ML model is lower than, higher than or equal to a defined threshold or the occurrence of an event related to the monitoring of at least one AI/ML model under the same function or the monitoring of a function implemented based on the AI/ML model, and the first condition is related to the second condition.
  • the performance of AI/ML models/functions is related to their generalization
  • Monitoring the AI/ML models/functions will increase the processing complexity of the network or user equipment 120.
  • some monitoring results may need to be reported, resulting in high reporting overhead and other issues.
  • the decision-making process for an LCM event only requires monitoring of the currently activated/normally operating AI/ML models/functions, the impact on the processing complexity, power consumption, and reporting overhead of the network device 110 or user device 120 is relatively controllable.
  • the decision-making process for some LCM events may require monitoring multiple AI/ML models/functions simultaneously. These multiple AI/ML models/functions may be activated or operating, or they may be non-activated or non-operating AI/ML models/functions. These non-activated or non-operating AI/ML models/functions need to be activated before monitoring. This increases the processing complexity and power consumption of the network device 110 or user device 120, and may also increase the reporting overhead of the user device 120. Therefore, the network device 110 and/or user device 120 needs to determine when to start/activate simultaneous monitoring of multiple AI/ML models/functions based on the monitoring results of the currently operating AI/ML models/functions.
  • At least one AI/ML model under the same function or a function implemented based on the AI/ML model can be monitored in an event-triggered manner.
  • the monitoring window length can be determined using standard constraints, can be configured by the network-side device 110, or can be unconstrained.
  • the specific monitoring duration is controlled by the user device 120.
  • the triggering/starting/starting of the simultaneous monitoring of at least one AI/ML model under the same function or a function implemented based on the AI/ML model can be determined based on one or more conditions. For example, the monitoring of the AI/ML model/function is triggered/started/started when the first condition and/or the second condition are met.
  • an event triggering method can be used to determine whether to monitor at least one AI/ML model under the same function or a function implemented based on the AI/ML model simultaneously.
  • the monitoring window length can be determined by a standard constraint method or the window length can be unconstrained.
  • the specific monitoring duration is controlled by the network side device 110.
  • the triggering/starting/starting of the monitoring of at least one AI/ML model under the same function or the monitoring of a function implemented based on the AI/ML model can be determined based on one or more conditions. For example, the monitoring of at least one AI/ML model under the same function or the monitoring of a function implemented based on the AI/ML model is triggered/started/started when the first condition and/or the second condition are met.
  • LCM events include, for example, activation, deactivation, selection, switching, fallback and other operations of the AI/ML model/function.
  • events and cycles are used to determine when to monitor at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model.
  • the above method can reduce the power consumption of the system to a certain extent while ensuring performance.
  • whether to perform monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model is determined according to the instruction of the network side device, and the instruction of the network side device is based on the first condition and/or the second condition to determine whether to perform monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model.
  • the triggering/starting/starting of simultaneous monitoring of at least one AI/ML model under the same function or a function implemented based on an AI/ML model can be based on a request sent by the user device 120. If the user device 120 meets the first condition and/or the second condition, the user device 120 sends a request to the network side device 110, requesting the network side device 110 to monitor at least one AI/ML model under the same function or a function implemented based on an AI/ML model.
  • the window length monitored by the network side device 110 can be determined by standard constraints, or the window length can be unconstrained, and the specific monitoring duration is controlled by the network.
  • whether to perform monitoring of at least one AI/ML model under the same function or monitoring of a function implemented based on the AI/ML model is determined based on the request information and the indication information in response to the request information, wherein the UE decides whether to send the request information to the network side device based on the first condition and/or the second condition.
  • the user device 120 may first initiate a request message to the network side device 110 for simultaneously monitoring at least one AI/ML model under the same function or a function implemented based on the AI/ML model.
  • the network side device 110 issues an instruction message based on the request of the user device 120, instructing the user device 120 to simultaneously monitor at least one AI/ML model under the same function or a function implemented based on the AI/ML model.
  • the user device 120 side initiates the request is determined based on one or more conditions.
  • the triggering/starting/starting of the simultaneous monitoring of at least one AI/ML model under the same function may be initiated by one side and then instructed or requested to the other side, or it may be based on the request of one side, and then the other side initiates and instructs the other side to trigger/start/start the simultaneous monitoring of at least one AI/ML model under the same function.
  • the user device 120 first requests the network side to trigger/start/start the simultaneous monitoring of at least one AI/ML model under the same function, and the request information may also include specific triggering/starting/starting time information, and may also include information about the corresponding AI/ML model to be monitored.
  • the network side decides to simultaneously monitor at least one AI/ML model under the same function based on the request of the user device 120.
  • the network side may also instruct the user device 120 which AI/ML models to monitor, and may also indicate the time information of triggering/starting/starting the monitoring, as well as the monitoring duration information.
  • the network side may also send an indication message to the user equipment 120, indicating that the user equipment 120 project has been started.
  • AI/ML models on the network side if the user device 120 requests to monitor at least one AI/ML model under the same function or a function implemented based on an AI/ML model, the specific AI/ML models to be monitored may be determined by the user device 120's request, or the network may independently decide which AI/ML models to monitor.
  • the network device 110 instructs the user device 120 to monitor at least one AI/ML model under the same function or a function implemented based on an AI/ML model
  • the network device 110 may also instruct the user device 120 to monitor the specific AI/ML models, or the user device 120 may independently decide which AI/ML models to monitor.
  • the occurrence of an event related to the monitoring of at least one AI/ML model under the execution of the same function or the monitoring of a function implemented based on the AI/ML model refers to whether the monitored quantity matches the reference value or the error is within a defined range.
  • the second condition is that the first condition persists for a period of time, or the second condition is how many times the first condition occurs consecutively within a period of time, or the total number of times the first condition occurs within a period of time is greater than, less than, or equal to the defined threshold, or the probability distribution of the occurrence of the first condition within a period of time is greater than, less than, or equal to the defined threshold.
  • the determination of the first condition and/or the second condition can be based on the monitoring results of the currently activated AI/ML model/function or the working AI/ML model/function.
  • the above-mentioned first condition can be at least one of the following forms: a certain quantity X is lower than/higher than/equal to a defined threshold Y; or it can be the occurrence of a certain event, such as whether a certain variable (monitored quantity) matches the actual measurement value (reference value), or the error is within a defined range.
  • the second condition can be that the first condition persists for a period of time, such as multiple time slots or frames; it can also be how many times the first condition appears consecutively; it can also be how many times the first condition appears consecutively within a period of time (consecutive multiple time slots or frames); it can also be that the total number of times the first condition appears within a period of time is greater than/less than/equal to a defined threshold; it can also be that the probability distribution of the first condition appearing within a period of time is greater than/less than/equal to a defined threshold.
  • the "continuous period of time” mentioned in the second condition can be a standard agreement for the network device 110 or an internal implementation of the network device 110.
  • the conditions for the user device 120 can also be a standard agreement, a configuration of the network device 110, or an internal implementation of the user device 120 itself.
  • the "continuous period of time" in the second condition can be in the form of a time window, the window length of which can be configured by the network or agreed upon by the standard.
  • a certain quantity X may be different, and the corresponding threshold value Y may also be different.
  • a certain quantity X can be an SINR value, a channel quality indicator (CQI) value, the system throughput, the BLER value, the modulation and coding scheme (MCS) value, or the correlation between the predicted CSI and the measured CSI.
  • the CSI here can be the channel, the channel eigenvector, or the eigenvector after SVD decomposition of the channel eigenvector.
  • the specific form of CSI is not limited here and can be any form of channel state information.
  • the correlation here can be represented by a value between 0 and 1, where 0 indicates complete no correlation and 1 indicates the best correlation between the two.
  • the quantity X can also be the difference between the positioning result of the AI/ML model/function and the ground truth label or the approximate ground truth label, or the difference between some intermediate positioning quantities and the ground truth label or the approximate ground truth label.
  • These intermediate positioning quantities can be downlink reference signal time difference (DL-RSTD), uplink RSTD (UL-RSTD), time of arrival (TOA), time difference of arrival (TDOA), timing time difference, round trip time (RTT), or reference signal received power (RSRP), reference signal received quality (RSRQ), power delay profile (PDP), channel impulse response (CIR), channel frequency response (CFR), etc.
  • AoA can be used to represent AoA in the azimuth plane and/or the elevation plane.
  • the AoA in the azimuth plane can be referred to as azimuth/horizontal AoA.
  • the AoA in the elevation plane can be referred to as elevation/zenith AoA.
  • AoD can be used herein to encompass AoD in the azimuth plane and/or the elevation plane.
  • the AoD in the azimuth plane can be referred to as azimuth/horizontal AoD
  • the AoD in the elevation plane can be referred to as elevation/zenith AoD.
  • the network-side device 110 or the user device 120 After the first condition is met, the network-side device 110 or the user device 120 starts/triggering/requesting/continues to monitor the activated AI/ML model/function or the currently working AI/ML model/function for a certain period of time, and then based on the monitoring results, the network-side device 110 or the user device 120 decides whether to start/trigger/request simultaneous monitoring of at least one AI/ML model under the same function or a function implemented based on the AI/ML model. The process of the network-side device 110 or the user device 120 making the decision can be based on the internal implementation of the network-side device 110 or the user device 120.
  • the network side performs performance monitoring and decides whether to activate/deactivate/update/switching the AI/ML model/function.
  • the user equipment 120 side performs performance monitoring and decides whether to activate/deactivate/update/switching the AI/ML model/function.
  • the corresponding LCM event can be a fallback to a traditional processing method, a switching of at least one AI/ML model under the same function or a function implemented based on the AI/ML model, the activation of at least one AI/ML model under the same function or a function implemented based on the AI/ML model, the deactivation of at least one AI/ML model under the same function or a function implemented based on the AI/ML model, the selection of at least one AI/ML model under the same function or a function implemented based on the AI/ML model, or the upgrade of at least one AI/ML model under the same function or a function implemented based on the AI/ML model.
  • the user device 120 may request the network side device 110 to implement the corresponding LCM event, for example, to activate/deactivate/select/switch/upgrade one or more AI/ML models/functions, or to fall back to the traditional processing method. If it is an AI/ML model/function on the user device 120 side, the network side device 110 may instruct the user device 120 to implement the corresponding LCM event, including activating/deactivating/selecting/switching/upgrading one or more AI/ML models/functions, or to fall back to the traditional processing method.
  • the decision of the LCM event applies to multiple AI/ML models/functions or a group of AI/ML models/functions under multiple network side devices 110/TRPs at the same time.
  • the acquisition of the first and second conditions mentioned in the seventh embodiment can also be considered as the result of monitoring the AI/ML model/function.
  • the monitoring of the AI/ML model/function can be considered to be real-time, that is, as long as the performance indicators of the AI/ML model/function are detected in the working state, it can be understood as monitoring the AI/ML model/function.
  • LCM events can also be decided based on the real-time monitoring results of the AI/ML model/function. The decision-making process of the LCM event is as described in some of the above embodiments.
  • first condition the duration of monitoring the currently active AI/ML model/function after the first condition is met
  • second condition the duration of monitoring the currently active AI/ML model/function after the first condition is met
  • first condition and second condition may correspond to multiple threshold values, and different threshold values correspond to different LCM events.
  • the first condition corresponds to two thresholds, threshold 1 and threshold 2. If the first condition is less than/equal to threshold 1, the corresponding LCM event is a fallback operation.
  • the RSRP or SINR for beam management mentioned in the seventh embodiment can be the RSRP or SINR for beam management mentioned in the seventh embodiment that is less than threshold 1; if the first condition is greater than threshold 1 and less than/equal to threshold 2, the corresponding LCM event is the switching/selection/upgrade/activation/deactivation of the AI/ML model/function, etc.; if the first condition is greater than threshold 2, the current state is maintained; at the same time, if the first condition is greater than threshold 1 and less than/equal to threshold 2, it can also be the start/trigger/start of monitoring at least one AI/ML model under the same function or a function implemented based on the AI/ML model, including the currently activated or working AI/ML model/function, and some or all of the AI/ML models/functions that are not activated or in working state.
  • the first condition is greater than threshold 2
  • the current state is maintained.
  • the relationship between the first condition and thresholds 1 and 2 may also be different for different monitoring quantities. For example, if the first condition is greater than/equal to threshold 2, the corresponding LCM event is a rollback operation. If the first condition is less than threshold 2 and greater than/equal to threshold 1, the corresponding LCM event is switching/selecting/upgrading/activating/deactivating the AI/ML model/function, or starting/triggering/starting monitoring of at least one AI/ML model under the same function or a function implemented based on the AI/ML model.
  • the network device 110 or the user device 120 monitors at least one AI/ML model under the same function or a function implemented based on the AI/ML model, and decides on possible LCM events based on the monitoring results.
  • the monitoring duration can be controlled by a time window.
  • the window length of the time window can be agreed upon by the standard or determined by the network device 110 or the user device 120.
  • the AI/ML model/function on the user device 120 side it can be configured by the network side or configured by the network side based on the user device 120 request.
  • the AI/ML model/function on the network side it can be requested by the user device 120 or constrained by the standard.
  • the network device 110 or user device 120 may also select multiple or multiple groups of AI/ML models to support the normal operation of the functionality. Furthermore, for those AI/ML models whose performance falls below a threshold, the network device 110 or user device 120 will deactivate these AI/ML models. Furthermore, if the monitoring results of the quantity X corresponding to all AI/ML models do not meet the threshold, which can be greater than, less than, or equal to the threshold, the functionality will need to fall back to traditional signal processing methods.
  • the first condition and the second condition can be a compound condition.
  • the first condition or the second condition can be lower than/higher than/equal to the defined threshold and last for a period of time; or the first condition or the second condition can be lower than/higher than/equal to the defined threshold and appear consecutively for a plurality of times; or a certain event can occur and appear consecutively for a certain number of times or appear a certain number of times over a period of time thereafter.
  • a short-term event + long-term event joint decision-making mechanism is adopted to avoid mismonitoring of the AI/ML model and leave a certain amount of operating space for processing on the network side or the terminal side. At the same time, this mechanism can also be used for decision-making of LCM events.
  • an event and periodic method is used to determine when to monitor at least one AI/ML model under the same function or monitoring of a function implemented based on an AI/ML model.
  • the user device 120 reaches/encounters a periodic monitoring time window, the user device 120 starts/begins to monitor at least one AI/ML model under the same function or a function implemented based on the AI/ML model, wherein the multiple AI/ML models may be AI/ML models that are currently activated or working, or AI/ML models that are not currently activated or working. These AI/ML models that are not currently activated or working need to be activated before monitoring.
  • the user device 120 or the network device 110 starts/begins monitoring at least one AI/ML model under the same function or a function implemented based on the AI/ML model, or the network device 110 instructs the user device 120 to start/start monitoring at least one AI/ML model under the same function or a function implemented based on the AI/ML model, or the user device 120 requests the network device 110 to start/start monitoring at least one AI/ML model under the same function or a function implemented based on the AI/ML model, where the multiple AI/ML models can be AI/ML models that are currently activated or working, or can be AI/ML models that are not currently activated or working.
  • each periodic monitoring time window starts/starts to monitor at least one AI/ML model under the same function or a function implemented based on the AI/ML model can be determined by the AI/ML model on the user device 120 side according to the network side indication, or it can be decided by the user device 120 itself; for the AI/ML model on the network side, it can also be based on the network side's self-determination or the user device 120's request.
  • This method can periodically monitor the performance of the AI/ML model/function continuously to ensure that the system performance does not continue to deteriorate; on the other hand, it also avoids the performance deterioration of the AI/ML model/function due to channel mutations or external environment mutations or other emergencies; it also avoids the increase in the processing complexity of the user equipment 120 caused by monitoring the AI/ML model/function when the AI/ML model/function is working relatively stably, and the possible waste of uplink resources caused by reporting the monitoring results; at the same time, the above method can reduce the power consumption of the system to a certain extent while ensuring performance.
  • any one of the first condition, the second condition, the pre-monitoring time window and threshold value based on event triggering/starting can be configured through at least one of RRC, MAC CE, DCI, and standard constraints.
  • the monitoring time window triggered by the event and the monitoring time window triggered by the period may overlap. If the two overlap, the time window with the earlier start time will be used, and the monitoring of the AI/ML model/function will be processed based on that time window. It should be noted that if the monitoring time window of the period is earlier and the AI/ML model/function is not monitored in the current period monitoring time window, the event-based monitoring of the AI/ML model/function will be processed. Furthermore, if event-based monitoring of the AI/ML model/function occurs between the two period monitoring time windows, the time window of the next period monitoring can be skipped/ignored and the AI/ML model/function will not be monitored.
  • the scheme of the ninth embodiment may be implemented in conjunction with the schemes of the first, second, third, fourth, and fifth embodiments, and the sixth, seventh, and/or eighth embodiments, or may be implemented independently of the schemes of the first, second, third, fourth, and fifth embodiments, and the sixth, seventh, and/or eighth embodiments.
  • FIG. 5 is a schematic structural diagram of a wireless communication device 700 provided in an embodiment of the present application.
  • the wireless communication device can be a user equipment, a base station, or a network element.
  • the wireless communication device 700 shown in Figure 5 includes a processor 710, which can call and execute a computer program from a memory to implement the method in the embodiment of the present application.
  • the wireless communication device 700 may further include a memory 720.
  • the processor 710 may call and execute a computer program from the memory 720 to implement the method in the embodiment of the present application.
  • the memory 720 may be a separate device independent of the processor 710 or may be integrated into the processor 710.
  • the wireless communication device 700 may further include a transceiver 730.
  • the processor 710 may control the transceiver 730 to communicate with other devices.
  • the transceiver 730 may send information or data to other devices or receive information or data sent by other devices.
  • the transceiver 730 may include a transmitter and a receiver.
  • the transceiver 730 may further include one or more antennas.
  • the wireless communication device 700 may specifically be the network side device 110 of the embodiment of the present application, and the wireless communication device 700 may implement the corresponding processes implemented by the network side device 110 in each method of the embodiment of the present application. For the sake of brevity, they will not be repeated here.
  • An embodiment of the present application also provides a computer program product, including computer program instructions.
  • the computer program may be applied to the network-side device in the embodiments of the present application.
  • the computer program When the computer program is run on a computer, the computer executes the corresponding processes implemented by the network-side device in the various methods of the embodiments of the present application.
  • the computer program may be applied to the user equipment in the embodiments of the present application.
  • the computer program When the computer program is run on a computer, the computer executes the corresponding processes implemented by the user equipment in the various methods of the embodiments of the present application. For the sake of brevity, no further details are given here.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

Les modes de réalisation de la présente demande concernent un procédé de communication sans fil et un dispositif de communication sans fil conçus pour surveiller un modèle/une fonction d'intelligence artificielle/apprentissage automatique (AI/ML). Le procédé de communication sans fil est exécuté par un équipement utilisateur (UE). Le procédé comprend les étapes consistant à : recevoir une pluralité de fenêtres temporelles de surveillance configurées par un dispositif côté réseau ; et, avant de surveiller un modèle/une fonction d'AI/ML pendant une première fenêtre temporelle de surveillance parmi la pluralité de fenêtres temporelles de surveillance, recevoir une première signalisation envoyée par le dispositif côté réseau. La pluralité de fenêtres temporelles de surveillance a une périodicité. La première signalisation est utilisée pour indiquer s'il faut exécuter, pendant la première fenêtre temporelle de surveillance, la surveillance d'au moins un modèle d'AI/ML relevant de la même fonction ou la surveillance d'une fonction exécutée sur la base du modèle d'AI/ML.
PCT/CN2024/077477 2024-02-18 2024-02-18 Procédé de communication sans fil et dispositif de communication sans fil conçus pour surveiller un modèle/une fonction d'ai/ml Pending WO2025171677A1 (fr)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2024/077477 WO2025171677A1 (fr) 2024-02-18 2024-02-18 Procédé de communication sans fil et dispositif de communication sans fil conçus pour surveiller un modèle/une fonction d'ai/ml

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2024/077477 WO2025171677A1 (fr) 2024-02-18 2024-02-18 Procédé de communication sans fil et dispositif de communication sans fil conçus pour surveiller un modèle/une fonction d'ai/ml

Publications (1)

Publication Number Publication Date
WO2025171677A1 true WO2025171677A1 (fr) 2025-08-21

Family

ID=96772477

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2024/077477 Pending WO2025171677A1 (fr) 2024-02-18 2024-02-18 Procédé de communication sans fil et dispositif de communication sans fil conçus pour surveiller un modèle/une fonction d'ai/ml

Country Status (1)

Country Link
WO (1) WO2025171677A1 (fr)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023012359A1 (fr) * 2021-08-05 2023-02-09 Telefonaktiebolaget Lm Ericsson (Publ) Procédé de surveillance des performances d'un modèle ou d'un algorithme d'intelligence artificielle (ai)/d'apprentissage machine (ml)
CN116349279A (zh) * 2023-02-08 2023-06-27 北京小米移动软件有限公司 Ai模型的性能监测方法、装置、网络节点及存储介质
WO2024031692A1 (fr) * 2022-08-12 2024-02-15 富士通株式会社 Procédé et appareil de surveillance pour modèle ai/ml
WO2024032762A1 (fr) * 2022-08-12 2024-02-15 Qualcomm Incorporated Protocoles et signalisation pour la surveillance des performances de modèles d'intelligence artificielle et d'apprentissage automatique

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023012359A1 (fr) * 2021-08-05 2023-02-09 Telefonaktiebolaget Lm Ericsson (Publ) Procédé de surveillance des performances d'un modèle ou d'un algorithme d'intelligence artificielle (ai)/d'apprentissage machine (ml)
WO2024031692A1 (fr) * 2022-08-12 2024-02-15 富士通株式会社 Procédé et appareil de surveillance pour modèle ai/ml
WO2024032762A1 (fr) * 2022-08-12 2024-02-15 Qualcomm Incorporated Protocoles et signalisation pour la surveillance des performances de modèles d'intelligence artificielle et d'apprentissage automatique
CN116349279A (zh) * 2023-02-08 2023-06-27 北京小米移动软件有限公司 Ai模型的性能监测方法、装置、网络节点及存储介质

Similar Documents

Publication Publication Date Title
US11451452B2 (en) Model update method and apparatus, and system
CN109983720B (zh) 无线系统中的前向纠错码选择
CN111801909B (zh) 用于在无线通信系统中调整载波聚合操作的方法和装置
EP1434364A2 (fr) Contrôler la puissance de transmission d'une station de base radio pour la suppression d'interférences
US20220295396A1 (en) Systems and methods for performance-aware energy saving in a radio access network
US12015559B2 (en) Service processing method, device, chip, and computer program
KR20220038100A (ko) 무선 통신 방법, 단말 디바이스 및 네트워크 디바이스
US20240195745A1 (en) Network entity and user equipment for transmission rate control
CN116918424A (zh) 信道状态信息的上报方法及装置、终端设备、网络设备
US20240276311A1 (en) Voice fallback target cell selection method and apparatus, storage medium and electronic device
US20220338032A1 (en) Communication method and communication apparatus
US20210152278A1 (en) Interaction based thermal mitigation
US10911968B2 (en) Method for wireless communication, terminal, and non-transitory computer-readable storage medium
EP3966972B1 (fr) Procédé et dispositif de réseau pour adaptation de liaison utilisant des statistiques de retour d'ack et de nack
US20190393981A1 (en) Method and apparatus for handover aware cqi adjustment in wireless networks
WO2025171677A1 (fr) Procédé de communication sans fil et dispositif de communication sans fil conçus pour surveiller un modèle/une fonction d'ai/ml
WO2025044392A1 (fr) Procédé de surveillance de performances et dispositif de communication
WO2024169924A1 (fr) Procédé et appareil d'évaluation de performance, terminal et dispositif côté réseau
CN117956604A (zh) 无线通信方法、装置及存储介质
WO2024109393A1 (fr) Procédé et appareil de réglage de paramètre de qualité de service, et support de stockage
WO2019145171A1 (fr) Procédé, appareil et programme informatique pour modification de critères de contrôle d'admission
US20220408361A1 (en) Energy saving information transmission method, terminal and network device
US20230362712A1 (en) Method and apparatus for adjusting attribute parameter, and communication device
CN111436142A (zh) 周期资源配置、确定方法和设备
WO2025001885A1 (fr) Procédé et dispositif d'envoi d'informations et procédé et dispositif de réception d'informations

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 24924443

Country of ref document: EP

Kind code of ref document: A1