Communication method for configuring a terminal device

Abstract

Communication Method for Configuring a Terminal Device The present invention is related to a communication method for configuring a terminal device. The method comprises the following steps: sending, by the terminal device, a capability message to a network device, the capability message comprising a model identification (ID) based on a machine learning, ML, model of the terminal device, and an functionality ID based on functionalities of an Artificial Intelligence / Machine Learning (AI / ML) inference function of the terminal device, and receiving, by the terminal device from the network device, a configuration message, the configuration message comprising a terminal device-specific configuration based on the capability message for configuring the terminal device.

Description

[0001] 202405220 1

[0002] Description

[0003] Communication Method for Configuring a Terminal Device

[0004] The present disclosure relates to a system and a method for configuring terminal devices in a communication network based on artificial intelligence and machine learning capabilities. Specifically, the present disclosure pertains to the exchange of capability information and device-specific configuration between terminal devices and network devices utilizing AI / ML model identifications and inference functionalities.

[0005] Background

[0006] In the field of wireless communications, the integration of artificial intelligence (Al) and machine learning (ML) functionalities into terminal devices has become increasingly prevalent. These advancements have enabled terminal devices to perform advanced functions such as beam prediction, radio resource management (RRM) measurements, handover operations, and various forms of inference directly at the device level. AI / ML models are tailored on a per device-basis during a lifecycle management (LCM) process before the devices are used. Depending on the device’s capabilities (such as processing power) or the required functionalities (e.g. beam prediction), different AI / ML models, enabled to serve different tasks, are employed on devices. This leads to the problem that a variety of devices with different possibilities are part of a communication network. Conventional systems typically involve the exchange of capability information between terminal devices and network infrastructure to facilitate optimal configuration and operation. However, the manner in which these capabilities are signaled and utilized by network entities often lacks granularity and adaptability, particularly with respect to AI / ML-enabled functionalities. As a result, network devices may not be able to fully exploit the diverse and evolving capabilities of modern terminal devices, leading to suboptimal configuration and performance.

[0007] According to known approaches, capability signaling between terminal devices and network infrastructure is generally limited to basic hardware or protocol features, such as supported frequency bands, maximum data rates, or general processing capabilities. These systems do not adequately account for the specific AI / ML models or inference functions present on terminal devices, nor do they provide mechanisms for dynamically 202404028

[0008] 2 adapting configurations based on the actual functional capabilities or operational context of the device. Furthermore, the signaling mechanisms for instructing terminal devices to switch between different operational configurations are often inefficient or lack the flexibility required to support real-time adaptation to changing network conditions, user requirements, or AI / ML model updates. Despite the substantial advances in the field of wireless communications and the increasing adoption of AI / ML at the device level, there remains a need for more sophisticated methods for capability exchange and configuration management that can address the heterogeneity and dynamic nature of modern terminal devices.

[0009] Consequently, the technical problem lies in providing improved mechanisms for configuring terminal devices in wireless communication systems in a way that at least partially overcomes the disadvantages of known systems.

[0010] Summary

[0011] The present invention provides methods that enable more efficient and flexible configuration of terminal devices based on their artificial intelligence and machine learning capabilities, which overcome one or more of the disadvantages of known methods.

[0012] A first aspect of the invention provides a communication method for configuring a terminal device, comprising the steps of: sending, by the terminal device, a capability message to a network device, the capability message comprising a model identification, ID, based on a machine learning, ML, model of the terminal device, and an functionality ID based on functionalities of an Artificial Intelligence / Machine Learning, AI / ML, inference function of the terminal device, and receiving, by the terminal device from the network device, a configuration message, the configuration message comprising a terminal device-specific configuration based on the capability message for configuring the terminal device. Functionalities of an AI / ML inference function may refer to functionalities that are supported by the AI / ML inference function, hence, are supported functionalities. Which functionalities may be supported can differ for the applied model. Models may offer different possible prediction scopes (what is predicted - different inputs, different 202404028

[0013] 3 outputs), but may also have different technical requirements (esp. processing power and / or memory space) in dependence of the required time frame (how time-sensitive is the prediction result). Hence, for example, complex models may be used on terminal devices if there is a stable connection to a cloud service taking over the model processing. On the other hand, a rather simple model prediction may be required for time-sensitive predictions that have to be run on a terminal device with limited processing power.

[0014] The subject matter describes a method by which a terminal device, such as a user equipment (UE) in a wireless communication system, communicates its artificial intelligence and machine learning capabilities to a network device, such as a base station (NodeB). This method is however also applicable for scenarios where two same device types are communication (e.g. one UE as terminal device and another UE as network device). In this method, the terminal device sends a capability message to the network device. This capability message contains two key pieces of information: a model identification, which is an identifier associated with a particular machine learning model identification for lifecycle management (LCM), present on the terminal device, and a functionality identification, which is an identifier that reflects the supported functionalities of the AI / ML inference function that the terminal device is capable of supporting or is currently applying. The sending of a capability message may comprise that before sending a capability message, the terminal may determine the model ID and functionality ID as predefined values, or, optionally, based on a predefined classification table.

[0015] The term "model identification" may refer to a unique code or identifier that corresponds to a specific machine learning model implemented or available on the terminal device for the purpose of AI / ML model related operation during the lifecycle management procedure. This allows the network device to recognize which AI / ML models are present and potentially active on the terminal device.

[0016] The term " functionality identification" may refer to an identifier that indicates which functionalities or features, enabled by AI / ML inference, are supported or applicable on the terminal device. This could include, for example, the ability to perform certain types of predictions or optimizations, depending on the device's hardware and software capabilities.

[0017] Upon receiving this capability message, the network device responds by sending a configuration message back to the terminal device. This configuration message contains 202404028

[0018] 4 a terminal device-specific configuration, which is tailored based on the information provided in the capability message. The configuration may include parameters such as the frequency of AI / ML inference, the size of the data collection window, or the enabling or disabling of certain AI / ML functionalities, depending on the terminal device's processing capacity, memory, and other capabilities.

[0019] The configuration message received by the terminal device comprises information that may allow the terminal device to configure the ML model and the AI / ML functionalities. The advantages of this approach are that the network configures each terminal device individually based on its model ID and functionality ID, enabling adaptive management of AI / ML functionalities across diverse devices. This ensures that devices with varying processing capacities, memory constraints, or supported features receive configurations that are best suited to their operational context, thereby enhancing efficiency, performance, and user experience. Furthermore, this method is particularly beneficial in environments where terminal devices exhibit diverse characteristics, such as differing mobility profiles or device types, as it enables the network to accommodate these variations dynamically. Such a mechanism is especially relevant for advanced wireless systems, including those envisioned for beyond 5G and 6G, where AI / ML-driven optimization and energy efficiency are critical.

[0020] Consequently, an improved method for configuring the terminal device is provided.

[0021] In a further embodiment, the method may further comprise that the model ID comprises at least one information on a power capability of the terminal device, a processing power of the terminal device, a buffering capability of the terminal device, and / or dataset collection capability of the terminal device. Further the model ID may also comprise additional characteristics related to the lifecycle management of the AI / ML related operation during the complete AI / ML function lifetime. This may enable to add a high level component of any future (to be determined) LCM operation which may also be indicated using the model ID.

[0022] The model ID may represent a capability of a (terminal) device. The capability of a device may refer to its performance for example the energy or power status level (describing a battery charging status of full to empty or charging, or stationary power supply), computing power (describing available CPU, GPU, or other processing units), buffering power (describing available memory and its performance for running models, such as 202404028

[0023] 5

[0024] RAM), and / or a dataset collection capability (what is collected, and how many different sets of datasets are collected, a higher amount of collectible data causes means an increased model inference). The model ID may be a single (distinct) value that is unique for a device or at least for devices of similar (performance) capabilities.

[0025] In the context of the implementation, the mechanisms of communication between the terminal device and the network device remain centered on the exchange of messages, specifically the capability message sent from the terminal device to the network device and the configuration message received in return. The implementation introduces a refinement to the content of the capability message by specifying that the model identification included in the message comprises of at least one type of information related to the terminal device’s operational characteristics. These characteristics include but may not be limited to the power capability, which refers to the available energy resources or battery status of the terminal device, the processing power, which relates to the computational resources or the ability of the terminal device to execute complex operations, the buffering capability, which concerns the terminal device’s capacity to temporarily store data during processing or transmission, and the dataset collection capability, which addresses the terminal device’s ability to gather, store, or process data sets necessary for Al or ML inference functions. The communication mechanism itself is not altered in terms of the protocol or the sequence of message exchanges; rather, the enhancement lies in the granularity and specificity of the information conveyed within the model identification field of the capability message. By including such detailed operational parameters, the network device is enabled to make more informed and precise decisions when generating the terminal device-specific configuration message. This results in a configuration that is better tailored to the actual capabilities and limitations of the terminal device, thereby optimizing the deployment and operation of Al or ML inference functions on the device. The new feature brought by this implementation is the explicit inclusion of device-specific operational metrics within the model identification, which allows for a more nuanced and effective configuration process. This enhancement supports improved resource allocation, more efficient energy usage, and potentially higher performance of Al or ML tasks on the terminal device, as the network device can adapt its configuration instructions to the real-time status and capabilities of the terminal device. The implementation thus enhances adaptability by making the configuration process 202404028

[0026] 6 responsive to the terminal device’s specific and current capabilities as conveyed through the model identification information.

[0027] In a further embodiment, the method may further comprise that the functionality ID comprises at least one information on functionalities of the terminal device such as beam prediction, radio resource management (RRM) measurements, handover capability, channel state information (CSI) prediction and compression, and / or mobility enhancements.

[0028] The functionality ID may represent the functionalities of the (terminal) device.

[0029] The functionality ID may be a single (distinct) value that is unique for a device or at least for devices of similar functionalities.

[0030] The implementation introduces an additional mechanism of communication between the terminal device and the network device by specifying that the terminal device is configured to receive an operation message from the network device. This operation message serves as an instruction to the terminal device, directing it to operate either with the terminal device-specific configuration, which was previously determined based on the capability message, or alternatively with a default configuration. The operation message is not arbitrary but is generated based on considerations such as global model convergence, required use-case, and inference reporting. The communication mechanism thus involves the network device transmitting an explicit operational directive to the terminal device, which is distinct from the initial configuration message. The operation message acts as a dynamic control signal, allowing the network device to adapt the operational mode of the terminal device in real time according to broader system-level factors or changing requirements. The new feature brought by this implementation is the introduction of a feedback or control loop, whereby the network device can actively manage the operational state of the terminal device beyond the initial configuration. This enables the network device to ensure that the terminal device operates in a manner that is consistent with the current state of the global machine learning model, the specific use-case requirements, or the outcomes of inference reporting. For example, if global model convergence has not been achieved, the network device may instruct the terminal device to revert to a default configuration to maintain system stability or consistency. Alternatively, if a particular use-case demands a specific configuration for optimal performance, the operation message can direct the terminal device accordingly. Including 202404028

[0031] 7 inference reporting as a basis for the operation message further allows for adaptive operation based on the real-time performance or results of the AI / ML inference function. This feature enhances the flexibility and responsiveness of the system, allowing for more granular and context-aware management of terminal device configurations. It also introduces a mechanism for centralized oversight and coordination by the network device, which can be critical in distributed AI / ML systems where consistent operation across multiple terminal devices is necessary. Overall, the implementation enables the network device to dynamically instruct the terminal device to switch between configurations based on system-wide intelligence and reporting, improving robustness and efficiency.

[0032] In a further embodiment, the method may further comprise: receiving, by the terminal device from the network device, an operation message, the operation message indicating the terminal device to operate with the terminal device-specific configuration or with a default configuration, wherein the operation message is based on global model convergence, and / or required use-case and inference reporting.

[0033] The implementation introduces an additional mechanism of communication between the terminal device and the network device by specifying that the terminal device is configured to receive an operation message from the network device. This operation message serves as an instruction to the terminal device, directing it to operate either with the terminal device-specific configuration, which was previously determined based on the capability message, or alternatively with a default configuration (which may be commonly shared in the network or predefined). The operation message enables the network device to adapt the operational mode of the terminal device in real time based on considerations such as global model convergence (such as federated or distributed learning), required use-case, and inference reporting (such as dataset quality or accuracy thresholds). The communication mechanism thus involves the network device transmitting an explicit operational directive to the terminal device, which is distinct from the initial configuration message. The operation message acts as a dynamic control signal, allowing the network device to adapt the operational mode of the terminal device in real time according to broader system-level factors or changing requirements. The new feature brought by this implementation is the introduction of a feedback or control loop, whereby the network device can actively manage the operational state of the terminal device beyond the initial 202404028

[0034] 8 configuration. This enables the network device to ensure that the terminal device operates in a manner that is consistent with the current state of the global machine learning model, the specific use-case requirements, or the outcomes of inference reporting. For example, if global model convergence has not been achieved, the network device may instruct the terminal device to revert to a default configuration to maintain system stability or consistency. Alternatively, if a particular use-case demands a specific configuration for optimal performance, the operation message can direct the terminal device accordingly. The inclusion of inference reporting as a basis for the operation message further allows for adaptive operation based on the real-time performance or results of the AI / ML inference function. This feature enables more granular and context-aware management of terminal device configurations and provides centralized coordination by the network device, which is critical for consistent operation in distributed AI / ML systems.

[0035] In a further embodiment, the method may further comprise that the operation message has a length of 1 -bit, and the operation message is signaled via any of physical downlink control channel, PDCCH, physical random access channel, PRACH, physical downlink shared channel, PDSCH, or any of radio link control, RLC, system information block SIB, UE assistance information, UAI, or radio resource control, RRC.

[0036] PDCCH, PRACH and / or PDSCH may refer to Layer 1 (L1 ) signaling.

[0037] RLC may refer to Layer 2 (L2) signaling, SIB is Layer 3 content but mapped to L1 / L2 for delivery, and UAI is not a layer related signal but may rather be understood as a parametric encapsulation.

[0038] RRC may refer to Layer 3 (L3) signaling.

[0039] Optionally, the operation message may also be longer than 1 bit. This may be helpful if further information (than enable / disable) shall be signaled.

[0040] The implementation introduces specific mechanisms for the communication of an operation message between the terminal device and the network device, refining the general communication framework established in the broader claim set. In particular, it specifies that the operation message, which is a signaling message relevant to the configuration or operation of the terminal device in the context of Al or ML functionalities, is constrained to a length of 1 bit. This 1 -bit length indicates a highly efficient signaling approach, minimizing overhead and latency, and is particularly advantageous for 202404028

[0041] 9 time-sensitive or resource-constrained scenarios typical in wireless communications involving terminal devices with Al or ML capabilities. The implementation further delineates the possible channels and protocols through which this operation message may be transmitted. These include physical layer channels such as the physical downlink control channel, physical random-access channel, and physical downlink shared channel, as well as higher-layer signaling mechanisms like radio link control, system information block, user equipment assistance information, and radio resource control. By specifying these diverse options, the implementation ensures flexibility and compatibility with various network architectures and deployment scenarios. The inclusion of both physical and higher-layer signaling paths enables seamless integration of the 1 -bit operation message into control signal and data exchanges, supporting both initial access and ongoing configuration of the terminal device's Al or ML inference functions. The key feature is the explicit definition of a minimal, 1 -bit operation message signaled over multiple standardized channels and protocols, enhancing signaling efficiency, interoperability, and adaptability. This allows rapid, resource-efficient signaling between the network and terminal device, enabling dynamic configuration or operational changes with minimal overhead. The implementation addresses signaling efficiency and integration with existing procedures, ensuring Al or ML capabilities of the terminal device can be managed with minimal impact on network resources and battery life. This is especially relevant for frequent or low-latency updates or when signaling resources are limited, such as in massive machine-type or ultra-reliable low-latency communications. The implementation thus provides an efficient signaling mechanism for Al or ML-enabled terminal device configuration in modem wireless networks.

[0042] A second aspect of the invention provides a communication method for configuring a terminal device, comprising: receiving, by a network device from the terminal device, a capability message, the capability message comprising a model identification, ID, based on a machine learning, ML, model of the terminal device, and an functionality ID based on functionalities of an Artificial Intelligence / Machine Learning, AI / ML, inference function of the terminal device; configuring, by the network device, a terminal device-specific configuration for configuring the terminal device based on the capability message; and sending, by the network device to the terminal device, a configuration message, the configuration message comprising the terminal device-specific configuration. 202404028

[0043] 10

[0044] The subject matter concerns a method for configuring a terminal device, such as a user equipment (UE), in a wireless communication system, particularly in the context of supporting artificial intelligence and machine learning functionalities. The process begins with the network device, for example a base station (NodeB), receiving a capability message from the terminal device. This method is however also applicable for scenarios where two same device types are in communication (e.g. one UE as terminal device and another UE as network device). This capability message includes two key pieces of information: a model identification (ID) and a functionality ID. The model ID may refer to an identifier that specifies which machine learning model is present or supported on the terminal device. The functionality ID may refer to an identifier that indicates which functionalities of the AI / ML inference function are available or applicable on the terminal device, reflecting the device’s capabilities, such as processing power, memory, or supported features.

[0045] Upon receiving this capability message, the network device configures a terminal device-specific configuration. This means that the configuration is tailored to the particular characteristics and capabilities of the terminal device, as indicated by the model ID and functionality ID. Such configuration may include parameters like inference period, dataset collection window, or enabling / disabling certain functionalities, depending on the device’s reported capabilities, model performance, and / or the network’s requirements. After preparing this device-specific configuration, the network device sends a configuration message back to the terminal device. This message contains the specific configuration intended for that terminal device, ensuring that the device operates optimally within the network, taking into account its unique AI / ML capabilities and functionalities. The combination of capabilities and functionalities of a terminal device may also be referred to as applicability or applicability conditions.

[0046] Applicability conditions may include the functionalities that the UE is prepared to support or execute for model inference. The network device may use these applicability conditions to determine and configure the terminal device-specific configuration. A terminal device-specific configuration may instruct the terminal device to enable / disable certain functionalities and / or adjusting parameters such as inference periods or dataset collection windows. Applicability conditions may be distinct from general terminal device capabilities, and may specifically indicate which AI / ML-enabled functionalities are "applicable" (i.e. , ready to be used) for a given terminal device at a given time. 202404028

[0047] 11

[0048] The advantage of this approach is that it allows for individualized configuration of terminal devices based on their specific AI / ML capabilities and functionalities. This is particularly beneficial in scenarios where devices have diverse characteristics, such as varying processing capacities, memory limits, or support for different AI / ML features. By enabling the network to tailor configurations for each device, the method supports more efficient and effective deployment of AI / ML functions across a heterogeneous device population, improving overall system performance and adaptability to different use cases, including those relevant to future wireless standards.

[0049] In a further embodiment, the method may further comprise that the model ID comprises at least one information on a power capability of the terminal device, a processing power of the terminal device, a buffering capability of the terminal device, and / or dataset collection capability of the terminal device. The model ID may also be used as an identifier to address any or all of the procedures involved during the lifecycle management of the AI / ML functionality (esp. comprising future procedures). This may enable keeping the model updated for advances or changes during lifecycle management.

[0050] The model ID may represent the capability of a (terminal) device. The capability of a device may refer to its performance for example the energy or power status level (describing a battery charging status of full to empty or charging, or stationary power supply), computing power (describing available CPU, GPU, or other processing units), buffering power (describing available memory and its performance for running models, such as RAM), and / or a dataset collection capability (what is collected, and how many different sets of datasets are collected, a higher amount of collectible data causes means an increased model inference). The model ID may be a single (distinct) value that is unique for a device or at least for devices of similar (performance) capabilities.

[0051] In the context of the implementation, the mechanisms of communication between the terminal device and the network device remain centered on the exchange of messages, specifically the capability message sent from the terminal device to the network device and the configuration message received in return. The implementation introduces a refinement to the content of the capability message by specifying that the model identification included in the message comprises at least one type of information related to the terminal device’s operational characteristics. These characteristics include the power capability, which refers to the available energy resources or battery status of the 202404028

[0052] 12 terminal device, the processing power, which relates to the computational resources or the ability of the terminal device to execute complex operations, the buffering capability, which concerns the terminal device’s capacity to temporarily store data during processing or transmission, and the dataset collection capability, which addresses the terminal device’s ability to gather, store, or process data sets necessary for Al or ML inference functions. The communication mechanism itself is not altered in terms of the protocol or the sequence of message exchanges; rather, the enhancement lies in the granularity and specificity of the information conveyed within the model identification field of the capability message. By including such detailed operational parameters, the network device is enabled to make more informed and precise decisions when generating the terminal device-specific configuration message. This results in a configuration that is better tailored to the actual capabilities and limitations of the terminal device, thereby optimizing the deployment and operation of Al or ML inference functions on the device. The new feature brought by this implementation is the explicit inclusion of device-specific operational metrics within the model identification, which allows for a more nuanced and effective configuration process. This enhancement supports improved resource allocation, more efficient energy usage, and potentially higher performance of Al or ML tasks on the terminal device, as the network device can adapt its configuration instructions to the real-time status and capabilities of the terminal device. The implementation thus enhances adaptability by making the configuration process responsive to the terminal device’s specific and current capabilities as conveyed through the model identification information.

[0053] In a further embodiment, the method may further comprise that the functionality ID comprises at least one information on functionalities of the terminal device such as beam prediction, radio resource management (RRM) measurements, handover capability, channel state information (CSI) prediction and compression, and / or mobility enhancements.

[0054] The functionality ID may represent the functionalities of the (terminal) device.

[0055] The functionality ID may be a single (distinct) value that is unique for a device or at least for devices of similar functionalities.

[0056] The implementation introduces an additional mechanism of communication between the terminal device and the network device by specifying that the terminal device is 202404028

[0057] 13 configured to receive an operation message from the network device. This operation message serves as an instruction to the terminal device, directing it to operate either with the terminal device-specific configuration, which was previously determined based on the capability message, or alternatively with a default configuration. The operation message is not arbitrary but is generated based on considerations such as global model convergence, required use-case, and inference reporting. The communication mechanism thus involves the network device transmitting an explicit operational directive to the terminal device, which is distinct from the initial configuration message. The operation message acts as a dynamic control signal, allowing the network device to adapt the operational mode of the terminal device in real time according to broader system-level factors or changing requirements. The new feature brought by this implementation is the introduction of a feedback or control loop, whereby the network device can actively manage the operational state of the terminal device beyond the initial configuration. This enables the network device to ensure that the terminal device operates in a manner that is consistent with the current state of the global machine learning model, the specific use-case requirements, or the outcomes of inference reporting. For example, if global model convergence has not been achieved, the network device may instruct the terminal device to revert to a default configuration to maintain system stability or consistency. Alternatively, if a particular use-case demands a specific configuration for optimal performance, the operation message can direct the terminal device accordingly. The inclusion of inference reporting as a basis for the operation message further allows for adaptive operation based on the real-time performance or results of the AI / ML inference function. This feature enhances the flexibility and responsiveness of the system, allowing for more granular and context-aware management of terminal device configurations. It also introduces a mechanism for centralized oversight and coordination by the network device, which can be critical in distributed AI / ML systems where consistent operation across multiple terminal devices is necessary. Overall, the implementation enables the network device to dynamically instruct the terminal device to switch between configurations based on system-wide intelligence and reporting, improving robustness and efficiency.

[0058] In a further embodiment, the method may further comprise that the functionality ID indicates a maximum of available functionalities of the terminal device, and the method 202404028

[0059] 14 further comprising: configuring, by the network device, a terminal device-specific configuration based on the capability message comprises associating the model ID to a subset of the maximum of available functionalities of the terminal device.

[0060] In the context of the implementation under analysis, the mechanisms of communication between the network device and the terminal device remain grounded in the exchange of messages, specifically the capability message sent from the terminal device to the network device and the configuration message sent in return. The capability message is enhanced by the inclusion of a functionality ID, which indicates the maximum set of supported functionalities of a terminal device. This functionality ID, therefore, provides the network device with a comprehensive overview of the terminal device's AI / ML inference function capabilities. The new feature introduced by this implementation is the explicit association, by the network device, between the model ID (representing the ML model of the terminal device) and a subset of the maximum available functionalities as indicated by the functionality ID. This enables the network device to selectively activate a certain subset of functionalities supported by a terminal device. Thus, the configuration is tailored by mapping the identified ML model to those functionalities most relevant for that model. As a result, the network device generates a terminal device-specific configuration that is more precise and efficient, leveraging the terminal's capabilities and the requirements of the ML model. This targeted association allows finer granularity in configuring terminal devices, improving performance and resource allocation in AI / ML inference tasks. The communication mechanism thus conveys information about the terminal device's capabilities and enables a context-aware configuration process, wherein the network device selects and activates only functionalities suited to the identified ML model. This approach reduces overhead, avoids enabling irrelevant features, and ensures the terminal device operates within its optimal functional envelope as determined by its capabilities and the specific AI / ML model. The net effect is a more adaptive and efficient system for configuring terminal devices in a networked AI / ML environment, with enhanced communication protocols supporting model-aware configuration decisions.

[0061] In a further embodiment, the method may further comprise: sending, by the network device to the terminal device, an operation message, the operation message indicating the terminal device to operate with the terminal device-specific configuration or with a 202404028

[0062] 15 default configuration, wherein the operation message is sent based on global model convergence, and / or required use-case and inference reporting.

[0063] The implementation introduces an additional mechanism of communication between the terminal device and the network device by specifying that the terminal device is configured to receive an operation message from the network device. This operation message serves as an instruction to the terminal device, directing it to operate either with the terminal device-specific configuration, which was previously determined based on the capability message, or alternatively with a default configuration (which may be commonly shared in the network or predefined). The operation message enables the network device to adapt the operational mode of the terminal device in real time based on considerations such as global model convergence (such as federated or distributed learning), required use-case, and inference reporting (such as dataset quality or accuracy thresholds). The communication mechanism thus involves the network device transmitting an explicit operational directive to the terminal device, which is distinct from the initial configuration message. The operation message acts as a dynamic control signal, allowing the network device to adapt the operational mode of the terminal device in real time according to broader system-level factors or changing requirements. The new feature brought by this implementation is the introduction of a feedback or control loop, whereby the network device can actively manage the operational state of the terminal device beyond the initial configuration. This enables the network device to ensure that the terminal device operates in a manner that is consistent with the current state of the global machine learning model, the specific use-case requirements, or the outcomes of inference reporting. For example, if global model convergence has not been achieved, the network device may instruct the terminal device to revert to a default configuration to maintain system stability or consistency. Alternatively, if a particular use-case demands a specific configuration for optimal performance, the operation message can direct the terminal device accordingly. The inclusion of inference reporting as a basis for the operation message further allows for adaptive operation based on the real-time performance or results of the AI / ML inference function. This feature enables more granular and context-aware management of terminal device configurations and provides centralized coordination by the network device, which is critical for consistent operation in distributed AI / ML systems. 202404028

[0064] 16

[0065] In a further embodiment, the method may further comprise that the operation message has a length of 1 -bit, and the operation message is signaled via any of physical downlink control channel, PDCCH, physical random access channel, PRACH, physical downlink shared channel, PDSCH, or any of radio link control, RLC, system information block SIB, UE assistance information, UAI, or radio resource control, RRC.

[0066] PDCCH, PRACH and / or PDSCH may refer to Layer 1 (L1 ) signaling.

[0067] RLC may refer to Layer 2 (L2) signaling, SIB is Layer 3 content but mapped to L1 / L2 for delivery, and UAI is not a layer related signal but may rather be understood as a parametric encapsulation.

[0068] RRC may refer to Layer 3 (L3) signaling.

[0069] The implementation introduces specific mechanisms for the communication of an operation message between the terminal device and the network device, refining the general communication framework established in the broader claim set. In particular, it specifies that the operation message, which is a signaling message relevant to the configuration or operation of the terminal device in the context of Al or ML functionalities, is constrained to a length of 1 bit. This 1 -bit length indicates a highly efficient signaling approach, minimizing overhead and latency, and is particularly advantageous for time-sensitive or resource-constrained scenarios typical in wireless communications involving terminal devices with Al or ML capabilities. The implementation further delineates the possible channels and protocols through which this operation message may be transmitted. These include physical layer channels such as the physical downlink control channel, physical random-access channel, and physical downlink shared channel, as well as higher-layer signaling mechanisms like radio link control, system information block, user equipment assistance information, and radio resource control. By specifying these diverse options, the implementation ensures flexibility and compatibility with various network architectures and deployment scenarios. The inclusion of both physical and higher-layer signaling paths enables seamless integration of the 1 -bit operation message into control and data exchanges, supporting both initial access and ongoing configuration of the terminal device's Al or ML inference functions. The key feature is the explicit definition of a minimal, 1 -bit operation message signaled over multiple standardized channels and protocols, enhancing signaling efficiency, interoperability, and adaptability. This allows rapid, resource-efficient signaling between the network and terminal device, enabling dynamic configuration or operational changes 202404028

[0070] 17 with minimal overhead. The implementation addresses signaling efficiency and integration with existing procedures, ensuring Al or ML capabilities of the terminal device can be managed with minimal impact on network resources and battery life. This is especially relevant for frequent or low-latency updates or when signaling resources are limited, such as in massive machine-type or ultra-reliable low-latency communications. The implementation thus provides an efficient signaling mechanism for Al or ML-enabled terminal device configuration in modem wireless networks.

[0071] A third aspect of the invention provides a terminal device configured to carry out the method according to any one of the implementations of the first aspect.

[0072] The method carried out by the terminal device involves receiving and applying configuration information from the network, where the configuration is tailored to the device's reported capabilities and the specific AI / ML functionalities it supports. These capabilities may include the device's processing power, memory, dataset collection ability, or other relevant characteristics. The configuration may also take into account applicability conditions, which may refer to the functionalities that the device is ready or able to apply for model inference.

[0073] The advantages of this arrangement are significant. By enabling the network to provide device-specific configurations, each terminal device can be optimized according to its own characteristics, such as its computational resources or mobility profile. This ensures that devices with varying capabilities receive configurations that are best suited to their performance envelope, leading to more efficient operation and potentially improved AI / ML inference accuracy. Furthermore, this approach is particularly beneficial in heterogeneous network environments where devices differ widely in their hardware and supported features. The ability to enable or disable specific configurations or functionalities dynamically, for example through a simple signaling mechanism, allows for flexible adaptation to changing network conditions or device states, thereby supporting advanced use cases in AI / ML-enabled wireless communication systems.

[0074] A fourth aspect of the invention provides a network device configured to carry out the method according to any one of the implementations of the second aspect.

[0075] The method carried out by the network device allows for the assignment of a configuration that is tailored to the unique capabilities and operational context of each UE. 202404028

[0076] 18

[0077] This may involve, for example, adjusting the frequency of AI / ML inference operations, the size of data collection windows, or enabling and disabling certain functionalities through a simple signaling mechanism, such as a 1 -bit indicator. The configuration is determined based on the information provided by the terminal device regarding its capabilities and the functionalities it is prepared to support.

[0078] The advantage of this approach is that it enables the network to optimize the use of AI / ML features on a per-device basis, taking into account the diversity of devices in terms of mobility, device type, and resource constraints. This leads to more efficient network operation, as each terminal device receives a configuration that matches its abilities, potentially improving performance, resource utilization, and user experience.

[0079] Furthermore, this method supports the dynamic and flexible management of AI / ML functionalities in advanced wireless networks.

[0080] A fifth aspect of the invention provides a computer program product, comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method according to any one of the implementations of the first aspect, or the method according to any one of the implementations of the second aspect.

[0081] The subject matter described is a computer program product that includes instructions which, when executed by a computer, cause the computer to perform a method for configuring terminal device in a wireless communication network, specifically in the context of AI / ML-enabled features. The method referenced in the aspect is based on associating a specific configuration for a terminal device according to its model identification (model ID) and functionality identification (functionality ID).

[0082] A computer program product may refer to any form of software or firmware, such as a non-transitory computer-readable medium containing program instruction. When these instructions are executed, the computer is caused to carry out the steps of the method as described in earlier implementations, which relate to configuring UEs in the network. Model ID may refer to an identifier that uniquely represents the AI / ML model or models supported or used by the UE. The model ID may further serve as an identifier to identify the process involved in the lifecycle management of the AI / ML model during its lifecycle. Functionality ID may refer to an identifier that represents the specific AI / ML-enabled feature or functionality that the terminal device can support or is configured to use. LCM 202404028

[0083] 19 may refer to lifecycle management, which is the process of managing the deployment, configuration, and operation of AI / ML models and functionalities on the UE.

[0084] The approach allows the network, such as a gNB (next-generation NodeB), to configure each terminal device with a specific set of parameters or operational modes that are tailored to the UE’s individual characteristics, such as processing capacity, memory, dataset collection capability, and other relevant capabilities. These characteristics are typically reported by the terminal device to the network via a capability message. Based on this information, the network can assign configurations such as inference period length, dataset collection window size, or enable or disable certain AI / ML functionalities. This configuration can be highly granular, potentially down to enabling or disabling specific functionalities using a simple indicator, such as a 1 -bit flag, which can be signaled to the terminal device through various layers of network signaling (L1 , L2, or L3). The model ID may be associated with multiple functionality IDs, and each functionality can be assigned a specific configuration depending on the terminal device’s capabilities and the network’s requirements.

[0085] The advantages of this subject matter include the ability to optimize the operation of AI / ML-enabled features on a per-device basis, taking into account the diverse capabilities and roles of different UEs in the network. This allows for more efficient use of network and device resources, improved performance for AI / ML applications, and greater flexibility in supporting a wide variety of device types and use cases, such as those found in automotive or advanced mobile communication scenarios. By enabling device-specific configurations, the network can ensure that each terminal device operates within its optimal parameters, enhancing overall system efficiency and user experience.

[0086] A sixth aspect of the invention provides a computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to carry out the method according to any one of the implementations of the first aspect, or the method according to any one of the implementations of the second aspect.

[0087] The aspect defines a computer-readable storage medium that contains instructions which, when executed by a computer, cause the computer to perform the method as described in any of previous implementations or previous implementations. The subject matter relates to the field of configuring user equipment (UE) in a wireless communication 202404028

[0088] 20 network, particularly in the context of artificial intelligence and machine learning (AI / ML) functionalities.

[0089] A computer-readable storage medium may refer to any physical medium capable of storing data or instructions, such as a hard drive, solid-state drive, optical disc, or other non-transitory storage devices. The instructions stored on this medium are executable by a computer, meaning that when the computer reads and executes these instructions, it performs a specific method.

[0090] The method referenced in previous implementations involves configuring a specific terminal device based on its model ID and functionality ID. The model ID may refer to an identifier that distinguishes different AI / ML models supported or used by the UE. The functionality ID may refer to an identifier that specifies particular AI / ML-enabled features or functionalities that the terminal device can support, as determined during the lifecycle management (LCM) process. The LCM process is used to manage the deployment, update, and removal of AI / ML models and functionalities on the UE.

[0091] In this context, the network, such as a gNB (next-generation NodeB, or base station), determines and signals a specific configuration to the UE. This configuration is tailored according to the UE’s reported capabilities, which may include processing capacity, memory limitations, dataset collection capabilities, and other relevant characteristics. The configuration may also depend on applicability conditions, which may refer to the functionalities that the terminal device is ready or able to apply for model inference.

[0092] The instructions on the storage medium thus enable the computer to perform the steps of receiving capability information from the UE, determining a suitable configuration based on the model ID and functionality ID, and signaling this configuration to the UE. The configuration may include parameters such as inference period length, dataset collection window size, or enabling / disabling specific functionalities, potentially using a simple indicator such as a 1 -bit flag.

[0093] The advantages of this subject matter include the ability to provide device-specific configurations that are optimized for the individual characteristics of each UE. This allows for more efficient and effective use of AI / ML functionalities in the network, particularly in scenarios where UEs have diverse capabilities, mobility profiles, or device types. By tailoring configurations to each UE, the network can ensure that AI / ML features are applied in a manner that matches the UE’s capabilities, leading to improved performance, resource efficiency, and adaptability in next-generation wireless systems. 202404028

[0094] 21

[0095] Furthermore, a system, comprising of at least one terminal device according to the third aspect and a network device according to the fourth aspect, may be defined.

[0096] Although, only described for one terminal device, the application is not limited thereto. The described method may be applied to networks with at least two terminal devices. The described method may also be applied to a network with more than one network device.

[0097] Overall, the solution provides a flexible and scalable way to manage AI / ML model configurations in heterogeneous wireless networks, ensuring that each terminal device operates with settings that are best suited to the terminal’s own capabilities and the network's current requirements. This leads to improved efficiency, adaptability, and performance of AI / ML functionalities in advanced wireless communication systems.

[0098] Figures

[0099] Fig. 1 shows a schematic sequence diagram illustrating the communication in a system according to an embodiment of the present invention.

[0100] Fig. 2 shows a schematic sequence diagram illustrating the communication between terminal devices and a network device according to an embodiment of the present invention.

[0101] Detailed description

[0102] Figure 1 illustrates a schematic sequence diagram comprising a terminal device and a network device according to the present disclosure and optional aspects. The communication between the terminal device (referred to as UE) and network device (referred to as base station) is illustrated in a sequence of message exchanges.

[0103] The terminal device transmits a capability message to the network device. This message is directed from the terminal device to the network device and serves to inform the network device of the terminal device’s capabilities, which may include model ID and functionality ID information relevant to artificial intelligence or machine learning inference functions. 202404028

[0104] 22

[0105] Upon reception of the capability message, the network device processes the provided information and generates a terminal device-specific configuration. This terminal device-specific configuration is tailored to the terminal device. The network device may determine the terminal device-specific configuration on the basis of the terminal device’s capability and the terminal device’s available functionalities, which are taken into consideration with the current state of the communication network which can be assessed by the network device. Based on that information, the network device selects a combination of the functionalities, supported by the terminal device, and indicates the selected functionalities in the terminal device-specific configuration. The terminal device-specific configuration is based on the terminal device’s model ID and functionality ID.

[0106] This specific configuration is then transmitted from the network device back to the terminal device.

[0107] The figure further indicates that optional subsequent operations may be governed by model convergence and / or applicability conditions. Model convergence may refer to the state wherein a global or federated machine learning model has achieved a desired level of accuracy or stability. This may be relevant for the network side.

[0108] Applicability conditions may relate to the operational context or use-case requirements. This may be relevant for the terminal side.

[0109] Following the evaluation of these conditions, the network device may transmit a message to the terminal device to indicate that the terminal device shall enable or disable specific configuration message, also referred to as operation message. This operation message may instruct the terminal device either to activate or to deactivate the previously provided terminal device-specific configuration, depending on the outcome of the model convergence or applicability assessment. Hence, the terminal device may either operate in accordance with terminal device-specific configuration, or in accordance with a default configuration.

[0110] Thereby, the network device may be able to configure the terminal device to use different possible modes of operation.

[0111] Figure 2 illustrates a schematic sequence diagram comprising multiple terminal devices and a network device according to the present disclosure and optional aspects. The 202404028

[0112] 23 communication between two terminal devices (referred to as UE1 and UE2) and network device (referred to as base station) is illustrated in a sequence of message exchanges. The figure depicts that UE1 and UE2, each may be communicating independently with the base station. UE1 may be a high capability UE (e.g. in terms of processing power and / or available memory), and UE2 may be a low capability UE.

[0113] UE1 may transmit a capability message to the base station. This message may be followed by a response from the base station, which sends a UE-specific configuration back to UE1 .

[0114] Then, the same sequence may be performed by UE2, which sends its own capability message to the network device and subsequently receives a UE-specific configuration from the base station.

[0115] The capability message from each UE1 / UE2 may be directed to the base station and contains information necessary for the base station to determine an appropriate configuration tailored to the corresponding specific UE1 / UE2. Upon reception of the capability message, the base station may process the information and may generate a specific configuration, which is then transmitted back to the originating UE.

[0116] Based on the UE-specific configuration, when compared to UE2, UE1 may be enabled to achieve advantages such as higher accuracy, higher dataset threshold, and / or smaller inference window. Based on the UE-specific configuration, when compared to UE1 , UE2 may only achieve lower accuracy, lower dataset threshold, and / or bigger inference window.

[0117] The figure further indicates that optional subsequent operations may be governed by model convergence and / or applicability conditions.

[0118] Following the evaluation of model convergence on the network side and / or applicability conditions for the UE side model, the base station may transmit an operation message to UE1 so that UE1 device shall enable or disable the UE-specific configuration message. This operation message may instruct the UE either to activate or to deactivate the previously provided UE-specific configuration, depending on the outcome of the model convergence or applicability assessment. Hence, the UE may either operate in accordance with UE-specific configuration, or in accordance with a default configuration. Thereby, the base station may be able to configure each UE to use different possible modes of operation. 202404028

[0119] 24

[0120] While the invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered exemplary and not restrictive. The invention is not limited to the disclosed embodiments. In view of the foregoing description and drawings it will be evident to a person skilled in the art that various modifications may be made within the scope of the invention, as defined by the claims.

Claims

20240402825Patent claims1. A communication method for configuring a terminal device, comprising: sending, by the terminal device, a capability message to a network device, the capability message comprising a model identification, ID, based on a machine learning, ML, model of the terminal device, and a functionality ID based on functionalities of an Artificial Intelligence / Machine Learning, AI / ML, inference function of the terminal device; receiving, by the terminal device from the network device, a configuration message, the configuration message comprising a terminal device-specific configuration based on the capability message for configuring the terminal device.

2. The method according to the previous claim, characterized in that the model ID comprises at least one information on a power capability of the terminal device, a processing power of the terminal device, a buffering capability of the terminal device, and / or dataset collection capability of the terminal device.

3. The method according to any of the previous claims, characterized in that the functionality ID comprises at least one information on functionalities of the terminal device such as beam prediction, radio resource management, RRM, measurements, handover capability, channel state information, CSI, prediction and compression, and / or mobility enhancements.

4. The method according to any of the previous claims, characterized in further comprising: receiving, by the terminal device from the network device, an operation message, the operation message indicating the terminal device to operate with the terminal device-specific configuration or with a default configuration, wherein the operation message is based on global model convergence, and / or required use-case and inference reporting.

5. The method according to claim 4, characterized in that the operation message has a length of 1 -bit, and20240402826 the operation message is signaled via any of physical downlink control channel, PDCCH, physical random access channel, PRACH, physical downlink shared channel, PDSCH, radio link control, RLC, system information block, SIB, UE assistance information, UAI, and / or radio resource control, RRC.

6. A communication method for configuring a terminal device, comprising: receiving, by a network device from the terminal device, a capability message, the capability message comprising a model identification, ID, based on a machine learning, ML, model of the terminal device, and a functionality ID based on functionalities of an Artificial Intelligence / Machine Learning, AI / ML, inference function of the terminal device; configuring, by the network device, a terminal device-specific configuration for configuring the terminal device based on the capability message; and sending, by the network device to the terminal device, a configuration message, the configuration message comprising the terminal device-specific configuration.

7. The method according to the previous claim, c h a r a c t e r i z e d i n that the model ID comprises at least one information on a power capability of the terminal device, a processing power of the terminal device, a buffering capability of the terminal device, and / or dataset collection capability of the terminal device.

8. The method according to any of the previous claims, c h a r a c t e r i z e d i n that the functionality ID comprises at least one information on functionalities of the terminal device such as beam prediction, radio resource management, RRM, measurements, handover capability, channel state information, CSI, prediction and compression, and / or mobility enhancements.

9. The method according to claim 8, c h a r a c t e r i z e d i n that the functionality ID indicates a maximum of available functionalities of the terminal device, and the method further comprising: configuring, by the network device, a terminal device-specific configuration based on the capability message comprises associating the model ID to a subset of the maximum of available functionalities of the terminal device.2024040282710. The method according to any of the previous claims, c h a r a c t e r i z e d i n further comprising: sending, by the network device to the terminal device, an operation message, the operation message indicating the terminal device to operate with the terminal device-specific configuration or with a default configuration, wherein the operation message is sent based on global model convergence, and / or required use-case and inference reporting.11 . The method according to claim 10, c h a r a c t e r i z e d i n that the operation message has a length of 1 -bit, and the operation message is signaled via any of physical downlink control channel, PDCCH, physical random-access channel, PRACH, physical downlink shared channel, PDSCH, radio link control, RLC, system information block, SIB, UE assistance information, UAI, and / or radio resource control, RRC.

12. A terminal device configured to carry out the method according to any one of claims 1 to 5.

13. A network device configured to carry out the method according to any one of claims 6 to 11.

14. A computer program product, comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method according to any one of claims 1 to 5, or the method according to any one of claims 6 to 11 .

15. A computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to carry out the method according to any one of claims 1 to 5, or the method according to any one of claims 6 to 11 .

Images