Communication method and apparatus

The method addresses inaccurate ML model inference errors in wireless networks by configuring settings and using predefined events to update ML models, enhancing accuracy and resource efficiency.

JP2024541475AInactive Publication Date: 2024-11-08NEC CORP

Patent Information

Application Number
JP2024531023
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2021-11-24
Publication Date
2024-11-08
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Inaccurate inference results from ML models in wireless communication networks lead to communication failures and wastage of radio resources, necessitating a solution for detecting and addressing these errors.

Method used

A method for detecting inference errors in ML models by transmitting setting configurations associated with ML models, including inference and measurement results, and applying predefined events to trigger updates, using specific radio network identities and configurations for efficient model distribution.

Benefits of technology

Enhances the accuracy of ML model inference by detecting and correcting errors, thereby improving communication performance and optimizing resource utilization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024541475000001_ABST
    Figure 2024541475000001_ABST
Patent Text Reader

Abstract

An exemplary embodiment of the present disclosure relates to an effective mechanism for dealing with discontinuous coverage scenarios. In this solution, a device in a network may obtain at least one setting associated with a machine learning (ML) model. Specifically, the at least one setting indicates at least one first time interval for obtaining both an inference result generated by the ML model and a measurement result corresponding to the ML model. By applying the at least one setting, inference errors can be detected and erroneous inferences can be dealt with successfully.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] Exemplary embodiments of the present disclosure relate generally to the field of communications technology, and more particularly, to methods, apparatus, and media for detecting and addressing inference errors in machine learning (ML) models. [Background technology]

[0002] As communication networks and services grow in size, complexity, and number of users, communications in communication networks may become increasingly complex. To improve communication performance, it has been proposed to use ML / artificial intelligence (AI) techniques in wireless communication networks. In particular, it has been proposed that ML / AI techniques may be used in air interface interactions (i.e., interactions between terminal devices and network devices). In this case, it is expected to efficiently distribute ML / AI models to relevant network elements, including terminal devices and network devices.

[0003] Furthermore, the inference results generated by ML / AI models are not always accurate. Inaccurate inference results may lead to multiple adverse effects, such as communication failure, waste of wireless resources, etc. Therefore, it is desirable to propose a solution to detect and address the inference errors of ML models. Summary of the Invention [Problem to be solved by the invention]

[0004] Overall, the exemplary embodiments of the present disclosure provide a solution for detecting and addressing inference errors in ML models. The embodiments (if any) not included in the claims are to be construed as examples useful for understanding various embodiments of the present disclosure. [Means for solving the problem]

[0005] In a first aspect, a method of communication is provided, the method including: receiving, at a first device, from a second device, at least one setting associated with an ML model, the at least one setting indicating at least one first time interval for obtaining both inference results generated by the ML model and measurement results corresponding to the ML model; and applying the at least one setting.

[0006] In a second aspect, a method of communication is provided, the method including: detecting, in a terminal device, a predefined event indicative of an inference error of an ML model, the predefined event being specifically defined for a particular function on an air interface, and sending, to a network device, a request to update the ML model.

[0007] In a third aspect, a method of communication is provided, the method including receiving, at a first device, one or more ML models from a second device, the one or more ML models being received via at least one of a specific radio network temporary identity (RNTI), a specific control resource set (CORESET), a specific search space set, or one of a group common transmission and broadcast.

[0008] In a fourth aspect, a method of communication is provided, the method including: generating, in a second device, at least one configuration associated with an ML model, the at least one configuration indicating at least one first time interval for obtaining both an inference result generated by the ML model and a measurement result corresponding to the ML model; and transmitting the at least one configuration.

[0009] In a fifth aspect, a method of communication is provided, the method including transmitting, at a second device, one or more ML models to a first device, the one or more ML models being transmitted via at least one of a specific RNTI, a specific CORESET, a specific search space set, or one of a group common transmission and broadcast.

[0010] In a sixth aspect, a first apparatus is provided, the first apparatus comprising a processing unit and a memory coupled to the processing unit storing instructions, the instructions being executed by the processing unit to cause the terminal device to perform the method of the first aspect.

[0011] In a seventh aspect, a terminal device is provided, the terminal device comprising: a processing unit; and a memory coupled to the processing unit storing instructions, the instructions being configured to, when executed by the processing unit, perform the method of the second aspect.

[0012] In an eighth aspect, a first apparatus is provided, the first apparatus comprising a processing unit and a memory coupled to the processing unit storing instructions, the instructions being executed by the processing unit to cause the terminal device to perform the method of the third aspect.

[0013] In a ninth aspect, a second apparatus is provided, the second apparatus comprising a processing unit and a memory coupled to the processing unit storing instructions, the instructions being executed by the processing unit to cause the terminal device to perform the method of the fourth aspect.

[0014] In a tenth aspect, a second apparatus is provided, the second apparatus comprising a processing unit and a memory coupled to the processing unit storing instructions, the instructions, when executed by the processing unit, causing the network device to perform the method of the fifth aspect.

[0015] In an eleventh aspect, there is provided a computer readable medium storing instructions which, when executed on at least one processor, cause the at least one processor to perform a method according to any one of the first to fifth aspects above.

[0016] It should be understood that this summary is not intended to identify key or essential features of the embodiments of the present disclosure, nor to limit the scope of the present disclosure. Other features of the present disclosure will be readily apparent from the following description. [Brief description of the drawings]

[0017] The above and other objects, features and advantages of the present disclosure will become more apparent from the following detailed description of several exemplary embodiments of the present disclosure in the accompanying drawings.

[0018] [Figure 1] FIG. 1 illustrates a proposed functional framework of the ML process.

[0019] [Diagram 2] FIG. 1 illustrates an exemplary communication environment in which exemplary embodiments of the present disclosure may be practiced.

[0020] [Figure 3A] FIG. 1 illustrates the types of AI / ML operations supported by the present disclosure. [Figure 3B] FIG. 1 illustrates the types of AI / ML operations supported by the present disclosure. [Figure 3C] FIG. 1 illustrates the types of AI / ML operations supported by the present disclosure.

[0021] [Figure 4] FIG. 2 is a signaling diagram illustrating a process for communication in accordance with some embodiments of the present disclosure.

[0022] [Figure 5A] FIG. 11 illustrates an exemplary operation timing diagram for a first / second device according to some embodiments of the present disclosure. [Figure 5B] FIG. 11 illustrates an exemplary operation timing diagram for a first / second device according to some embodiments of the present disclosure.

[0023] [Figure 6] FIG. 11 is a diagram showing another exemplary operation timing in a first / second device according to some embodiments of the present disclosure.

[0024] [Figure 7] FIG. 2 illustrates an exemplary method performed by a first device according to some embodiments of the present disclosure.

[0025] [Figure 8] FIG. 2 illustrates an exemplary method performed by a terminal device in accordance with some embodiments of the present disclosure.

[0026] [Figure 9] FIG. 2 illustrates an exemplary method performed by a first device according to some embodiments of the present disclosure.

[0027] [Figure 10] FIG. 10 illustrates an exemplary method performed by a second device according to some embodiments of the present disclosure.

[0028] [Figure 11] FIG. 10 illustrates an exemplary method performed by a second device according to some embodiments of the present disclosure.

[0029] [Figure 12] FIG. 1 is a schematic block diagram of a device suitable for implementing exemplary embodiments of the present disclosure.

[0030] In the drawings, the same or similar reference numbers represent the same or similar elements. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0031] The principles of the present disclosure will now be described with reference to some exemplary embodiments. It should be understood that these embodiments are provided for illustrative purposes only, to assist those skilled in the art in understanding and implementing the present disclosure, and do not imply any limitations on the scope of the present disclosure. The embodiments described herein can be implemented in various ways different from those described below.

[0032] In the following description and claims, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0033] References in this disclosure to "one embodiment," "an embodiment," "an exemplary embodiment," and the like indicate that the described embodiment may include a particular feature, structure, or characteristic, but each embodiment does not necessarily include the particular feature, structure, or characteristic. Moreover, such phrases do not necessarily refer to the same embodiment. Moreover, when a particular feature, structure, or characteristic is described in connection with an embodiment, it is believed to be within the knowledge of one of ordinary skill in the art to affect such feature, structure, or characteristic in connection with other embodiments, whether or not explicitly described.

[0034] It should be understood that although the terms "first" and "second" etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used only to distinguish one element from another. For example, a first element may be termed a second element, and similarly, a second element may be termed a first element, without departing from the scope of the exemplary embodiments. As used herein, the term "and / or" includes any and all combinations of one or more of the listed terms.

[0035] The terms used herein are used only for the purpose of describing particular embodiments and are not intended to limit the exemplary embodiments. As used herein, the singular forms "a," "an," and "said" include the plural forms unless the context clearly indicates otherwise. It should be further understood that, as used herein, the terms "including," "including," "having," "comprising," "having," and / or "having" specify the presence of stated features, elements, and / or components, etc., but do not exclude the presence or addition of one or more other features, elements, components, and / or combinations thereof.

[0036] In some instances, values, procedures, or devices are referred to as "best," "lowest," "highest," "minimum," "maximum," etc. It should be understood that such descriptions are intended to illustrate that selections can be made from among many functional alternatives used, and that such selections are not necessarily better, smaller, higher, or otherwise more preferred than other selections.

[0037] As used herein, the term "communication network" refers to a network conforming to any suitable communication standard, such as New Radio (NR), Long Term Evolution (LTE), LTE-Advanced (LTE-A), Wideband Code Division Multiple Access (WCDMA), High-Speed ​​Packet Access (HSPA), Narrow Band Internet of Things (NB-IoT), etc. Furthermore, communication between terminal devices and network devices in a communication network may be realized according to any suitable generation of communication protocols, including, but not limited to, first generation (1G), second generation (2G), 2.5G, 2.75G, third generation (3G), fourth generation (4G), 4.5G, fifth generation (5G), 5.5G, 5G-Advanced networks, or sixth generation (6G) communication protocols, and / or any other protocols now known or developed in the future. Embodiments of the present disclosure may be applied to various communication systems. In view of the rapid development of communication, there will naturally be future types of communication technologies and systems in which the present disclosure can be embodied, which should not be considered as limiting the scope of the present disclosure only to the aforementioned systems.

[0038] As used herein, the term "terminal device" refers to any device having wireless or wired communication capabilities. Examples of terminal devices include User Equipment (UE), personal computers, desktops, mobile phones, cellular phones, smartphones, personal digital assistants (PDAs), portable computers, tablets, wearable devices, Internet of Things (IoT) devices, Ultra-Reliable Low Latency Communication (URLLC) devices, Internet of Everything (IoE) devices, machine type communication (MTC) devices, vehicle-mounted devices for V2X communications where X stands for pedestrian, vehicle, or infrastructure / network, devices for integrated access and integrated access and backhaul (IAB), satellite-based or airborne vehicles in a non-terrestrial network (NTN) including High Altitude Platforms (HAPs) including satellites and Unmanned Aircraft Systems (UASs), extended reality (XR) devices including different types of reality such as augmented reality (AR), mixed reality (MR), virtual reality (VR), unmanned aerial vehicles (UAVs), which are aircraft without a human pilot and commonly referred to as drones. The term "terminal device" includes, but is not limited to, a vehicle (V2X), a device on a high speed train (HST), or an image capture device such as a digital camera, a sensor game device, a music storage and playback device, or an Internet appliance that allows wireless or wired Internet access and browsing. The "terminal device" may further have "multicast / broadcast" capabilities to support V2X applications, transparent IPv4 / IPv6 multicast distribution, IPTV, smart TV, wireless services, over-the-air software distribution, group communication, and IoT applications that are public safety and mission critical. It may also incorporate one or more Subscriber Identity Modules (SIMs), known as multi-SIMs.The term "terminal equipment" may be used interchangeably with UE, mobile station, subscriber station, mobile terminal, user terminal, or wireless equipment.

[0039] The term "network device" as used herein means a device capable of providing or hosting a cell or coverage in which a terminal device can communicate. Examples of network devices include, but are not limited to, satellites, unmanned aerial systems (UAS) platforms, Node Bs (NodeBs or NBs), evolved Node Bs (eNodeBs or eNBs), next generation Node Bs (gNBs), transmit / receive points (TRPs), remote radio units (RRUs), radio heads (RHs), remote radio heads (RRHs), IAB nodes, femto nodes, pico nodes, and low power nodes such as reconfigurable intelligent surfaces (RISs).

[0040] The terminal device or network device may have the capability of artificial intelligence (AI) or machine learning. Generally, it includes a model trained from a large amount of data collected for a specific function and can be used to predict some information.

[0041] The terminal device or network device may operate on several frequency ranges, such as FR1 (410 MHz-7125 MHz), FR2 (24.25 GHz-71 GHz), frequency bands greater than 100 GHz, and Tera Hertz (THz). It can also operate on licensed / unlicensed / shared spectrum. The terminal device may have two or more connections with the network device under Multi-Radio Dual Connectivity (MR-DC) application scenarios. The terminal device or network device can operate in full duplex, flexible duplex, and cross division duplex modes.

[0042] Embodiments of the present disclosure may be implemented in test equipment, such as, for example, a signal generator, a signal analyzer, a spectrum analyzer, a network analyzer, a test terminal equipment, a test network equipment, a channel emulator, and the like.

[0043] Embodiments of the present disclosure may be performed in accordance with any generation of communication protocols now known or developed in the future, including, but not limited to, first generation (1G), second generation (2G), 2.5G, 2.75G, third generation (3G), fourth generation (4G), 4.5G, fifth generation (5G) communication protocols, 5.5G, 5G-Advanced networks, or sixth generation (6G) networks.

[0044] The term "circuitry" as used herein may refer to a hardware circuit and / or a combination of a hardware circuit and software. For example, a circuit may be a combination of analog and / or digital hardware circuitry and software / firmware. As yet another example, a circuit may be any portion of a hardware processor having software, including a digital signal processor, software, and one or more memories, that cooperate to cause a device, such as a terminal device or a network device, to perform various functions. In yet another example, a circuit may be a hardware circuit and / or a processor, such as a microprocessor or a portion thereof, that requires software / firmware for operation, but the software may not be present if not required for operation. As used herein, the term "circuitry" also includes an implementation of only a hardware circuit or one or more processors, or a portion of a hardware circuit or one or more processors and its (or their) associated software and / or firmware.

[0045] As mentioned above, it is expected that ML / AI techniques may be used for air interface interactions. Specifically, ML models (sometimes referred to as AI models) may be applied to either or both of the terminal device and the network device. Generally speaking, an ML model refers to a data-driven algorithm by applying ML techniques that generates a set of outputs comprising predictive information based on a set of inputs. The ML model may be described and specified in different manners, for example. Characterize the definition stages of AI / ML related algorithms and the associated complexity: Model generation, e.g. model training (including input / output, pre-processing / post-processing (if applicable)), model validation, model testing (if applicable). Inference operations, e.g. input / output, pre-processing / post-processing (if applicable). Identifying different levels of cooperation between UE and gNB, e.g. Uncoordinated: AI / ML algorithms that are only based on realizations, without exchanging information for comparison. Various levels of UE / gNB cooperation targeting separate or combined ML operations. Identify AI / ML model lifecycle management, e.g., model deployment (initiation / configuration), model monitoring, model updates, and model transfer. ●Datasets for training, inference, validation and testing. Identify common notations and terminology for AI / ML-related functions, procedures, and interfaces.

[0046] Furthermore, ML training is required to generate the ML model. For wireless communication scenarios, data used for ML training may be collected from network nodes, management entities, or UEs. An online or offline process may then be performed to train the ML model by learning features and patterns that best present the data and obtain a trained ML model for inference. During the ML inference process, the trained ML model may be used to make predictions or guide decisions based on the collected data and the ML model. Now, referring to FIG. 1, FIG. 1 illustrates a proposed functional framework 100 of the ML process. As shown in FIG. 1, the proposed functional framework 100 includes four important functional blocks, namely, a data collection function 110, a model training function 120, a model inference function 130, and an actor function 140. Details of the above four important functional blocks are described below. The data collection function 110 is a function that provides input data to the model training function 120 and the model inference function 130. AI / ML algorithm-specific data preparation (e.g., data pre-processing and cleaning, formatting, and conversion) is not performed in the data collection function 110. Examples of input data may include measurements from the UE or different network entities, feedback from actors, and output from AI / ML models. Training Data 150: This is the data required as input for the AI / ML model training function 120. Inference data 160: Data required as input for the AI / ML model inference function 130. The model training function 120 is a function that performs ML model training, validation, and testing. The model training function 120 is also responsible for data preparation (e.g., data pre-processing and cleaning, formatting, transformation) based on the training data 150 delivered by the data collection function 110, as necessary. ● Model inference function 130 is a function that provides AI / ML model inference output (i.e., output 170, e.g., prediction or decision). Model inference function 130 is also responsible for data preparation (e.g., data pre-processing and cleaning, formatting, transformation) based on inference data delivered by data collection function 110, as necessary. Output 170: The inference output of the AI / ML model generated by the model inference function 130. Actor functions 140 are functions that receive output from the model reasoning function 130 and trigger or perform corresponding actions. Actor functions 140 may trigger actions on other entities or on themselves. Feedback 180: Training data 150, or inference data 160, or information that may be needed to derive performance feedback.

[0047] It has been agreed that the principles for applying ML / AI technologies to wireless communications include that detailed AI / ML algorithms and models are left to the implementation, the privacy of user data must be protected, and support for AL / ML should reuse existing RAN architecture and not introduce new interfaces.

[0048] In the upcoming 3rd Generation Partnership Project (3GPP) Release 18 (Rel-18), several research and work items will focus on work on AI / ML for the air interface and AI / ML for Next-Generation Radio Access Networks (NG-RAN).

[0049] Recently, there has been some discussion on AI / ML for air interfaces. For example, it is expected that AI / ML may be applied to multiple use cases. One exemplary use case is channel state information (CSI) feedback, e.g., reducing overhead, improving accuracy, prediction. Another exemplary use case is beam management, e.g., beam prediction in time and / or spatial domains for reducing overhead and delay, improving beam selection accuracy. Another exemplary use case is enhancing positioning accuracy for various scenarios with severe non-line of sight (NLOS) conditions. Other use cases include reducing reference signal (RS) overhead, radio resource management (RRM) mobility (e.g., prediction in time or frequency for robustness, reducing interruptions and overhead), etc.

[0050] However, as mentioned above, the inference results generated by the ML / AI model are not always accurate. Inaccurate inference results can result in multiple adverse effects / performance degradation. Similarly, in a scenario of CSI feedback compression, if the recovered CSI is worse than the actual channel, it can cause unnecessary additional resource allocation, while if the recovered CSI is better than the actual channel, it can cause failed decoding. Similarly, in another scenario of CSI prediction (i.e., low-overhead CSI-RS), if the predicted CSI is worse than the actual channel, it can cause unnecessary additional resource allocation, while if the predicted CSI is better than the actual channel, it can cause failed decoding. Additionally, indicating an incorrect beam in beam management can cause low signal-to-noise ratio (SNR) or even beam failure, and incorrect time / frequency compensation in low-overhead demodulation reference signal (DMRS) can cause failed decoding.

[0051] From the above discussion of example scenarios, it becomes apparent that it is desirable to propose a solution for detecting and addressing inference errors in ML models in wireless communication networks.

[0052] Moreover, there are usually two or more models that need to be distributed and many users that need to be configured with the two or more ML models, so it is also expected that the ML / AI models are efficiently distributed to the relevant network elements.

[0053] According to some embodiments of the present disclosure, at least some of the above problems can be solved, but it should be understood that the problems solved by the present disclosure are not limited to the above problems.

[0054] Generally speaking, a cell is associated with a particular frequency resource, which may be further identified by a number of metrics, such as a band, a band combination, a frequency range, a physical cell, a bandwidth part (BWP), a BWP group, a carrier, a component carrier (CC), a carrier group, a CC group, etc. In the following, some embodiments are described with respect to a "cell" / "cell group." It should be understood that such description of a "cell" / "cell group" may apply to a band, a band combination, a frequency range, a physical cell, a BWP, a BWP group, a carrier, a CC, a carrier group, a CC group, etc.

[0055] It should be appreciated that a scenario in which an ML model is used for the air interface would particularly benefit from the exemplary embodiments of the present disclosure. However, that does not mean that the embodiments of the present disclosure are only applicable to this particular scenario. In fact, the embodiments of the present disclosure may be implemented in any scenario. The present disclosure is not limited in this respect.

[0056] In the following text, for the sake of better understanding, the terminal device and the network device are used as examples of the first device and the second device, respectively. It should be understood that, unless otherwise specified, the embodiments described herein may be implemented between any suitable network elements. In particular, either the first device or the second device may be any of the terminal device, the network device, the core network device, or other suitable network elements in wireless communication.

[0057] In the following text, ML-based or ML-assisted, ML-based or ML-assisted beam management, and ML-based or ML-assisted low-overhead RS are described merely as exemplary use cases for better understanding. It should be understood that the embodiments described herein may be applied to any suitable use case. In particular, ●ML-based or ML-assisted CSI feedback may include CSI feedback compression and recovery, limited CSI-RS transmission (in the time domain, frequency domain, spatial domain, antenna port domain, etc.), sounding reference signal (SRS) transmission or joint CSI prediction in the network (NW) and UE based on CSI feedback, CSI prediction in the NW based on limited CSI feedback (in the time domain, frequency domain, spatial domain, antenna port domain, etc.), CSI prediction in the NW based on limited SRS transmission (in the time domain, frequency domain, spatial domain, antenna port domain, etc.), CSI prediction in the UE based on limited CSI-RS transmission (in the time domain, frequency domain, spatial domain, antenna port domain, etc.), etc. ●ML-based or ML-assisted beam management may include combined beam selection or prediction in the NW and UE based on limited CSI-RS transmission, SRS transmission or beam report (in the time domain, frequency domain, spatial domain, antenna port domain, etc.), beam selection in the NW to infer the best beam based on a subset of resources for measurement and reporting, beam selection in the UE to infer the best beam based on limited CSI-RS transmission, beam prediction in the NW based on limited beam report (in the time domain, frequency domain, spatial domain, antenna port domain, etc.), beam prediction in the NW based on limited SRS transmission in the time domain, frequency domain, spatial domain, antenna port domain, etc., beam prediction in the UE based on limited CSI-RS transmission (in the time domain, frequency domain, spatial domain, antenna port domain, etc.), etc. ML-based or ML-assisted low-overhead RS may include low-overhead SSB / CSI-RS / SRS / PTRS / UL DMRS / DL DMRS, etc. (in time domain, frequency domain, spatial domain, antenna port domain, etc.).

[0058] Additionally, other use cases need to be supported, including positioning accuracy, RRM, etc.

[0059] For ease of explanation, some terms and expressions used in the following description are listed below. ●The term "first time interval" refers to the time interval for obtaining both the inference results generated by the ML model and the measurement results corresponding to the ML model, and is sometimes referred to as the "validation time interval." ●The term “second time interval” refers to the time interval for obtaining inference results generated by the ML model, and is sometimes referred to as the “inference-only time interval.” ● The term "inference cycle" refers to a time interval that includes a first time interval, a second time interval, at least one optional other time interval, and at least one optional gap in the time domain, and may also be referred to as a "verification cycle." ● The term "time unit" refers to a unit when describing a length of time, and further, the time unit may be any one of a symbol, a slot, a subframe, a frame, a SFN, a millisecond, etc.

[0060] In this disclosure, some terms may refer to the same or similar physical meaning and may be used interchangeably. Some illustrative examples are given below. ●The terms “ML model”, “AI model” and “algorithm” may be used interchangeably. ● The terms "time interval", "window", "period", "timer" and "interval" may be used interchangeably. ● The terms "function" and "use case" may be used interchangeably. ●The terms "beam selection", "beam measurement", and "beam reporting" may be used interchangeably, and these terms may also refer to "Layer 1 Reference Signal Received Power (L1-RSRP) measurements and / or reports" or "Layer 1 Signal-to-Interference and Noise Ratio (L1-SINR) measurements and / or reports". Example environment

[0061] FIG. 2 illustrates an example communications environment 200 in which example embodiments of the present disclosure may be practiced.

[0062] In the communication environment 100, first devices 210-1 and 210-2 (collectively or individually referred to as first devices 210) and second devices 220-1 and 220-2 (collectively or individually referred to as second devices 220) are shown. The first devices 210 and the second devices 220 may communicate with each other via physical communication channels or links, and some exemplary connections are shown in Figure 2. It should be understood that the connection between the first devices 210 and the second devices 220 may be adjusted according to the actual network deployment.

[0063] In the embodiment of Fig. 2, the first devices 210-1 and 210-2 are shown as terminal devices. The second device 220-1 is shown as a network device serving the first devices 210-1 and 210-2. The service area provided by the second device 220 is referred to as a cell 230. Furthermore, the second device 220-2 is shown as an ML function controller (sometimes referred to as an ML server).

[0064] It should be understood that the number of devices and their connections shown in Figure 2 are provided for illustrative purposes only and do not imply any limitations. Environment 200 may include any suitable number of devices suitable for implementing embodiments of the present disclosure. It should also be understood that, although not shown, communication environment 200 may include other network elements, such as other core network devices, other application servers, etc.

[0065] It should also be appreciated that in some instances, communication environment 200 may include only homogeneous network deployments or only heterogeneous network deployments.

[0066] Communications in the communication environment 200 may conform to any suitable standard, including, but not limited to, Long Term Evolution (LTE), LTE-Evolution, LTE-Advanced (LTE-A), Wideband Code Division Multiple Access (WCDMA), Code Division Multiple Access (CDMA), Global System for Mobile Communications (GSM), and the like. Furthermore, communications may be performed according to any currently known or future developed generation of communications protocols. Examples of communications protocols include, but are not limited to, first generation (1G), second generation (2G), 2.5G, 2.75G, third generation (3G), fourth generation (4G), 4.5G, fifth generation (5G), 5.5G, 5G-Advanced networks, or sixth generation (6G) communications protocols.

[0067] Additionally, ML techniques may be applied in the communication environment 200. Furthermore, different types of AI / ML operations may be supported in the communication environment 200. Reference is now made to FIGS. 3A-3C, which show three exemplary types of AI / ML operations 300, 320, and 340 supported by the present disclosure. In the specific example of FIG. 3A, the AI / ML operations are divided between different AI / ML endpoints (i.e., end devices, e.g., the first device 210 as shown in FIG. 2, and network AI / ML endpoints, e.g., the second device 220 as shown in FIG. 2). Specifically, the AI / ML operations / models are divided into multiple parts depending on the current task and environment. The intention is to offload the computationally intensive and energy intensive parts to the network endpoints, but leave the privacy-sensitive and delay-sensitive parts at the end devices. The devices execute the operations / models up to a certain part / layer, and then send intermediate data to the network endpoints. The network endpoints execute the remaining parts / layers and feed back the inference results to the devices.

[0068] Another type of supported AI / ML operation is shown in FIG. 3B. In the example of FIG. 3B, a multifunction mobile device (e.g., the first device 210 as shown in FIG. 2) may need to switch AI / ML models depending on the changes in the task and environment. The condition for adaptive model selection is that the model to be selected is available to the mobile device. However, considering the situation where AI / ML models are increasingly diverse and the limited storage resources in the UE, it may be decided not to pre-load all candidate AI / ML models on-board. When an AI / ML model is required in the mobile device to adapt to the changed AI / ML task and environment, online model distribution (i.e., downloading of new models) is necessary, which can distribute the AI / ML model from a NW endpoint (e.g., the second device 220 as shown in FIG. 2) to the mobile device. For this purpose, the model performance in the UE needs to be continuously monitored.

[0069] Another type of supported AI / ML operation is shown in FIG. 3C. In the example of FIG. 3C, the AI / ML server 350 (e.g., a cloud server) trains a global model by integrating the local models partially trained by each end device. In each training iteration, the UE (e.g., the first device 210 as shown in FIG. 2) performs training based on the model downloaded from the AI / ML server 350 using local training data. The UE then notifies the AI / ML server 350 of intermediate training results, e.g., via a fifth generation (5G) uplink channel. The AI / ML server 350 integrates intermediate training results from the UE and updates the global model. The updated global model is then distributed back to the UE, so that the UE can perform training for the next iteration.

[0070] It should be understood that the above types of AI / ML operations are presented for illustrative purposes only and do not imply any limitation, and implementations of the present disclosure are not limited to any particular types of AI / ML operations. Process Example

[0071] Although features / operations are described separately in particular exemplary embodiments, it should be understood that these features / operations described in different exemplary embodiments may be used in any suitable combination, unless expressly indicated to the contrary.

[0072] It is clear that the first device 210 and the second device 220 described herein are counterparts to each other. Specifically, both the first device 210 and the second device 220 may exchange capability-related information, receive one or more ML models, apply one or more settings for the ML models, execute one or more ML models, generate inference results by the ML models, generate measurement results corresponding to the ML models, detect inference states of the ML models, exchange status information of the ML models, and initiate and / or execute updated procedures for the ML models. Furthermore, the above procedures / operations performed by the first device 210 and the second device 220 are similar and corresponding. Therefore, although some discussions have been given with respect to the first device 210 or the second device 220, these discussions may be applied to the other device of the first device 210 and the second device 220 as well. For the sake of brevity, some identical or similar contents are omitted.

[0073] Additionally, in the following description, some interactions occur between devices (e.g., the first device 210 and the second device 220, or the second device 220-1 and another second device 220-2). It should be understood that the interactions may be realized within one signaling / message or multiple signaling / messages, including system information, radio resource control (RRC) messages, DCI messages, uplink control information (UCI) messages, MAC CE, etc. The present disclosure is not limited in this respect.

[0074] Additionally, in the following exemplary embodiments, messages (including first message, second message, third message, fourth message, trigger message, activation message, deactivation message, and other messages) may be sent to either the first device 210 or the second device 220. It is apparent that any of the messages described in this disclosure may be sent specifically for a particular cell, a particular group of cells, a particular first device 210, a particular group of first devices, a particular function, a particular level of coordination, a particular architecture, a particular algorithm, and the like.

[0075] The principles and embodiments of the present disclosure will be described in detail below with reference to Fig. 4 showing a signaling chart illustrating a communication process 400 according to some embodiments of the present disclosure. For illustrative purposes, the process 400 will be described with reference to Fig. 2. The process 400 may involve a first device 210 and a second device 220.

[0076] Optionally, the first device 210 may communicate the capability-related information with the second device 220 so that the second device 220 can perform efficient ML model distribution and generate more suitable settings for the associated ML model. As shown in FIG. 4, the first device 210 transmits 410 the capability-related information to the second device 220.

[0077] An example of the capability-related information is whether the first device 210 supports detecting an inference error of an ML model. For example, the capability-related information indicates whether the first device 210 can support the introduced first time interval, second time interval, and / or ML model inference cycle.

[0078] Another example of the capability-related information is the number of ML models configured in the first device 210. For example, the capability-related information is the capability of the first device 210 with respect to multiple ML models configured / applied simultaneously.

[0079] Another example of capability-related information is information about the ML models supported by the first device 210 .

[0080] In one particular exemplary embodiment, the capability-related information is information regarding an ML model cooperation level supported by the first device 210, and the ML model cooperation level may be any of the cooperation levels including no cooperation, network (NW) only, UE only, node-to-node assistance, combined ML operation, split operation, federated learning (FL), downloadable, etc. In some embodiments, the ML model cooperation level "no cooperation" includes at least a non-cooperative framework in which one or more AI / ML algorithms are simply implemented without requiring improvements on air interface changes, and a non-cooperative framework with a modified air interface corresponding to the AI / ML algorithms based on the efficient implementation.

[0081] Alternatively, in some embodiments, when the ML model coordination level "node-to-node assistance" applies, either the UE (e.g., first device 210) or the network device (e.g., second device 220) gets assistance (for training, adaptation, etc.) from the other of the UE and the network device. The ML model coordination level "node-to-node assistance" does not require model exchange between the UE and the network device. Alternatively, in some embodiments, the ML model coordination level "coupled ML operations" requires exchange of ML models and / or instructions between the UE and the network device.

[0082] In another particular exemplary embodiment, the capability-related information is information regarding the ML model architecture supported by the first device 210. In particular, the ML model architecture may be associated with the number of layers or maximum number of layers, the number of hidden layers or maximum number of hidden layers, layer type, layer shape (i.e., filter size, number of channels / filters), the number and maximum number of neurons per layer, the number and maximum number of neurons, connections between layers, etc. For example, the ML model may be a multilayer perceptron (MLP), a convolution neural network (CNN), a recurrent neural network (RNN), etc.

[0083] In another particular exemplary embodiment, the capability-related information is information regarding ML features supported by the first device 210, where the ML features may be ML-based or ML-assisted CSI feedback, beam management, low overhead RS, positioning, mobility, etc.

[0084] In some embodiments, ML-based or ML-assisted CSI feedback may include CSI feedback compression and recovery, limited CSI-RS transmission (in the time domain, frequency domain, spatial domain, antenna port domain, etc.), sounding reference signal (SRS) transmission or joint CSI prediction in the NW and UE based on CSI feedback, CSI prediction in the NW based on limited CSI feedback (in the time domain, frequency domain, spatial domain, antenna port domain, etc.), CSI prediction in the NW based on limited SRS transmission (in the time domain, frequency domain, spatial domain, antenna port domain, etc.), CSI scheduling in the UE based on limited CSI-RS transmission (in the time domain, frequency domain, spatial domain, antenna port domain, etc.).

[0085] Alternatively, in some embodiments, ML-based or ML-assisted beam management may include combined beam selection or prediction at the NW and UE based on limited CSI-RS transmission, SRS transmission or beam report (in the time domain, frequency domain, spatial domain, antenna port domain, etc.), beam selection at the NW to infer a best beam based on a subset of resources for measurement and reporting, beam selection at the UE to infer a best beam based on limited CSI-RS transmission, beam prediction at the NW based on limited beam report (in the time domain, frequency domain, spatial domain, antenna port domain, etc.), beam prediction at the NW based on limited SRS transmission in the time domain, frequency domain, spatial domain, antenna port domain, etc., beam prediction at the UE based on limited CSI-RS transmission (in the time domain, frequency domain, spatial domain, antenna port domain, etc.).

[0086] Alternatively, in some embodiments, the ML-based or ML-assisted low-overhead RS may include low-overhead SSB / CSI-RS / SRS / PTRS / UL DMRS / DL DMRS (in the time domain, frequency domain, spatial domain, antenna port domain, etc.).

[0087] Yet another example of the capability-related information is information about ML computing capability. Specifically, the ML computing capability may be specified by a total ML computing capability, an ML computing capability of each ML model, and the like.

[0088] Other performance related information may directly indicate a particular index / type of ML model, an ML model for a particular algorithm, etc. Additionally, an ML model for a particular algorithm may be associated with learning rate, loss function, cost function, activation function, mini-batch size, number of training iterations, momentum, number of hidden units, weight decay, activation sparsity, nonlinearity, weight initialization, regularization constant, number of epochs, number of branches in a decision tree, number of clusters in a clustering algorithm, any other hyperparameters.

[0089] Similarly, the second device 220 may provide similar capability-related information to the first device 210. Furthermore, the second device 220 may provide its capability-related information to another second device 220. For example, the second device 220-1 (i.e., a network device) transmits its capability-related information to another second device 220-2 (i.e., an ML function controller).

[0090] The capability-related information of the second device 220 is similar to the capability-related information of the first device 210. For the sake of brevity, the same or similar content will be omitted here.

[0091] It should be understood that the above examples of capability-related information are provided for illustrative purposes only and do not imply any limitation. In some other exemplary embodiments, any suitable capability-related information may be exchanged between associated devices (e.g., between any two of first devices 210-1 and 210-2 and second devices 220-1 and 220-2). This disclosure is not limited in this respect.

[0092] Additionally, the capability related information may be carried within any suitable signaling / message, including but not limited to system information, RRC messages, DCI messages, UCI messages, MAC CE, and the like.

[0093] By exchanging capability-related information, the second device 220 (a network device or an ML function controller) may perform efficient ML model distribution and generate more appropriate configurations for the ML models.

[0094] Furthermore, according to some embodiments, an improved ML distribution procedure is proposed. In some embodiments, one or more ML models may be distributed within a network. In the particular example of FIG. 4, the second device 220 transmits (430) the one or more ML models to the first device 210.

[0095] Additionally, in some embodiments, more than one ML model may be sent to the first device 210 if the first device 210 supports multiple ML models being configured / applied simultaneously.

[0096] In some embodiments, the one or more ML models may be transmitted in a specific RNTI, i.e., one or more RNTIs may be defined / introduced specifically for ML model distribution. Additionally, in one embodiment, a new group-specific common RNTI may be defined / introduced.

[0097] Alternatively or additionally, in some embodiments, the one or more ML models may be transmitted with a specific scrambling ID, i.e., one or more scrambling IDs may be defined / introduced specifically for ML model distribution. Additionally, in one particular exemplary embodiment, a scrambling ID may be configured for PDCCH / PDSCH / PUCCH / PUSCH and / or each DMRS.

[0098] Alternatively or additionally, in some embodiments, the one or more ML models are transmitted in a specific RRC message, a specific MAC CE logic channel ID (LCID).

[0099] Alternatively or additionally, in some embodiments, the one or more ML models are transmitted via group common transmission or broadcast.

[0100] Alternatively or additionally, in some embodiments, the one or more ML models are transmitted via any or combination of a physical downlink control channel (PDCCH), a PDCCH scrambled with a specific RNTI or a specific scrambling ID, a group-common PDCCH, a physical downlink shared channel (PDSCH), a PDSCH scrambled with a specific RNTI or a specific scrambling ID, a group-common PDSCH, a paging message, system information, a short message, etc. Alternatively, in some other embodiments, the one or more ML models are transmitted via any or combination of a physical uplink control channel (PUCCH), a PUCCH scrambled with a specific RNTI or a specific scrambling ID, a physical uplink shared channel (PUSCH), a PUSCH scrambled with a specific RNTI or a specific scrambling ID, etc.

[0101] Thus, using group common transmission or broadcasting improves the efficiency of the ML model distribution procedure, since the NW (eg, the second device 220) is unlikely to train an individual ML model for each UE.

[0102] Alternatively, in some other embodiments, the one or more ML models are transmitted via unicast, thereby enabling UE-specific distribution of the ML models.

[0103] Alternatively or additionally, in some embodiments, the one or more ML models are monitored over a specific radio resource configuration, such as a specific CORESET, or a specific set of search spaces, such that only a specific search space or CORESET needs to be monitored to obtain a schedule for ML model distribution.

[0104] In some embodiments, the multiple transmitted ML models may be used for the same function / use case. Alternatively, the multiple transmitted ML models may be used for different functions / use cases.

[0105] In some embodiments, the multiple ML models may be listed and indexed. Specifically, each of the multiple ML models may be assigned an index / identity. In some embodiments, the index / identity may be in ascending or descending order for a particular cell, a particular group of cells, a particular first device 210, a particular group of first devices, or a particular function.

[0106] In some embodiments, the one or more ML models themselves may be transmitted. Alternatively, the one or more ML models may be represented to the relevant network using at least one parameter / information associated with the one or more ML models, such as ML model architecture, algorithm, hyperparameters, etc. Alternatively, the one or more ML models may be represented to the relevant network using an index / identity of the one or more ML models.

[0107] Additionally, the one or more ML models may be conditionally transmitted. Specifically, in some embodiments, the one or more ML models are transmitted only if the communication conditions between the first device 210 and the second device 220 satisfy a preset condition. The communication conditions may be identified by Quality of Service (QoS) requirements (e.g., rate, modulation and coding scheme (MCS), bandwidth, block error rate (BLER), delay, and any other suitable QoS requirements). In one exemplary embodiment, the ML models are distributed only if certain QoS requirements are satisfied, thus ensuring successful ML model distribution.

[0108] If the second device 220 needs to receive the ML model, the second device 220 may perform a similar procedure as described for the first device 210.

[0109] Additionally, one or more settings for the one or more ML models may be transmitted to either the first device 210 or the second device 220. In some embodiments, the one or more settings may be specific to a particular cell, a particular cell group, a particular first device 210, a particular first device group, or a particular function, thus allowing for flexible configuration / activation of the one or more ML models.

[0110] The trigger procedure, activation / deactivation procedure and configuration procedure may be realized by a single message / signaling or multiple separate messages / signaling, including system information, RRC, MAC-CE, UCI, DCI, etc. Additionally, in some embodiments, the message / signaling may indicate the specific ML model index / identity. Alternatively, some default rules may be defined for applying the one or more configurations. For example, the ML model with the lowest index may be considered as the ML model applied by default.

[0111] It should be understood that the transmission of the one or more messages / signaling for triggering, activation / deactivation and configuration may be performed in a manner similar to ML model distribution. For example, the one or more messages / signaling may be transmitted / monitored with a specific RNTI, a specific scrambling ID, a specific CORESET, a specific search space set, a specific MAC CE LCID, etc. For brevity, details for transmitting the one or more messages / signaling are omitted here.

[0112] Additionally, in some embodiments, the one or more messages / signaling may be generated based on the capability-related information described above. Alternatively, in some other exemplary embodiments, the one or more messages / signaling may be generated independently of the capability-related information described above.

[0113] As mentioned above, it is desirable to detect ML inference errors to avoid performance degradation. In other words, it is expected that each ML model applied should be verified. According to some embodiments, the detection of ML inference errors may be realized by setting at least one first time interval during which a device (either the first device 210 or the second device 220) can obtain both actual measurement results and inference results. In other words, the first time interval is used to monitor the performance of the ML model. Inference errors may be determined / detected by comparing the actual measurement results and the inference results.

[0114] As shown in Fig. 4, the first device 210 obtains (420-1) at least one setting associated with an ML model, the at least one setting indicating at least one first time interval for obtaining both an inference result generated by the ML model and a measurement result corresponding to the ML model. In the particular example of Fig. 4, the first device 210 receives (421) the at least one setting from a second device. In some embodiments, the second device 220 may be a network device. Alternatively, in some other embodiments, the second device 220 may be an ML function controller.

[0115] In response, the second device 220 may obtain (420-2) the corresponding at least one configuration. As one particular exemplary embodiment, the second device 220 (which may be a network device or an ML function controller) generates the corresponding at least one configuration by itself. Alternatively, in another particular exemplary embodiment, the second device 220 receives the at least one configuration from another second device 220. For example, the second device 220-1 (e.g., a network device) receives the at least one configuration from the second device 220-2 (e.g., an ML function controller).

[0116] In some embodiments, the at least one first time interval is aperiodic or semi-periodic. For example, one or more time intervals after applying a configuration may be set as the at least one first time interval. In some other embodiments, the at least one first time interval starts in response to a trigger message or an activation message. In some further embodiments, the at least one first time interval stops in response to a deactivation message. Additionally, the trigger message may be a DCI or a UCI, and the activation / deactivation message may be a MAC CE.

[0117] Alternatively, in some other embodiments, the at least one first time interval is periodic. In this case, the at least one setting may indicate some parameters used to determine the at least one first time interval. One exemplary parameter is the period of the at least one first time interval. Another exemplary parameter is the first length (e.g., X1 time units) of each of the at least one first time interval, specifically, the time unit may be any of symbols, slots, subframes, frames, SFNs, milliseconds, etc.

[0118] Further, in some embodiments, the first length may be expressed as a value of a first timer. Additionally, another exemplary parameter is information for determining a starting point for applying the at least one setting.

[0119] In some embodiments, the information for determining the starting point may be a system frame number (SFN) or a slot number, for example, the starting point may be SFN0 and slot0. Furthermore, a starting offset may be set. In one specific embodiment, the starting offset may be determined by (SFN*10+subframe number) mod (length of ML model inference cycle).

[0120] Alternatively, in some other embodiments, each of the at least one first time interval is part of an inference cycle. Referring now to Figures 5A and 5B, first time intervals 510-1 and 510-2 are part of inference cycles 550-1 and 550-2, respectively.

[0121] Additionally, in addition to the first time interval, each inference cycle may include a second time interval used to obtain an inference result. In one exemplary embodiment, the second time interval is after the first time interval in the inference cycle. As shown in FIG. 5A, second time intervals 520-1 and 520-2 are shown after the respective first time intervals 510-1 and 510-2, respectively. That is, first validate the ML model and then apply the model inference for a certain time. In some other embodiments, a first gap in the time domain may be defined or set between the first and second time intervals. Additionally, the length of the first gap (e.g., X3 time units, where X3 may be 0, 1, 2, ...) is associated with the capabilities of the first device 210 and / or the second device 220. Alternatively, the length of the first gap may be defined / set as a default value (e.g., 0) defined / set / defined by a wireless standard, a network operator, or a service provider.

[0122] Alternatively, in another embodiment, the second time interval is before the first time interval in the inference cycle. As shown in FIG. 5B, the second time intervals 520-1 and 520-2 are shown before the respective first time intervals 510-1 and 510-2, respectively. That is, the model inference is first applied for a certain time, and then the performance is confirmed. Additionally, in some other embodiments, a second gap in the time domain can be defined or set between the second time interval and the first time interval. Furthermore, the length of the second gap (e.g., X4 time units, where X4 may be 0, 1, 2, ...) is associated with the capabilities of the first device 210 and / or the second device 220. Alternatively, the length of the second gap may be specified / set as a default value (e.g., 0) defined / set / defined by a wireless standard, a network operator, or a service provider.

[0123] Additionally, a validation time interval may be set for an initial period. As shown in FIG. 5B, a validation time interval 530 is initially set. During this validation time interval, the ML model may be validated using an inference error detection procedure as described below. In some other embodiments, a third gap in the time domain may be defined or set between the validation time interval and the second time interval. Additionally, the length of the third gap (e.g., X5 time units, where X4 may be 0, 1, 2, ...) is associated with the capabilities of the first device 210 and / or the second device 220. Alternatively, the length of the third gap may be specified / set as a default value (e.g., 0) defined / set / specified by a wireless standard, a network operator, or a service provider.

[0124] Additionally, several parameters may be used to indicate the inference cycle, the first time interval, and / or the second time interval. One exemplary parameter is a period of the inference cycle. Another exemplary parameter is a first length of the first time interval as described above. Another exemplary parameter is a second length of the second time interval (e.g., X2 time units). Another exemplary parameter is a length of the inference cycle (e.g., X time units). Furthermore, in some embodiments, the second length may be expressed as a value of a second timer. In some other embodiments, the length of the inference cycle may be expressed as a value of a third timer. Another exemplary parameter may be information for determining a starting point for applying the at least one setting.

[0125] It should be understood that the above example parameters indicated by the at least one setting are provided for illustrative purposes only. Any of the above example parameters may be set as a default value. If so, it is not necessary to indicate the parameter. Furthermore, any other suitable parameters may be used to indicate the first time interval, the second time interval, and the inference cycle. The disclosure is not limited in this respect.

[0126] Additionally, in some embodiments, in addition to the first time interval and the second time interval, the inference cycle may further include a fourth time interval used for ML model training, a fifth time interval used for ML model testing, a sixth time interval used for ML model switching, etc.

[0127] Additionally, inference cycles / other time intervals (including a second time interval, a third time interval, a fourth time interval, etc.) may be set / indicated similarly to the first time interval.

[0128] For example, both the inference cycle and other time intervals may be set / generated by either the first device 210 or the second device 220.

[0129] Additionally, the inference cycle / other time interval may be aperiodic or semi-periodic. Specifically, one or more time intervals after applying the configuration may be set as the inference cycle / other time interval. In some other exemplary embodiments, the inference cycle / other time interval may be started in response to a trigger message or an activation message. In some other exemplary embodiments, the inference cycle / other time interval may be stopped in response to a deactivation message. Additionally, the trigger message may be a DCI or a UCI, and the activation / deactivation message may be a MAC CE.

[0130] Alternatively, the inference cycle / other time interval may be periodic. In this case, the at least one setting may indicate some parameters used to determine the inference cycle / other time interval, such as a length of each of the inference cycle / other time intervals, a period of each of the inference cycle / other time intervals, and other suitable parameters. Furthermore, in some embodiments, the respective lengths may be expressed as a value of a first timer. Additionally, another exemplary parameter is information for determining a starting point for applying the at least one setting.

[0131] For the sake of brevity, more details regarding inference cycles / other time intervals are omitted here.

[0132] Alternatively or additionally, both the inference cycles and the time intervals (i.e., the first time interval, the second time interval, the third time interval, the fourth time interval, etc.) may be configured to have respective starting points. Specifically, the inference cycle is configured to have a starting point T1, the first time interval is configured to have a starting point T2, the second time interval is configured to have a starting point T3, ...the nth time interval is configured to have a starting point T4, ...the nth time interval is configured to have a starting point T5, ...the nth time interval is configured to have a starting point T6, ...the nth time interval is configured to have a starting point T7, ...the nth time interval is configured to have a starting point T8, ...the nth time interval is configured to have a starting point T9, ...the nth time interval is configured to have a starting point T10, ...the nth time interval is configured to have a starting point T11, ...the nth time interval is configured to have a starting point T12, ...the nth time interval is configured to have a starting point T13, ...the nth time interval is configured to have a starting point T14, ...the nth time interval is configured to have a starting point T15, ...the nth time interval is configured to have a starting point T16, ...the nth time interval is configured to have a starting point T17, ...the nth time interval is configured to have a starting point T18, ...the nth time interval is configured to have a starting point T19 ...2, ...the nth time interval is configured to have a starting point T14, ...the nth time interval is configured to have a starting point T15, ...the nth time interval is configured to have a starting point T16, ...the nth time interval is configured to have a n+1 Furthermore, some of the inference cycles and time intervals may share a common starting point. For example, an inference cycle may be set to have a starting point T aThe first and second time intervals are set to have a starting point T b The third and fourth time intervals are set to have a starting point T c Additionally, the deactivation operations for the inference cycle and the time interval are similar. In summary, either the inference cycle or the time interval may be started and stopped separately. Thus, a more flexible setting is realized.

[0133] In some other embodiments, the at least one configuration may include different configurations that may be associated with either the first time interval or the second time interval. For example, multiple linked configurations may be used in the first time interval and the second time interval. Additionally, the linking of the at least one configuration may be explicitly set by pairing multiple configurations or implicitly set by associating IDs of the configurations. Specifically, the at least one configuration may be a reporting configuration, a measurement configuration, an RS configuration, a transmission configuration, etc.

[0134] Further, each of the different configurations may include a set of associated parameters, which may be defined for a particular function / use case, and the parameters within the different sets may differ.

[0135] In some other embodiments, the at least one configuration includes a first configuration (referred to as "configuration_1") used to obtain measurement results corresponding to an ML model and a second configuration (referred to as "configuration_2") used to obtain inference results generated by the ML model. Further, in some embodiments, the first configuration is unrelated to any ML model, e.g., the first configuration is a normal / traditional measurement configuration.

[0136] For a function / use case of CSI compression and recovery, the at least one configuration may include a reporting amount of CSI reporting, a number of configured bits of CSI reporting, etc. For example, the first configuration may include a first CSI, an amount of CSI report sets as a number of configured bits M for compressed CSI, etc., and the second configuration may include a second CSI, an amount of CSI report sets as a number of configured bits N for compressed CSI, etc. Furthermore, in some embodiments, the number of configured bits M in the first configuration is greater than the number of configured bits N in the second configuration. Alternatively or additionally, the first configuration may include a first CSI, an amount of CSI report sets as at least one of a high-resolution CSI, a Type-I CSI, a Type-II CSI, a rank indicator (RI), a precoding matrix indicator (PMI), a CQI, a layer indicator (LI), etc.

[0137] Alternatively, for a beam management function / use case, the at least one configuration may include a resource set for beam selection, a periodicity of beam measurement and reporting, etc. For example, a first configuration may include a first resource set for beam selection, a periodicity of beam management RS, a reporting period, etc., and a second configuration may include a second set for beam selection, etc. Furthermore, in some embodiments, the first set for beam selection included in the first configuration includes more resources than the second set for beam selection included in the second configuration.

[0138] Alternatively, for low overhead RS features / use cases, the at least one configuration may include an RS pattern, a frequency density, a periodicity, a number of antenna ports, a transmit power or energy per resource element (EPRE), etc. For example, a first configuration may include a first RS having a first RS pattern, a first RS having a first frequency density, a first periodicity, a first number of antenna ports, etc., and a second configuration may include a second RS having a second RS pattern, a second RS having a second frequency density, a second periodicity, a second number of antenna ports, etc. Furthermore, in some embodiments, the first frequency density of the first configuration is greater than the second frequency density of the second configuration, the first periodicity is less than the second periodicity, and the first number of antenna ports is greater than the second number of antenna ports.

[0139] In one particular example embodiment, either the first device 210 or the second device 220 may apply both the first setting and the second setting during at least a first time interval, while applying the second setting beyond at least one first time interval (e.g., within a second time interval), i.e., during the first time interval, the first device 210 and / or the second device 220 need to perform more measurements and reports according to the first setting.

[0140] The at least one setting may then be applied, i.e., inference and error detection may be performed. As shown in FIG. 4, the first device 210 and the second device 220 apply the at least one setting (440-1 and 440-2), respectively.

[0141] In some embodiments, the at least one setting may be applied according to a starting point indicated by the at least one setting as described above. Alternatively, in some other embodiments, the at least one setting may be applied from a determined starting point depending on a particular operation. In one particular exemplary embodiment, the starting point may be a time point of acknowledging receipt of the ML model distribution, and the starting offset may be a specific value (e.g., a first length of a first time interval, X1). Specifically, the first device 210 applies the at least one setting for a certain period of time (e.g., a length of a first time interval) after sending an ACK for receiving the ML model distribution. In another particular exemplary embodiment, the starting point may be a time point of sending or receiving information regarding the validation of the ML model, and the starting offset may be a specific value. In another particular exemplary embodiment, the second device 220 may send a trigger message to the first device 210. The first device 210 may apply the at least one setting for a certain period of time after receiving the trigger message.

[0142] For better understanding, the following describes one specific exemplary embodiment for applying the at least one setting, including the above first setting and second setting.

[0143] During the second time interval, configuration_2 is activated. Both the first device 210 and the second device 220 perform transmission and reception according to configuration_2 and / or the inferred result obtained based on configuration_2. For the CSI compression and recovery function / use case, the first device 210 and the second device 220 exchange the compressed version of CSI corresponding to the second device 220. Additionally, the first device or the second device adopts the restored version of CSI. Alternatively, for the beam management function / use case, the first device 210 and the second device 220 adopt and / or exchange the inferred best beam corresponding to the best beam of the second set. Alternatively, for the low overhead RS function / use case, the first device 210 and the second device 220 adopt and / or exchange the inferred channel estimation result corresponding to the result obtained by measuring the second RS.

[0144] During a first time interval, both configuration_1 and configuration_2 are activated. The first device 210 and the second device 220 transmit and receive according to configuration_1 and configuration_2, and the first device 210 and / or the second device 220 perform a verification procedure according to the actual measurement results (based on configuration_1) and the ML model inference results (based on configuration_2), which are described in the following text.

[0145] Alternatively, during the first time interval, configuration_1 is activated and configuration_2 is deactivated, and both the first device 210 and the second device 220 perform transmission and reception according to configuration_1 and / or measurement results obtained based on configuration_1.

[0146] Furthermore, to reduce power, the first device 210 may operate in several power saving modes, such as a discontinuous reception (DRX) mode. In accordance with some of the exemplary embodiments of the present disclosure, several further improvements to the power saving modes are proposed, as described below.

[0147] In the first time interval, power consumption and complexity are higher, which makes the first time interval unsuitable to be set within the inactive period of DRX (i.e., non-DRX ON time interval). Therefore, in some embodiments, when the first device 210 operates in DRX mode, the inference cycle is aligned to the DRX cycle of the DRX mode. Also, in some embodiments, the first time interval is aligned to the DRX ON time interval. In some embodiments, the second device 220 may generate aligned configurations for the DRX mode and the ML model, respectively.

[0148] Alternatively, when the first device 210 operates in DRX mode, the inference results and the measurement results may be obtained only during the active time intervals of the DRX cycle in the DRX mode. That is, the first device 210 performs the verification only during the intersection of the DRX ON time interval and the at least one configured first time interval. Thus, when the first device 210 operates in DRX mode, the first time intervals are skipped during the inactive time intervals of the DRX cycle.

[0149] Optionally, in some embodiments, an initial validation procedure may be performed to select an ML model from a plurality of ML models when multiple ML models are applied by each device. In the particular exemplary embodiment of Figures 5A and 5B, the initial validation procedure may be performed during a first time interval 510-1 as shown in Figure 5A and a validation time interval 530 as shown in Figure 5B.

[0150] In one particular exemplary embodiment, if multiple ML models are configured in the first device 210, the first device 210 may select an ML model from the multiple ML models and send a first message indicating the selected ML model to the second device 220. Similarly, if multiple ML models are configured in the second device 220, the second device 220 may also select an ML model from the multiple ML models and send a message indicating the selected ML model to the first device 210.

[0151] Further, when applying at least the configuration, either the first device 210 or the second device 220 may generate measurement and / or inference results themselves and exchange the measurement and / or inference results. Additionally, either the first device 210 or the second device 220 may determine an inference state of the ML model. According to some embodiments, relevant information may be exchanged between the devices so that the inference states of the ML models can be determined and aligned with each other.

[0152] 4, the first device 210 and the second device 220 exchange relevant information. The following text describes some exemplary exchange procedures.

[0153] Any of the following exchange procedures are optional, and the procedures shown below may be performed independently or in any suitable combination. For example, if the first device 210 transmits an actual measurement result to the second device 220, the second device 220 may not necessarily transmit the actual measurement result to the first device 210. Similarly, if the first device 210 transmits a first indication to the second device 220 indicating whether an inference error of the ML model has occurred, the second device 220 may not necessarily transmit a corresponding second indication to the first device 210.

[0154] In short, the actual exchange procedure may be performed as desired and according to the particular application scenario, although the present disclosure is not limited in this respect.

[0155] For example, during a first time interval, the first device 210 obtains a first inference result generated by the ML model and a first measurement result (i.e., an actual measurement result) corresponding to the ML model, and transmits the first inference result and the first measurement result to the second device 220.

[0156] Alternatively or additionally, in some embodiments, the first device 210 may transmit a first difference between the first inference result and the first measurement result to the second device 220.

[0157] Alternatively or additionally, in some embodiments, the first device 210 may determine an inference state of the ML model (as described below) and send a first indication to the second device 220 indicating whether an inference error of the ML model has occurred.

[0158] Additionally, the first device 210 may convey the above information using one or more messages (referred to as second messages). In some embodiments, the second message may include a cause for sending the second message. One exemplary cause is an ML model inference error. Other exemplary causes include, but are not limited to, a dead battery, a dead ML computing power, etc.

[0159] In some embodiments, the second message is transmitted periodically. Alternatively, in some embodiments, the second message is transmitted only in response to detecting an inference error in the ML model.

[0160] In response, the first device 210 may receive relevant information from the second device 220 .

[0161] In some embodiments, the first device 210 receives a second inference result generated by the ML model at the second device 220 and a second measurement result corresponding to the ML model measured by the second device 220.

[0162] Alternatively or additionally, in some embodiments, the first device 210 may receive a second difference between the second inference result and the second measurement result from the second device 220.

[0163] Similar to the first device 210, the second device 220 may determine an inference state of the ML model (as described below), in which case the first device 210 may receive a second indication from the second device 220 indicating whether an inference error of the ML model has occurred.

[0164] Additionally, the second device 220 may convey the above information using one or more messages (referred to as a fourth message). In some embodiments, the fourth message may include a cause for sending the fourth message.

[0165] In some embodiments, the fourth message is transmitted periodically. Alternatively, in some embodiments, the fourth message is transmitted only in response to detecting an inference error in the ML model.

[0166] With the above information, either the first device 210 or the second device 220 may determine whether an inference error of the ML model has occurred (ie, perform a validation procedure).

[0167] In some embodiments, the first device 210 may determine an inference state based on inference results (generated by the ML model at either the first device 210 or the second device 220) and measurement results corresponding to the ML model (measured by either the first device 210 or the second device 220).

[0168] Alternatively or additionally, in some embodiments, the first device 210 may determine the inference state based on the difference between the inferred result and the measured result. In some embodiments, a validation procedure may be performed based on whether the difference between the inferred result and the measured result satisfies a defined condition for the ML model inference error.

[0169] It should be understood that the detailed conditions for different use cases may be different. For a CSI compression and recovery function / use case, the condition may be a mean square error (MSE) threshold between the recovered CSI and the first CSI. For a beam management function / use case, the condition may be the probability that the L1-RSRP between the inferred beam based on the second set and the best beam of the first set is smaller than a threshold, e.g., 3 dB. For a low-overhead RS function / use case, the condition is an MSE threshold between the channel inferred from measurements of the second RS and the channel inferred from measurements of the first RS.

[0170] In some embodiments, the validation procedure is not performed as a single comparison, but is instead based on statistical data (average values, filter values, etc.) over the entire validation time interval or over a sliding window within the validation time interval.

[0171] Alternatively or additionally, in some embodiments, the first device 210 may determine the inference state directly based on a second indication received from the second device 220 indicating whether an inference error in the ML model has occurred.

[0172] Similar operations may be performed by the second device 220. Specifically, in some embodiments, the second device 220 may determine an inference state based on inference results (generated by the ML model at either the first device 210 or the second device 220) and measurement results corresponding to the ML model (measured by either the first device 210 or the second device 220).

[0173] Alternatively or additionally, in some embodiments, the second device 220 may also determine an inference state based on a difference between the inferred result and the measured result.

[0174] Alternatively or additionally, in some embodiments, the second device 220 may also determine the inference state directly based on a first indication received from the first device 210 indicating whether an inference error in the ML model has occurred.

[0175] According to the above procedure, either the first device 210 or the second device 220 may determine whether an inference error occurs. Thus, according to some embodiments of the present disclosure, inference errors can be better addressed by providing support for monitoring the performance of ML models by comparing inference results of the ML models with measurement results.

[0176] In some embodiments, when the first device 210 / or the second device 220 determines that an inference error of the ML model occurs, the first device 210 / or the second device 220 disables the generation of the inference result by the ML model until the update of the ML model is completed. In other words, when an inference error is detected, the inference cycle of the ML model is stopped. In one embodiment, disabling the generation of the inference result is achieved by stopping the first time interval or setting a timer for maintaining the first time interval to "0" or expiration. In another embodiment, disabling the generation of the inference result is achieved by setting a timer for maintaining the inference cycle to "0" or expiration. Additionally, configuration_1 is activated and configuration_2 is deactivated before the update of the ML mode is completed.

[0177] Reference is now made to Fig. 6. As shown in Fig. 6, during time period 610, the first device 210 / or the second device 220 disables generating inference results. Additionally, during time period 610, the first device 210 / or the second device 220 may perform conventional measurements according to a previously received configuration (e.g., the first configuration as described above).

[0178] Alternatively or additionally, the first device 210 and the second device 220 may continue to use the ML model in which the inference error occurred, for example activating configuration_2.

[0179] Alternatively or additionally, the first device 210 / or the second device 220 may exchange some information to detail the inference error or to initiate an update procedure. As shown in Fig. 4, the first device 210 and the second device 220 update the ML model (450-1 and 450-2). Furthermore, some information may be exchanged between the first device 210 and the second device 220 (451).

[0180] In some embodiments, if the first device 210 determines that an inference error occurred in the ML model, the first device 210 may transmit information about the ML model in which the inference error occurred to the second device 220. For example, the first device 210 transmits a third message to the second device 220 including an index / identity of the ML model.

[0181] Alternatively or additionally, in some embodiments, the first device 210 may transmit information about the candidate ML model. For example, the first device 210 may be configured with multiple ML models, and may determine that an inference error occurs in a currently applied ML model, and the first device 210 may perform a validation procedure on other configured ML models to select a candidate ML model. Additionally, the selected candidate ML model may be notified to the second device 220.

[0182] Similar operations for selecting candidate ML models may be performed by the second device 220. For the sake of brevity, identical or similar content will be omitted here.

[0183] Furthermore, if the candidate ML model needs to be distributed, the distribution of the candidate ML model may be performed according to the above-mentioned ML distribution procedure. For example, the candidate ML model may be delivered via a specific RNTI. Alternatively or additionally, the candidate ML model is distributed via one of group common transmission and broadcast. Alternatively or additionally, the candidate ML model is monitored via a specific radio resource configuration, for example a specific CORESET, or a specific search space set. Alternatively or additionally, the candidate ML model is distributed only if the communication situation between the first device 210 and the second device 220 satisfies a pre-set condition. Alternatively or additionally, the candidate ML model is transmitted with a specific scrambling ID. Alternatively or additionally, the candidate ML model is transmitted with a specific RRC message, a specific MAC CE LCID. Alternatively or additionally, the candidate ML model is transmitted on any or a combination of the PDCCH, a PDCCH scrambled with a specific RNTI or a specific scrambling ID, a group-common PDCCH, a physical downlink shared channel (PDSCH), a PDSCH scrambled with a specific RNTI or a specific scrambling ID, a group-common PDSCH, a paging message, a system information, a short message, a PUCCH, a PUCCH scrambled with a specific RNTI or a specific scrambling ID, a physical uplink shared channel (PUSCH), a PUSCH scrambled with a specific RNTI or a specific scrambling ID, etc.

[0184] Additionally, the candidate ML models themselves may be transmitted. Alternatively, the candidate ML models may be represented to the relevant network using at least one parameter / information associated with the one or more ML models, such as ML model architecture, algorithm, hyper-parameters, etc. Alternatively, the candidate ML models may be represented to the relevant network using an index / identity of the one or more ML models.

[0185] For brevity, more details about the distribution of candidate ML models are omitted here.

[0186] In the above procedure, the inference error is detected by setting at least one first time interval. In the following text, some other embodiments for detecting the inference error are described. For ease of explanation, in the embodiments described below, the first device 210 functions as a terminal device, and the second device 220 functions as a network device.

[0187] In operation, the first device 210 and the second device 220 may optionally exchange capability-related information and at least one setting. These procedures are similar to those described above, except that the at least one first time interval does not need to be set. Furthermore, in some embodiments, the capability-related information may be information about whether the terminal device supports an event-triggered ML model inference error report. Alternatively or additionally, in some embodiments, the capability-related information may be information about whether the terminal device supports detailed supported values ​​of those events. In summary, any capability-related information associated with an event-triggered ML model inference error report may be indicated by the terminal device. The present disclosure is not limited in this respect.

[0188] In some embodiments, the concept of ML model inference error events and event-triggered reporting is introduced. In some embodiments, the first device 210 detects predefined events indicative of an inference error of the ML model, the predefined events being specifically defined for a particular function related to the air interface. In particular, the predefined events may be defined via one or more indirect metrics that may reflect the inference state of the ML model. In this case, the first device 210 may detect the inference error without making actual measurements and comparing the actual measurement results with the inference results.

[0189] Additionally, predefined events may be defined according to different features / use cases: For the CSI compression and recovery (or low-overhead CSI-RS) feature / use case, the predefined events may include one or more instances where the indicated MCS clearly differs from the measured CSI (which implies erroneous compression and / or recovery / prediction of the CSI).

[0190] In one particular exemplary embodiment, the predefined event may be that the difference between the indicated MCS and the MCS mapped according to the measured CQI exceeds a threshold.

[0191] Alternatively, for beam management functionality / use cases, the predefined event may include one or more instances where an indicated beam has a much worse L1-RSRP than a reported beam, signifying an incorrect selection of an applied beam. Specifically, during beam management, an indicated beam may be represented by an RS ID associated with an indicated or activated state, and a reported beam may be represented by an RS ID reported in an L1-RSRP or L1-SINR report. In one particular exemplary embodiment, the predefined event may be that the L1-RSRP difference between the most recently indicated beam and the most recently reported beam exceeds a threshold.

[0192] Alternatively, for low overhead DMRS features / use cases, the predefined event may include one or more instances of receiving a negative acknowledgement (NACK) while in good channel conditions, meaning that the failed decoding is due to ML model inference error of DMRS channel estimation rather than low SNR. In one particular exemplary embodiment, the predefined event may be that the number of consecutive NACKs exceeds a threshold, but the indicated MCS and the mapping MCS according to SNR is less than a threshold.

[0193] Alternatively or additionally, the predefined event may be associated with an MCS-CQI mapping relationship, an MCS-SNR mapping relationship, etc. In some embodiments, the MCS-CQI mapping relationship, the MCS-SNR mapping relationship may be transmitted by the second device 220 to the first device 210. Alternatively, in some other embodiments, the MCS-CQI mapping relationship, the MCS-SNR mapping relationship may be transmitted by the first device 210 to the second device 220.

[0194] In some embodiments, the first device 210 determines that a predefined event has been detected based on statistical results over a certain period of time by utilizing a timer or counter. In one particular exemplary embodiment, the first device 210 determines that a predefined event has been detected if the predefined event continues over a certain period of time. In one particular exemplary embodiment, the first device 210 determines that a predefined event has been detected if the number of instances exceeds a threshold.

[0195] If the first device 210 detects a predefined event, the first device 210 may send some information detailing the inference error or initiate an updated procedure, as described above (some of the same or similar content is omitted for brevity).

[0196] In this way, by defining predefined events, we can effectively deal with inference errors in ML models. Example method

[0197] 7 is a flowchart of an example method 700 according to some embodiments of the present disclosure. For example, the method 700 may be implemented in the first device 210 as shown in FIG.

[0198] In block 710, the first device 210 receives from the second device 220 at least one setting associated with the ML model, the at least one setting indicating at least one first time interval for obtaining both an inference result generated by the ML model and a measurement result corresponding to the ML model.

[0199] In block 720, the first device 210 applies the at least one setting.

[0200] In some embodiments, the at least one first time interval is periodic, and the at least one setting indicates at least one of: a period of the at least one first time interval, a first length of each of the at least one first time interval, or information for determining a starting point for applying the at least one setting.

[0201] In some embodiments, the at least one configuration includes a first configuration for obtaining the measurement results corresponding to the ML model and a second configuration for obtaining the inference results generated by the ML model.

[0202] In some embodiments, each of the at least one first time interval is part of an inference cycle, the inference cycle including a first time interval of the at least one first time interval and a second time interval used by the first device 210 to obtain the inference result, the second time interval being before or after the first time interval in the inference cycle.

[0203] In some embodiments, the at least one setting indicates at least one of a period of the inference cycle, a first length of the first time interval, a second length of the second time interval, or information for determining a starting point for applying the at least one setting.

[0204] In some embodiments, when the first device 210 operates in a DRX mode, the inference cycle is aligned to the DRX cycle of the DRX mode.

[0205] In some embodiments, when the first device 210 operates in the DRX mode, the first device 210 obtains inference results and measurement results within the first time interval during an active time interval of a DRX cycle in the DRX mode, or the first device 210 skips the first time interval during an inactive time interval of the DRX cycle.

[0206] In some embodiments, the first device 210 selects one ML model from the multiple ML models when multiple ML models are configured in the first device 210, and sends a first message to the second device 220 indicating the selected ML model.

[0207] In some embodiments, the first device 210 sends to the second device 220 a second message indicating at least one of: a first inference result generated by the ML model at the first device 210; a first measurement result corresponding to the ML model measured by the first device 210; a first difference between the first inference result and the first measurement result; a first indication determined by the first device 210 indicating whether an inference error of the ML model has occurred; or a cause for sending the second message.

[0208] Alternatively, in some embodiments, the second message is sent periodically or only in response to detecting an inference error in the ML model.

[0209] In some embodiments, when an inference error occurs in the ML model, the first device 210 sends a third message to the second device 220 indicating at least one of information about the ML model in which the inference error occurred, information about a candidate ML model, or a cause for sending the third message.

[0210] In some embodiments, the first device 210 receives a fourth message from the second device 220 indicating at least one of: a second indication determined by the second device 220 indicating whether an inference error of the ML model occurred; a second inference result generated by the ML model at the second device 220; a second measurement result corresponding to the ML model measured by the second device 220; a second difference between the second inference result and the second measurement result; information about the ML model in which the inference error occurred; information about a candidate ML model; or a cause for sending the fourth message.

[0211] In some embodiments, the first device 210 determines whether an inference error of the ML model has occurred based on at least one of an inference result generated by the ML model in either the first device or the second device 220, a measurement result corresponding to the ML model measured by either the first device or the second device 220, a difference between the inference result and the measurement result, or a second indication received from the second device 220 indicating whether an inference error of the ML model has occurred.

[0212] In some embodiments, if the first device 210 determines that an inference error has occurred in the ML model, the first device 210 updates the ML model to a candidate ML model.

[0213] In some embodiments, if the first device 210 determines that an inference error has occurred in the ML model, the first device 210 disables the ML model from generating the inference results until an update of the ML model is completed.

[0214] In some embodiments, the first device 210 is configured to apply said at least one setting from a determined starting point depending on the particular operation.

[0215] In some embodiments, the first device 210 is a terminal device, the second device 220 is a network device or an ML function controller, and the ML model is an ML model used for the air interface.

[0216] 8 is a flowchart of an example method 800 according to some embodiments of the present disclosure. For example, the method 800 may be implemented in a terminal device such as the first device 210 shown in FIG. 2.

[0217] In block 810, the terminal device detects a predefined event indicative of an inference error in the ML model, the predefined event being specifically defined for a particular function related to the air interface.

[0218] In block 820, the terminal device sends a request to update the ML model to the network device.

[0219] In some embodiments, the terminal device determines that the predefined event has been detected based on statistical results over a certain period of time.

[0220] In some embodiments, the request is determined by the terminal device and indicates at least one of a third indication indicating that an inference error has occurred in the ML model, information about the ML model in which the inference error occurred, or information about a candidate ML model.

[0221] In some embodiments, in response to detecting the predefined event, the terminal device disables generation of inference results by the ML model until an update of the ML model is completed.

[0222] 9 is a flowchart of an example method 900 according to some embodiments of the present disclosure. For example, the method 900 may be implemented in the first device 210 as shown in FIG.

[0223] In block 910, the first device 210 receives one or more ML models from the second device 220 via at least one of an RNTI, a specific CORESET, a specific search space set, or one of a group common transmission and a broadcast.

[0224] In some embodiments, the first device 210 sends to the second device 220 capability-related information of the first device 210 indicating at least one of: whether the first device 210 supports detecting inference errors in ML models, the number of ML models configured in the first device 210, information regarding the ML model coordination level supported by the first device 210, information regarding the ML model architecture supported by the first device 210, or information regarding the ML features supported by the first device 210.

[0225] In some embodiments, the first device 210 receives from the second device 220 at least one setting for the one or more ML models, the at least one setting being specific to at least one of a particular cell, a particular group of cells, a particular first device 210, a particular group of first devices 210, or a particular function.

[0226] In some embodiments, the first device 210 is a terminal device, the second device 220 is a network device or an ML function controller, and the ML model is an ML model used for the air interface.

[0227] 10 is a flowchart of an example method 1000 according to some embodiments of the present disclosure. For example, the method 1000 may be implemented in the second device 220 as shown in FIG.

[0228] In block 1010, the second device 220 generates at least one setting associated with the ML model, the at least one setting indicating at least one first time interval for obtaining both an inference result generated by the ML model and a measurement result corresponding to the ML model.

[0229] In block 1020, the second device 220 transmits the at least one setting.

[0230] In some embodiments, the at least one first time interval is periodic, and the at least one setting indicates at least one of: a period of the at least one first time interval, a first length of each of the at least one first time interval, or information for determining a starting point for applying the at least one setting.

[0231] In some embodiments, the at least one configuration includes a first configuration for obtaining the measurement results corresponding to the ML model and a second configuration for obtaining the inference results generated by the ML model.

[0232] In some embodiments, each of the at least one first time interval is part of an inference cycle, the inference cycle including a first time interval of the at least one first time interval and a second time interval used by the first device 210 to obtain the inference result, the second time interval being before or after the first time interval in the inference cycle.

[0233] In some embodiments, the at least one setting indicates at least one of a period of the inference cycle, a first length of the first time interval, a second length of the second time interval, or information for determining a starting point for applying the at least one setting.

[0234] In some embodiments, when the first device 210 operates in a DRX mode, the inference cycle is aligned to the DRX cycle of the DRX mode.

[0235] In some embodiments, the second device 220 transmits a plurality of ML models to the first device 210 and receives a first message from the first device 210 indicating an ML model selected by the first device 210 from the plurality of ML models.

[0236] In some embodiments, the second device 220 receives a second message from the first device 210 indicating at least one of a first inference result generated by the ML model at the first device 210, a first measurement result corresponding to the ML model, a first difference between the first inference result and the first measurement result, a first indication determined by the first device 210 indicating whether an inference error in the ML model has occurred, or a cause for sending the second message.

[0237] In some embodiments, the second device 220 receives a third message from the first device 210 indicating at least one of information about the ML model in which the inference error occurred, information about a candidate ML model, or a cause for sending the third message.

[0238] In some embodiments, the second device 220 sends to the first device 210 a fourth message indicating at least one of: a second indication determined by the second device 220 indicating whether an inference error of the ML model occurred, a second inference result generated by the ML model at the second device 220, a second measurement result corresponding to the ML model measured by the second device 220, a second difference between the second inference result and the second measurement result, information about the ML model in which the inference error occurred, information about a candidate ML model, or a cause for sending the fourth message.

[0239] In some embodiments, the fourth message is sent periodically or only in response to detecting an inference error in the ML model.

[0240] In some embodiments, the second device 220 determines whether an inference error in the ML model has occurred based on at least one of an inference result generated by the ML model in either the first device or the second device 220, a measurement result corresponding to the ML model measured by either the first device or the second device 220, a difference between the inference result and the measurement result, or a first indication received from the first device 210 indicating whether an inference error in the ML model has occurred.

[0241] In some embodiments, the first device 210 is a terminal device, the second device 220 is a network device or an ML function controller, and the ML model is an ML model used for the air interface.

[0242] 11 is a flowchart of an example method 1100 according to some embodiments of the present disclosure. For example, the method 1100 may be implemented in the second device 220 as shown in FIG.

[0243] In block 1110, the second device 220 transmits to the first device 210 one or more ML models transmitted via at least one of the following: an RNTI, a specific CORESET, a specific search space set, or one of a group common transmission and a broadcast.

[0244] In some embodiments, the second device 220 receives capability-related information of the first device 210 from the first device 210 indicating at least one of: whether the first device 210 supports detecting inference errors in ML models, the number of ML models configured in the first device 210, information regarding the ML model coordination level supported by the first device 210, information regarding the ML model architecture supported by the first device 210, or information regarding the ML features supported by the first device 210.

[0245] In some embodiments, the second device 220 transmits to the first second device 220 at least one setting for the one or more ML models, the at least one setting being specific to at least one of a particular cell, a particular group of cells, a particular first device 210, a particular group of first devices 210, or a particular function.

[0246] In some embodiments, the second device 220 transmits the one or more ML models when a communication condition between the first device 210 and the second device 220 satisfies a preset condition.

[0247] In some embodiments, the first device 210 is a terminal device, the second device 220 is a network device or an ML function controller, and the ML model is an ML model used for the air interface. Equipment example

[0248] In some embodiments, the first device 210 comprises a circuit configured to receive from the second device 220 at least one setting associated with an ML model, the at least one setting indicating at least one first time interval for obtaining both an inference result generated by the ML model and a measurement result corresponding to the ML model, and to apply the at least one setting.

[0249] In some embodiments, the at least one first time interval is periodic, and the at least one setting indicates at least one of: a period of the at least one first time interval, a first length of each of the at least one first time interval, or information for determining a starting point for applying the at least one setting.

[0250] In some embodiments, the at least one configuration indicates a first configuration for obtaining the measurement results corresponding to the ML model and a second configuration for obtaining the inference results generated by the ML model.

[0251] In some embodiments, each of the at least one first time interval is part of an inference cycle, the inference cycle including a first time interval of the at least one first time interval and a second time interval used by the first device 210 to obtain the inference result, the second time interval being before or after the first time interval in the inference cycle.

[0252] In some embodiments, the at least one setting indicates at least one of a period of the inference cycle, a first length of the first time interval, a second length of the second time interval, or information for determining a starting point for applying the at least one setting.

[0253] In some embodiments, when the first device 210 operates in a DRX mode, the inference cycle is aligned to the DRX cycle of the DRX mode.

[0254] In some embodiments, when the first device 210 operates in a DRX mode, the circuitry is configured to obtain inferences and measurements within a first time interval during an active time interval of a DRX cycle in the DRX mode, or to skip the first time interval during an inactive time interval of the DRX cycle.

[0255] In some embodiments, the circuitry is further configured to, when multiple ML models are configured in the first device 210, select one ML model from the multiple ML models and send a first message to the second device indicating the selected ML model.

[0256] In some embodiments, the circuitry is further configured to send to the second device 220 a second message indicating at least one of: a first inference result generated by the ML model at the first device 210; a first measurement result corresponding to the ML model measured by the first device 210; a first difference between the first inference result and the first measurement result; a first indication determined by the first device 210 indicating whether an inference error in the ML model has occurred; or a cause for sending the second message.

[0257] In some embodiments, the second message is sent periodically or only in response to detecting an inference error in the ML model.

[0258] In some embodiments, the circuitry is further configured to, if an inference error in the ML model occurs, send a third message to the second device 220 indicating at least one of information about the ML model in which the inference error occurred, information about a candidate ML model, or a cause for sending the third message.

[0259] In some embodiments, the circuitry is further configured to receive from the second device 220 a fourth message indicating at least one of: a second indication determined by the second device 220 indicating whether an inference error of the ML model occurred; a second inference result produced by the ML model in the second device 220; a second measurement result corresponding to the ML model measured by the second device 220; a second difference between the second inference result and the second measurement result; information about the ML model in which the inference error occurred; information about a candidate ML model; or a cause for sending the fourth message.

[0260] In some embodiments, the circuitry is further configured to determine whether an inference error of the ML model has occurred based on at least one of an inference result generated by the ML model in either the first device or the second device 220, a measurement result corresponding to the ML model measured by either the first device or the second device 220, a difference between the inference result and the measurement result, or a second indication received from the second device 220 indicating whether an inference error of the ML model has occurred.

[0261] In some embodiments, if the first device 210 determines that an inference error of the ML model has occurred, the circuitry is further configured to update the ML model to a candidate ML model.

[0262] In some embodiments, if the first device 210 determines that an inference error has occurred in the ML model, the circuitry is further configured to disable the ML model from generating the inference result until an update of the ML model is completed.

[0263] In some embodiments, the circuitry is further configured to apply the at least one setting from a determined starting point in response to a particular operation.

[0264] In some embodiments, the first device 210 is a terminal device, the second device 220 is a network device or an ML function controller, and the ML model is an ML model used for the air interface.

[0265] In some embodiments, a terminal device (e.g., a first device 210 as shown in FIG. 2) comprises circuitry configured to detect a predefined event indicative of an inference error in the ML model, the predefined event being specifically defined for a particular function related to the air interface, and to send a request to a network device to update the ML model.

[0266] In some embodiments, the circuitry is further configured to determine that the predefined event has been detected based on statistical results over a period of time.

[0267] In some embodiments, the request is determined by the terminal device and indicates at least one of a third indication indicating that an inference error has occurred in the ML model, information about the ML model in which the inference error occurred, or information about a candidate ML model.

[0268] In some embodiments, in response to detecting the predefined event, the circuitry is further configured to disable generation of inference results by the ML model until an update of the ML model is completed.

[0269] The first device 210 comprises a circuit configured to receive one or more ML models received from the second device 220 via at least one of an RNTI, a particular CORESET, a particular search space set, or one of a group common transmission and a broadcast.

[0270] In some embodiments, the circuitry is further configured to send to the second device 220 capability-related information of the first device 210 indicating at least one of: whether the first device 210 supports detecting inference errors in ML models, the number of ML models configured in the first device 210, information regarding the ML model coordination level supported by the first device 210, information regarding the ML model architecture supported by the first device 210, or information regarding the ML features supported by the first device 210.

[0271] In some embodiments, the circuitry is further configured to receive from the second device 220 at least one setting for the one or more ML models, the at least one setting being specific to at least one of a particular cell, a particular group of cells, a particular first device 210, a particular group of first devices 210, or a particular function.

[0272] In some embodiments, the first device 210 is a terminal device, the second device 220 is a network device or an ML function controller, and the ML model is an ML model used for the air interface.

[0273] In some embodiments, the second device 220 comprises circuitry configured to generate at least one setting associated with an ML model, the at least one setting indicating at least one first time interval for obtaining both an inference result generated by the ML model and a measurement result corresponding to the ML model, and to transmit the at least one setting.

[0274] In some embodiments, the at least one first time interval is periodic, and the at least one setting indicates at least one of: a period of the at least one first time interval, a first length of each of the at least one first time interval, or information for determining a starting point for applying the at least one setting.

[0275] In some embodiments, the at least one configuration includes a first configuration for obtaining the measurement results corresponding to the ML model and a second configuration for obtaining the inference results generated by the ML model.

[0276] In some embodiments, each of the at least one first time interval is part of an inference cycle, the inference cycle including a first time interval of the at least one first time interval and a second time interval used by the first device 210 to obtain the inference result, the second time interval being before or after the first time interval in the inference cycle.

[0277] In some embodiments, the at least one setting indicates at least one of a period of the inference cycle, a first length of the first time interval, a second length of the second time interval, or information for determining a starting point for applying the at least one setting.

[0278] In some embodiments, when the first device 210 operates in a DRX mode, the inference cycle is aligned to the DRX cycle of the DRX mode.

[0279] In some embodiments, the circuitry is further configured to transmit a plurality of ML models to the first device 210 and receive a first message from the first device 210 indicating an ML model selected by the first device 210 from the plurality of ML models.

[0280] In some embodiments, the circuitry is further configured to receive from the first device 210 a second message indicating at least one of a first inference result generated by the ML model at the first device 210, a first measurement result corresponding to the ML model, a first difference between the first inference result and the first measurement result, a first indication determined by the first device 210 indicating whether an inference error in the ML model has occurred, or a cause for sending the second message.

[0281] In some embodiments, the circuitry is further configured to receive a third message from the first device 210 indicating at least one of information about the ML model in which the inference error occurred, information about a candidate ML model, or a cause for sending the third message.

[0282] In some embodiments, the circuitry is further configured to send to the first device 210 a fourth message indicating at least one of: a second indication determined by the second device 220 indicating whether an inference error of the ML model occurred; a second inference result produced by the ML model at the second device 220; a second measurement result corresponding to the ML model measured by the second device 220; a second difference between the second inference result and the second measurement result; information about the ML model in which the inference error occurred; information about a candidate ML model; or a cause for sending the fourth message.

[0283] In some embodiments, the fourth message is sent periodically or only in response to detecting an inference error in the ML model.

[0284] In some embodiments, the circuitry is further configured to determine whether an inference error in the ML model has occurred based on at least one of an inference result generated by the ML model in either the first device or the second device 220, a measurement result corresponding to the ML model measured by either the first device or the second device 220, a difference between the inference result and the measurement result, or a first indication received from the first device 210 indicating whether an inference error in the ML model has occurred.

[0285] In some embodiments, the first device 210 is a terminal device, the second device 220 is a network device or an ML function controller, and the ML model is an ML model used for the air interface.

[0286] In some embodiments, the second device 220 comprises circuitry configured to transmit to the first device 210 one or more ML models transmitted via at least one of an RNTI, a particular CORESET, a particular search space set, or one of a group common transmission and a broadcast.

[0287] In some embodiments, the circuitry is further configured to receive capability-related information of the first device 210 from the first device 210 indicating at least one of: whether the first device 210 supports detecting inference errors in ML models, the number of ML models configured in the first device 210, information regarding a level of ML model coordination supported by the first device 210, information regarding an ML model architecture supported by the first device 210, or information regarding ML features supported by the first device 210.

[0288] In some embodiments, the circuitry is further configured to transmit to the first second device 220 at least one setting for the one or more ML models, the at least one setting being specific to at least one of a particular cell, a particular group of cells, a particular first device 210, a particular group of first devices 210, or a particular function.

[0289] In some embodiments, the circuitry is further configured to transmit the one or more ML models when a communication condition between the first device 210 and the second device 220 satisfies a preset condition.

[0290] In some embodiments, the first device 210 is a terminal device, the second device 220 is a network device or an ML function controller, and the ML model is an ML model used for the air interface.

[0291] Fig. 12 is a schematic block diagram of an apparatus 1200 suitable for implementing embodiments of the present disclosure. The apparatus 1200 can be considered as another exemplary implementation of the first apparatus 210 and the second apparatus 220 shown in Fig. 2. Thus, the apparatus 1200 can be implemented in the first apparatus 210 and the second apparatus 220 or as at least a part thereof.

[0292] As shown, the apparatus 1200 comprises a processor 1212, a memory 1220 coupled to the processor 1212, a suitable transmitter (TX) and receiver (RX) 1240 coupled to the processor 1212, and a communication interface coupled to the TX / RX 1240. The memory 1212 stores at least a portion of a program 1230. The TX / RX 1240 is used for bidirectional communication. The TX / RX 1240 has at least one antenna to facilitate communication, although the access nodes referred to herein may actually have multiple antennas. The communication interface may represent any interface required for communication with other network elements, such as an X2 interface for bidirectional communication between eNBs, an S1 interface for communication between a mobility management entity (MME) / serving gateway (S-GW) and an eNB, an Un interface for communication between an eNB and a relay node (RN), or an Uu interface for communication between an eNB and a terminal device.

[0293] The program 1230 is assumed to include program instructions that, when executed by the associated processor 1212, enable the device 1200 to operate according to embodiments of the present disclosure, as described herein with reference to Figures 2-11. The embodiments of the present disclosure may be implemented by computer software executable by the processor 1212 of the device 1200, or by hardware, or by a combination of software and hardware. The processor 1212 may be configured to implement various embodiments of the present disclosure. Furthermore, the combination of the processor 1212 and the memory 1220 may form a processing means 1250 suitable for implementing various embodiments of the present disclosure.

[0294] The memory 1220 may be of any type suitable for a local technology network and may be implemented using any suitable data storage technology, such as, by way of non-limiting example, non-transitory computer-readable storage media, semiconductor-based memory devices, magnetic memory devices and systems, optical memory devices and systems, fixed memory and removable memory. Although only one memory 1220 is shown in the device 1200, there may be several physically different memory modules in the device 1200. The processor 1212 may be of any type suitable for a local technology network and may include, by way of non-limiting example, one or more of a general purpose computer, a special purpose computer, a microprocessor, a digital signal processor (DSP) and a processor based on a multi-core processor architecture. The device 1200 may have multiple processors, for example application specific integrated circuit chips time-slaved to a clock that synchronizes the main processor.

[0295] Overall, various embodiments of the present disclosure may be implemented in hardware or special purpose circuits, software, logic, or any combination thereof. Some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software that may be executed by a controller, microprocessor, or other computing device. Although various aspects of the embodiments of the present disclosure have been illustrated and described using block diagrams, flow charts, or some other pictorial representations, it should be understood that the blocks, devices, systems, techniques, or methods described herein may be implemented, by way of non-limiting examples, in hardware, software, firmware, special purpose circuits or logic, general purpose hardware or controller or other computing device, or any combination thereof.

[0296] The present disclosure also provides at least one computer program product tangibly stored on a non-transitory computer-readable storage medium. The computer program product includes computer-executable instructions, such as instructions included in a program module, that execute in a device on a target real or virtual processor to perform the process or method described above with reference to Figures 7 to 11. Generally, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform particular tasks or implement particular abstract data types. In various embodiments, the functionality of the program modules may be combined or split between program modules as appropriate. The machine-executable instructions of the program modules may be executed in local or distributed devices. In a distributed device, the program modules may be located in both local and remote storage media.

[0297] Program codes for carrying out the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, a special purpose computer, or other programmable data processing device, and when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may run entirely on the machine, partially on the machine, as a separate software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0298] The above program code may be implemented on a machine-readable medium, which may be any tangible medium that can contain or store a program used by or associated with an instruction execution system, device, or apparatus. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or apparatus, or any suitable combination of the aforementioned media. More specific examples of the machine-readable storage medium may include an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the aforementioned.

[0299] It should be noted that although operations have been described in a particular order, it should not be understood that such operations need to be performed in the particular order shown, or in any sequential order, or that all of the operations described must be performed in order to achieve desired results. In some cases, multitasking and parallel processing may be advantageous. Similarly, although some specific implementation details are included in the above discussion, these should not be construed as limitations on the scope of the disclosure, but rather as descriptions of features that may be specific to certain embodiments. Some features described in the context of individual embodiments may be realized in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may be implemented separately in multiple embodiments or in any suitable subcombination.

[0300] Although the present disclosure has been described in language specific to structural features and / or methodological acts, it should be understood that the present disclosure, as defined in the appended claims, is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims.

Claims

1. 1. A method of communication comprising: receiving, at a first device, from a second device, at least one setting associated with a machine learning (ML) model, the at least one setting indicating at least one first time interval for obtaining both inference results generated by the ML model and measurement results corresponding to the ML model; applying said at least one setting; and The method includes:

2. The at least one first time interval is periodic, and the at least one setting comprises: a period of the at least one first time interval; a first length of each of the at least one first time interval; or information for determining a starting point for applying the at least one setting; The method of claim 1 , wherein the at least one of

3. The at least one setting may include: a first configuration for obtaining the measurement results corresponding to the ML model; a second setting for obtaining the inference results generated by the ML model; The method of claim 1 , wherein

4. Each of the at least one first time interval is part of an inference cycle, the inference cycle comprising: a first time interval of the at least one first time interval; and a second time interval used by the first device to obtain the inference result, the second time interval being either before or after the first time interval in the inference cycle; 2. The method of claim 1, comprising:

5. The at least one setting is: the period of said inference cycle; a first length of the first time interval; a second length of the second time interval; or information for determining a starting point for applying the at least one setting; The method of claim 4 , wherein the at least one of

6. If the first device operates in a discontinuous reception (DRX) mode, the inference cycle is aligned to a DRX cycle of the DRX mode. The method according to claim 4.

7. If the first device operates in a discontinuous reception (DRX) mode, applying the at least one setting includes: obtaining the inference and the measurement within the first time interval during an active time interval of a DRX cycle of the DRX mode; or skipping the first time interval during an inactive time interval of the DRX cycle; 5. The method of claim 4, comprising:

8. selecting one ML model from the plurality of ML models when a plurality of ML models are configured in the first device; sending a first message to the second device indicating a selected ML model; The method of claim 1 further comprising:

9. The second device comprises: a first inference result generated by the ML model in the first device; a first measurement result corresponding to the ML model, measured by the first device; a first difference between the first inference and the first measurement; a first indication, determined by the first device, of whether an inference error in the ML model has occurred; or a cause for sending the second message; transmitting said second message indicating at least one of: The method of claim 1 further comprising:

10. The second message is transmitted periodically or only in response to detecting an inference error in the ML model.

10. The method of claim 9.

11. When an inference error occurs in the ML model, the second device Information about the ML model in which the inference error occurred; Information about the candidate ML model, or a cause for sending the third message; transmitting the third message indicating at least one of: The method of claim 1 further comprising:

12. from the second device, a second indication, determined by the second device, of whether an inference error in the ML model occurred; a second inference result generated by the ML model at the second device; a second measurement result corresponding to the ML model, measured by the second device; a second difference between the second inference and the second measurement; Information about the ML model in which the inference error occurred; Information about the candidate ML model, or a cause for sending the fourth message; receiving the fourth message indicating at least one of: The method of claim 1 further comprising:

13. Whether or not an inference error occurs in the ML model is determined. inference results generated by the ML model in either the first device or the second device; a measurement result corresponding to the ML model, the measurement result being measured by either the first device or the second device; the difference between said inference and said measurement; or a second indication received from the second device indicating whether an inference error of the ML model occurred; determining based on at least one of: The method of claim 1 further comprising:

14. updating the ML model to a candidate ML model when the first device determines that an inference error has occurred in the ML model; The method of claim 1 further comprising:

15. when the first device determines that an inference error has occurred in the ML model, disabling generation of the inference result by the ML model until an update of the ML model is completed; The method of claim 1 further comprising:

16. Applying the at least one setting comprises: applying said at least one setting from a determined starting point depending on the particular operation; 2. The method of claim 1, comprising:

17. The first device is a terminal device, the second device is a network device or an ML function controller, and the ML model is an ML model used for an air interface. The method of claim 1.

18. 1. A method of communication comprising: detecting, in the terminal device, a predefined event indicative of an inference error of a machine learning (ML) model, the predefined event being specifically defined for a particular function on the air interface; sending a request to a network device to update the ML model; The method includes:

19. determining that the predefined event has been detected based on statistical results over a period of time; 20. The method of claim 18 further comprising:

20. The request is a third indication, determined by the terminal device, indicating that an inference error in the ML model has occurred; Information about the ML model in which the inference error occurred; or Information about the candidate ML models; 20. The method of claim 18, further comprising at least one of:

21. In response to detecting the predefined event, disabling generation of inference results by the ML model until an update of the ML model is completed; 20. The method of claim 18 further comprising:

22. 1. A method of communication comprising: In the first device, from the second device, A specific Radio Network Temporary Identity (RNTI); A specific control resource set (CORESET), A particular search space set, or One of group common transmission and broadcast, receiving one or more machine learning (ML) models via at least one of: The method includes:

23. The second device comprises: Whether the first device supports detecting inference errors in ML models; the number of ML models configured in the first device; Information regarding the ML cooperation level supported by the first device; Information regarding the ML architecture supported by the first device; or Information regarding ML capabilities supported by the first device; transmitting capability-related information of the first device indicative of at least one of:

23. The method of claim 22 further comprising:

24. at least one setting for the one or more ML models from the second device, A specific cell, A specific cell group, A specific first device, A specific first device group, or Specific features, receiving said at least one setting specific to at least one of 23. The method of claim 22 further comprising:

25. 1. A method of communication comprising: generating, in a second device, at least one configuration associated with a machine learning (ML) model, the at least one configuration indicating at least one first time interval for obtaining both an inference result generated by the ML model and a measurement result corresponding to the ML model; transmitting said at least one configuration; The method includes:

26. The at least one first time interval is periodic, and the at least one setting comprises: a period of the at least one first time interval; a first length of each of the at least one first time interval; or information for determining a starting point for applying the at least one setting; 26. The method of claim 25.

27. The at least one setting may include: a first configuration for obtaining the measurement results corresponding to the ML model; a second setting for obtaining the inference results generated by the ML model; 26. The method of claim 25, comprising:

28. Each of the at least one first time interval is part of an inference cycle, the inference cycle comprising: a first time interval of the at least one first time interval; and a second time interval used by the first device to obtain the inference result, the second time interval being either before or after the first time interval in the inference cycle; 26. The method of claim 25, comprising:

29. The at least one setting is: the period of said inference cycle; a first length of the first time interval; a second length of the second time interval; or information for determining a starting point for applying the at least one setting; 30. The method of claim 28, further comprising at least one of:

30. If the first device operates in a discontinuous reception (DRX) mode, the inference cycle is aligned to a DRX cycle of the DRX mode.

29. The method of claim 28.

31. transmitting a plurality of ML models to the first device; receiving a first message from the first device indicating an ML model selected by the first device from the plurality of ML models; 26. The method of claim 25 further comprising:

32. From the first device, a first inference result generated by the ML model in the first device; A first measurement result corresponding to the ML model; a first difference between the first inference and the first measurement; a first indication, determined by the first device, of whether an inference error in the ML model has occurred; or a cause for sending the second message; receiving the second message indicating at least one of:

26. The method of claim 25 further comprising:

33. From the first device, Information about the ML model in which the inference error occurred; Information about the candidate ML model, or a cause for sending the third message; receiving the third message indicating at least one of:

26. The method of claim 25 further comprising:

34. The first device comprises: a second indication, determined by the second device, of whether an inference error in the ML model occurred; a second inference result generated by the ML model at the second device; a second measurement result corresponding to the ML model, measured by the second device; a second difference between the second inference and the second measurement; Information about the ML model in which the inference error occurred; Information about the candidate ML model, or a cause for sending the fourth message; transmitting said fourth message indicating at least one of:

26. The method of claim 25 further comprising:

35. The fourth message is transmitted periodically or only in response to detecting an inference error in the ML model.

35. The method of claim 34.

36. Whether or not an inference error occurs in the ML model is determined. inference results generated by the ML model in either the first device or the second device; the measurement results corresponding to the ML model, measured by either the first device or the second device; the difference between said inference and said measurement; or a first indication received from the first device indicating whether an inference error of the ML model occurred; determining based on at least one of:

26. The method of claim 25 further comprising:

37. The first device is a terminal device, the second device is a network device or an ML function controller, and the ML model is an ML model used for an air interface.

26. The method of claim 25.

38. 1. A method of communication comprising: In the second device, the first device is A specific Radio Network Temporary Identity (RNTI); A specific control resource set (CORESET), A particular search space set, or One of group common transmission and broadcast, transmitting one or more machine learning (ML) models transmitted via at least one of The method includes:

39. From the first device, Whether the first device supports detecting inference errors in ML models; the number of ML models configured in the first device; Information regarding the ML cooperation level supported by the first device; Information regarding the ML architecture supported by the first device; or Information regarding ML capabilities supported by the first device; receiving capability-related information of the first device indicative of at least one of:

40. The method of claim 38, further comprising:

40. at least one setting for the one or more ML models, A specific cell, A specific cell group, A specific first device, A specific first device group, or Specific features, transmitting said at least one setting specific to at least one of 40. The method of claim 38, further comprising:

41. Transmitting the one or more ML models comprises: transmitting the one or more ML models when a communication situation between the first device and the second device satisfies a preset condition; 39. The method of claim 38, comprising:

42. A first apparatus comprising a processor and a memory coupled to the processor and storing instructions, The instructions, when executed by the processor, perform a method according to any one of claims 1 to 17 or claims 22 to 24. A first device.

43. A processor; a memory coupled to the processor and storing instructions; The instructions, when executed by the processor, perform a method according to any one of claims 18 to 21. Terminal device.

44. A second apparatus comprising a processor and a memory coupled to the processor and storing instructions, The instructions, when executed by the processor, perform a method according to any one of claims 25 to 41. A second device.

45. storing instructions which, when executed on at least one processor, cause said at least one processor to perform the method according to any one of claims 1 to 41; Computer-readable medium.

Citation Information

Patent Citations

  • Disaster prevention business support system

    JP2005301840A

  • Providing producer node machine learning based assistance

    WO2021063500A1

Cited By

  • Terminal, wireless communication method and system

    JP7823085B2

  • Terminal, wireless communication method, and base station

    JPWO2023135800A1