Terminal device method, network device and terminal device

Abstract

[0003] Exemplary embodiments of the present disclosure relate to an effective mechanism for communication. In one solution, a device, in a terminal device supporting multiple machine learning (ML) models, determines a respective priority or a respective number of ML processing resources for each ML model among at least one ML model running on the terminal device, and schedules the ML processing resources for the at least one ML model based on the respective priority or the respective number of ML processing resources. In this way, limited ML processing resources / capacity can be allocated to the most important ML model(s).

Description

[Technical Field]

[0001] FIELD Embodiments of the present disclosure relate generally to the field of communication technologies, and more particularly to methods, apparatus, and media for machine learning (ML) assisted communication. [Background technology]

[0002] As communication networks and services grow in size and complexity, and the number of users increases, the complexity of communications in communication networks increases. ML / artificial intelligence (AI) techniques have been proposed in wireless communication networks to improve communication performance. For example, terminal devices may use different ML models to support communication-related functions, such as multi-input multi-output (MIMO), channel state information (CSI), and beam management (BM). Summary of the Invention [Problem to be solved by the invention]

[0003] However, in the case of MIMO mode or multi-ML mode, the amount of calculations on the terminal device increases dramatically, which usually requires a lot of computational resources and is likely to increase the device temperature. This is expected to enable rational scheduling of limited ML processing resources among multiple ML modes and to effectively control the device temperature. [Means for solving the problem]

[0004] SUMMARY OF THE INVENTION Embodiments of the present disclosure generally provide a method, apparatus, and computer storage medium for ML-assisted communication.

[0005] In a first aspect, a method of communication is provided. The method includes, in a terminal device supporting a plurality of ML models, determining a respective priority or a respective number of ML processing resources for each ML model among at least one ML model operating on the terminal device based on at least one factor including a real-time requirement of the ML model, a delay of the ML model, a collaboration level of the ML model, a report transmission requirement of the ML model, an ML type of the ML model being one of a two-sided ML model or a one-sided ML model, a number of entities involved in the ML model, a function of the ML model, an ML group to which the ML model belongs, an accuracy requirement of the ML model, or a communication protocol layer associated with the ML model. The method further includes scheduling the ML processing resources for the at least one ML model based on the respective priority or the respective number of ML processing resources.

[0006] In a second aspect, a method of communication is provided, the method including generating, at a terminal device, auxiliary information associated with at least one of first information used to adjust a specification of an ML model operating at the terminal device and second information used to adjust a MIMO specification, the second information being associated with at least one of a MIMO measurement specification, a MIMO calculation specification, and a MIMO maintenance specification, the method further including transmitting the auxiliary information to a network device.

[0007] In a third aspect, a method of communication is provided, the method including: predicting, at a terminal device, a temperature change in a subsequent time period; determining auxiliary information used to mitigate the temperature change in the subsequent time period if the temperature change satisfies an adjustment condition; and transmitting the auxiliary information to a network device.

[0008] In a fourth aspect, a method of communication is provided, the method including receiving, at a network device, auxiliary information from a terminal device, the auxiliary information being associated with at least one of first information used to adjust specifications of an ML model operating in the terminal device and second information used to adjust specifications of a MIMO.

[0009] In a fifth aspect, there is provided a method of communication, the method including receiving, at a network device, auxiliary information from a terminal device, the auxiliary information being transmitted by the terminal device in response to detecting that a predicted temperature change satisfies an adjustment condition, and being used to mitigate a temperature change of the terminal device in a subsequent time period.

[0010] In a sixth aspect, there is provided a first apparatus, the terminal apparatus comprising: a processing unit; and a memory, coupled to the processing unit, having instructions stored thereon, the instructions, when executed by the processing unit, causing the apparatus to perform a method according to the first aspect.

[0011] In a seventh aspect there is provided a terminal apparatus comprising a processing unit and a memory, coupled to the processing unit, having instructions stored thereon, the instructions, when executed by the processing unit, causing the apparatus to perform a method according to the second aspect.

[0012] In an eighth aspect, there is provided a first apparatus, the terminal apparatus comprising: a processing unit; and a memory, coupled to the processing unit, having instructions stored thereon, the instructions, when executed by the processing unit, causing the apparatus to perform a method according to the third aspect.

[0013] In a ninth aspect, there is provided a second apparatus, the network apparatus comprising: a processing unit; and a memory, coupled to the processing unit, having instructions stored thereon, the instructions, when executed by the processing unit, causing the apparatus to perform a method according to the fourth aspect.

[0014] In a tenth aspect, there is provided a second apparatus, the network apparatus comprising: a processing unit; and a memory, coupled to the processing unit, having instructions stored thereon, the instructions, when executed by the processing unit, causing the apparatus to perform a method according to the fifth aspect.

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

[0016] It should be understood that this Summary of the Invention is not intended to identify key features or essential features of the embodiments of the present disclosure, nor is it intended to be used to limit the scope of the present disclosure. Other features of the present disclosure will be readily apparent through the following description. [Brief explanation 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 embodiments of the present disclosure in the accompanying drawings.

[0018] [Figure 1A] FIG. 10 is a diagram showing the results of deterioration in correlation with device temperature. [Figure 1B] A figure showing the correlation results between throughput and the number of NR channels. [Figure 1C] FIG. 10 is a diagram showing the results of a comparison between the case where an ice bag is used and the case where an ice bag is not used. [Figure 2A] FIG. 1 illustrates an exemplary communication environment in which exemplary embodiments of the present disclosure may be implemented. [Figure 2B] FIG. 1 illustrates a signaling chart illustrating a process for communication according to some embodiments of the present disclosure. [Figure 3] FIG. 1 illustrates an exemplary method performed by a terminal device according to some embodiments of the present disclosure. [Figure 4]FIG. 10 illustrates another exemplary method performed by a terminal device according to some embodiments of the present disclosure. [Figure 5] FIG. 10 illustrates another exemplary method performed by a terminal device according to some embodiments of the present disclosure. [Figure 6] FIG. 1 illustrates an exemplary method performed by a network device according to some embodiments of the present disclosure. [Figure 7] FIG. 10 illustrates another exemplary method performed by a network device according to some embodiments of the present disclosure. [Figure 8] FIG. 1 is a schematic block diagram of an apparatus suitable for implementing exemplary embodiments of the present disclosure.

[0019] Throughout the drawings, the same or similar reference numbers represent the same or similar elements. DETAILED DESCRIPTION OF THE INVENTION

[0020] The principles of the present disclosure will be described with reference to several embodiments. It should be understood that these embodiments are set forth for illustrative purposes only to aid those skilled in the art in understanding and practicing the present disclosure, and do not imply any limitations on the scope of the present disclosure. The present disclosure described herein can be embodied in various forms other than those described below.

[0021] In the following description and claims, unless defined otherwise, all technical and scientific terms used have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs.

[0022] In this disclosure, the term "terminal device" refers to any device with 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 and Low Latency Communications (URLLC) devices, Any Internet of Things (IoE) devices, Machine Type Communications (MTC) devices, in-vehicle devices for V2X communications where X represents pedestrian, vehicle, or infrastructure / network, devices for Integrated Access and Backhaul (IAB), spacecraft or airborne vehicles in Non-Terrestrial Networks (NTN) including Satellites and High Altitude Platforms (HAPs) including Unmanned Aircraft Systems (UAS), and Extended Reality (XR) including different types of reality such as Augmented Reality (AR), Mixed Reality (MR), and Virtual Reality (VR). This includes, but is not limited to, Reality devices, unmanned aerial vehicles (UAVs), commonly known as drones, i.e., aircraft without a human pilot, devices on high-speed trains (HST), image capture devices such as digital cameras, sensors, gaming devices, music storage and playback devices, and internet appliances that enable wireless or wired internet access and browsing. A "terminal device" may also have "multicast / broadcast" capabilities and support public safety and mission-critical applications, V2X applications, transparent IPV4 / IPV6 multicast delivery, IPTV, smart TV, wireless services, over-the-air software delivery, group communications, and IoT applications. It may also incorporate one or more subscriber identity modules (SIMs), referred to as multi-SIM. The term "terminal device" may be used interchangeably with UE, mobile station, subscriber station, mobile terminal, user terminal, or wireless device.

[0023] The term "network device" refers to a device capable of providing or hosting a cell or coverage area over which terminal devices can communicate. Examples of network devices include, but are not limited to, a Node B (NodeB or NB), an evolved Node B (eNodeB or eNB), a next generation Node B (gNB), a transmit / receive point (TRP), a remote radio unit (RRU), a radio head (RH), a remote radio head (RRH), a low power node such as an IAB node, a femto node, a pico node, and a reconfigurable intelligent surface (RIS).

[0024] A terminal device or network device may have artificial intelligence (AI) or machine learning capabilities, which typically include models trained from a large amount of collected data for a specific function and can be used to predict some information.

[0025] The terminal device or network device may operate in multiple frequency ranges, such as FR1 (410 MHz to 7125 MHz), FR2 (24.25 GHz to 71 GHz), frequency bands greater than 100 GHz, and TERA Hertz (THZ). It can also operate in licensed, unlicensed, and shared spectrum. The terminal device may have multiple connections with network devices in 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.

[0026] 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.

[0027] In some embodiments, a terminal device may be connected to a first network device and a second network device. One of the first network device and the second network device may be a master node, and the other may be a secondary node. The first network device and the second network device may use different radio access technologies (RATs). In some embodiments, the first network device may be a first RAT device, and the second network device may be a second RAT device. In some embodiments, the first RAT device is an eNB, and the second RAT device is a gNB. Information related to the different RATs may be transmitted to the terminal device from at least one of the first network device and the second network device. In some embodiments, the first information may be transmitted from the first network device to the terminal device, and the second information may be transmitted from the second network device directly or via the first network device to the terminal device. In some embodiments, information related to the terminal device configuration configured by the second network device may be transmitted from the second network device via the first network device. The information regarding the reconfiguration of the terminal device configured by the second network device may be transmitted to the terminal device directly from the second network device or via the first network device.

[0028] In this disclosure, the singular forms "a," "the," and "the" are intended to include the plural unless the context clearly indicates otherwise. The term "comprises" and variations thereof are interpreted as open-ended terms meaning "including, but not limited to." The term "based on" is interpreted as "based at least in part on." The terms "one embodiment" and "embodiment" are interpreted as "at least one embodiment." The term "another embodiment" is interpreted as "at least one other embodiment." Terms such as "first," "second," etc. may refer to different objects or the same object. The following content may include other definitions, both explicit and implicit.

[0029] In some instances, values, procedures, or devices are referred to as "optimal," "lowest," "highest," "minimum," "maximum," etc. It is understood that such descriptions are intended to indicate choices among multiple functional alternatives used, and that such choices are not necessarily better, smaller, higher, or more preferred than other choices.

[0030] Millimeter-wave band communication and MIMO are identified as features of wireless communication. ML / AI technologies have also been proposed for wireless communication networks, where AI / ML models / functions (e.g., AI / ML models used in the air interface) can be deployed on the terminal side. However, in the case of MIMO mode or multi-ML mode, the amount of computation on the terminal device increases dramatically, requiring more computational resources and potentially increasing the device temperature.

[0031] 1A and 1B show exemplary results 100 and 120 of throughput degradation according to the device temperature of a terminal device. Specifically, FIG. 1A shows the degradation result 100 in correlation with device temperature, and FIG. 1B shows the correlation result 120 between throughput and the number of NR channels. As can be seen from FIGS. 1A and 1B, at time 125 (i.e., 300 seconds) in FIG. 1B, the number of 5G mmWave channels is reduced from four to one, and at time 105 (i.e., 500 seconds) in FIG. 1A, the terminal device is handed over from the NR to the LTE network. Furthermore, at both time 125 (i.e., 300 seconds) and time 125 (i.e., 500 seconds), a secondary cell group (SCG) failure signaling packet is identified.

[0032] Another possible approach to controlling the device temperature is to use physical means (such as covering the terminal device with an ice bag). Referring to FIG. 1C, a comparison result 160 is shown for cases where an ice bag is used and not used. As shown in FIG. 1C, when an ice bag is used (when the device temperature is controlled), the throughput of the terminal device can be maintained at a relatively stable state without a sudden drop. This indicates that high device temperature is the main cause of the drop in throughput.

[0033] However, using an ice bag is not a feasible solution, so it is desirable to propose a feasible solution that can avoid high device temperatures without impairing communication performance.

[0034] According to an exemplary embodiment of the present disclosure, a terminal device can reasonably schedule limited ML processing resources among multiple ML modes and / or effectively control device temperature without compromising communication performance by performing at least one of scheduling limited ML processing resources based on different priorities for different ML modes, adjusting specifications of MIMO and / or ML models, and adopting adaptive settings associated with different device temperatures.

[0035] In the following description, when describing specific embodiments, a terminal device and a network device are used as examples for ease of understanding. Unless otherwise specified, the embodiments described herein may be implemented between any suitable network elements. Specifically, either the terminal device or the network device may be replaced with the other device type.

[0036] Here, the term "model" refers to an input-output association learned from training data, which, after training, can generate a corresponding output for a specified input. The generation of the model may be based on ML techniques, which are also called AI techniques. In general, an ML model can be constructed that accepts input information and makes a prediction based on the input information.

[0037] Furthermore, for ease of understanding, Table 1 below explains some terms related to ML models. [Table 1] JPEG2025526821000003.jpg223168

[0038] In this disclosure, the terms "ML model," "AI model," "ML function," "AI function," and "algorithm" may be used interchangeably.

[0039] Example environment FIG. 2A illustrates an exemplary communications environment 200 in which exemplary embodiments of the present disclosure may be implemented.

[0040] The communication environment 200 includes a network device 210 and a terminal device 220, with the network device 210 capable of communicating with the terminal device 220 via a physical communication channel or link.

[0041] In the specific example of communication environment 200, a link from terminal device 220 to network device 210 is referred to as an uplink, and a link from network device 210 to terminal device 220 is referred to as a downlink. Furthermore, MIMO is supported in communication environment 200, whereby network device 210 and terminal device 220 communicate with each other via different beams, enabling directional communication. In the downlink, network device 210 is a transmitting (TX) device (or transmitter) and terminal device 220 is a receiving (RX) device (or receiver), and network device 210 may send downlink transmissions to terminal device 220 via one or more beams. As shown in FIG. 2A, network device 210 sends downlink transmissions to terminal device 220 via beams 240-1 through 240-3.

[0042] Correspondingly, in the uplink, network device 210 is an RX device (or receiver), terminal device 220 is a TX device (or transmitter), and terminal device 220 may send uplink transmissions to network device 210 via one or more beams. As shown in FIG. 2A, terminal device 220 sends uplink transmissions to network device 210 via beams 230-1 to 230-3. For ease of explanation, beams 230-1 to 230-3 or beams 240-1 to 240-3 will be collectively or individually referred to as beams 230 or beams 240, respectively.

[0043] It should be understood that the number of devices and their connections in Figure 2A are shown for purposes of illustration and are not intended to limit the present disclosure. Communication environment 200 may include any suitable number of network devices and / or terminal devices adapted to implement embodiments of the present disclosure.

[0044] In some embodiments, terminal device 220 and network device 210 may communicate with each other via channels, such as wireless communication channels over an air interface (e.g., a Uu interface). The wireless communication channels may include a PUCCH, a PUSCH, a physical random access channel (PRACH), a physical downlink control channel (PDCCH), a PDSCH, and a physical broadcast channel (PBCH). Of course, any other suitable channels may also be used.

[0045] Communications in communication environment 200 may conform to any suitable standard, including, but not limited to, Global System for Mobile Communications (GSM), Long Term Evolution (LTE), LTE-Evolution, LTE-Advanced (LTE-A), New Radio (NR), Wideband Code Division Multiple Access (WCDMA), Code Division Multiple Access (CDMA), GSM EDGE Radio Access Network (GERAN), Machine Type Communication (MTC), etc. Embodiments of the present disclosure may be performed in accordance with any currently known or future-developed generation of communication protocols. Examples of communication 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) communication protocols, 5.5G, 5G-Advanced networks, or sixth generation (6G) networks.

[0046] In some embodiments, multiple ML modes may be configured in terminal device 220. According to some example embodiments of the present disclosure, terminal device 220 may take some actions to schedule limited ML processing resources and / or control device temperature. FIG. 2B shows a signaling diagram illustrating a process 250 for communication according to some embodiments of the present disclosure. For ease of explanation, process 250 is described with reference to FIG. 2A. Process 250 may involve terminal device 220 and network device 210.

[0047] As shown in FIG. 2B , in some exemplary embodiments, terminal device 220 and network device 210 may selectively exchange capability-related information with each other (252). In the specific example of FIG. 2B , terminal device 220 may determine one or more priorities for the ML mode(s) and / or the number of ML processing resources for each ML model (254) and further schedule ML processing resources for the ML models (255). In addition to the above, coordination interactions may be performed between terminal device 220 and network device 210 (256). Specifically, terminal device 220 may transmit auxiliary information to network device 210 (260) and selectively receive adjustment settings from network device 210 (270). Details of these actions are described in more detail in the next section of this disclosure.

[0048] In some embodiments, the assistance information may be transmitted as UE assistance information (UAI). In one specific embodiment, the assistance information may be included in an information element (IE) called OverheatingAssistance.

[0049] In some embodiments, the auxiliary information (i.e., UAI) may include any suitable parameters used for device management and communication. Exemplary parameters include, but are not limited to: - a delay budget report conveying the desired increment / decrement of the connected mode discontinuous reception (DRX) cycle length; or -Overheating auxiliary information, or In-device coexistence interference avoidance (IDC) auxiliary information, or - preferences for DRX parameters for power savings, or - a preference for maximum aggregate bandwidth for power savings, or - a preference for a maximum number of secondary component carriers for power savings, or -Preference for maximum number of MIMO layers for power savings, or - a preference for minimum scheduling offset for cross-slot scheduling to save power, or - Preference for the RRC condition, or - Configuration grant assistance information for NR sidelink communications, or -Preference for providing reference time information, or - a preference for frequency range 2 (FR2) uplink gaps, or -Preference to transition from RRC_CONNECTED state for MUSIM operation, or -MUSIM Gap preference, or - Relaxed state for Radio Link Monitoring (RLM) measurements, or - Relaxed state for beam failure detection (BFD) measurements, or - the availability of data mapped to a radio bearer that is not configured for small data transmission (SDT), or - Preference to deactivate the SCG, or -Indicating that the UE has uplink data to transmit for data radio bearers (DRBs) for which no master cell group (MCG) radio link control (RLC) bearer exists while the SCG is deactivated, or -Change of implementation status for Radio Resource Management (RRM) measurement relaxation criteria

[0050] In some embodiments, the auxiliary information may be provided when a reporting condition is detected to be met, such as internal overheating being detected or no overheating condition being present, and may be transmitted by the terminal device 220 in the RRC_CONNECTED state.

[0051] In some embodiments, the UEAssistanceInformation message may be used to indicate assistance information to the network device 210 . As an example, when the auxiliary information transmission procedure is initiated, the terminal device 220 If configured to provide thermal auxiliary information, An overtemperature condition has already been detected and the reporting timer (e.g., T345) is not running, or If the current overheating assistance information is different from that indicated in the previous transmission of a UEAssistanceInformation message containing overheatingAssistance and timer T345 is not running: starting a timer T345 whose value is set as overheatingIndicationProhibitTimer; Initiate sending of UEAssistanceInformation message to provide overheating assistance information.

[0052] For better understanding, here are two examples of some IE OverheatingAssistance: JPEG2025526821000004.jpg110168

[0053] In the specific embodiment described above, the parameter "reducedBW-FR2" may be reported. The parameter "reducedBW-FR2" indicates a preference for a reduced configuration for terminal device 220, which corresponds to the maximum aggregate bandwidth across all downlink carrier(s) and all uplink carrier(s) of FR2-1. This can avoid overheating and achieve power savings. Additionally, in some embodiments, reporting of this field is permitted only if terminal device 220 is configured with serving cell(s) operating in FR2-1.

[0054] Additionally, in some embodiments, the aggregate bandwidth across all downlink carriers of FR2-1 is the sum of the bandwidths of the active downlink bandwidth portions (BWPs) across all activated downlink carriers of FR2-1, and the aggregate bandwidth across all uplink carriers of FR2-1 is the sum of the bandwidths of the active uplink BWPs across all activated uplink carriers of FR2-1.

[0055] In some embodiments, the absence of a field in the IE MaxBW-Preference or IE OverheatingAssistance is interpreted as indicating that the terminal device 220 has no preference for the maximum aggregated bandwidth of FR2-1. Thus, when instructed to address overheating, this maximum aggregated bandwidth includes the carrier(s) in FR2 of both the NR Master Cell Group (MCG) and the NR SCG. In some embodiments, this maximum aggregated bandwidth includes only the carriers in FR2-1 of the SCG in the (NG)EN-DC.

[0056] In some embodiments, when instructed to accommodate power savings, this maximum aggregate bandwidth includes the carrier(s) of FR2-1 of the cell group with which this UE assistance information is associated. Additionally, in some embodiments, when instructed to accommodate power savings, the aggregate bandwidth can only be up to the current active configuration.

[0057] Process Example Although features / operations are described individually in particular exemplary embodiments, it should be understood that these features / operations described in different exemplary embodiments can be used in any suitable combination unless expressly indicated to the contrary.

[0058] Furthermore, in the following description, several interactions (e.g., exchange of capability-related information, auxiliary information, configuration, etc.) are performed between terminal device 220 and network device 210. It should be understood that the interactions may be implemented with either a single signaling / message or multiple signaling / messages, including system information (SI), radio resource control (RRC) messages, downlink control information (DCI) messages, uplink control information (UCI) messages, medium access control (MAC) control elements (CEs), etc. The disclosure is not limited in this respect.

[0059] As described above, device temperature needs to be properly controlled. In some embodiments, if the terminal device 220 becomes too hot, it may request a reduction in transmission specifications, such as reducing MIMO layers, transmission bandwidth, etc., which directly leads to a decrease in data transmission speed / throughput. However, in a real communication environment, in addition to the data transmission itself, measurement, calculation, and maintenance operations also require significant resource consumption, which must also be considered as a major factor affecting device temperature. Furthermore, introducing AI / ML models into the device's air interface may consume a large amount of resources, exacerbating the overheating problem.

[0060] In summary, in a real communication environment, device temperature is affected by multiple factors, including but not limited to: Environmental factors: season (summer / winter), time of day (day / night), environmental temperature, etc. Communications-related factors: high data rate requirements, frequent measurements, intensive baseband / IF / RF processing, etc. Non-communication related factors: Resource-intensive applications running on the terminal device, such as video, Extended Reality (XR) / Augmented Reality (AR) / games, rendering tasks, etc., may require high data rates when online processing is required. AI / ML-related factors (if AI / ML functions are deployed): AI / ML for communication functions (e.g., AI for air interface), AI / ML for non-communication functions (e.g., AI for photos / videos / games), etc.

[0061] This disclosure considers the above factors more comprehensively. Next, examples of ML resource scheduling and device temperature control are detailed.

[0062] Process for scheduling ML processing resources The following process for scheduling ML processing resources is described with reference to Figures 2A and 2B.

[0063] In some embodiments, terminal device 220 may optionally report 252 the maximum ML processing capability supported by terminal device 220 as the UE capability.

[0064] In some embodiments, ML processing resources may be identified by multiple artificial intelligence (AI) processing units (APUs) for more granular control over ML processing resource scheduling.

[0065] In some embodiments, an APU is considered as a logical concept, which may be further represented by other metrics that evaluate AI computational power. In one particular exemplary embodiment, one APU corresponds to a certain amount of physical resources, such as the number of operations per second (OPS), the number of floating point operations per second (FLOPS), etc., e.g., 1 APU = 1 Tera OPS (TOPS).

[0066] In some embodiments, the number of APUs supported by terminal device 220 simultaneously or in a time period (N APU ) may be reported as the UE capability (i.e., maximum ML processing capability).

[0067] It will be appreciated that the period may be measured in any suitable unit of time, including but not limited to seconds, milliseconds, frames, slots, OFDM symbols, and the like.

[0068] In some embodiments, terminal device 220 supports multiple ML models, with at least one ML model currently running on terminal device 220. To rationally schedule ML processing resources, terminal device 220 determines (254) a respective priority for each ML model among the ML model(s) running on terminal device 220 and schedules (256) ML processing resources based on the determined priorities. This allows limited ML processing resources / capacity on terminal device 220 to be allocated to the most important ML model(s).

[0069] In some embodiments, the priority is determined based on at least one factor. Additionally, the priority may be determined based on at least one rule.

[0070] In some embodiments, the at least one rule is predefined. As an example, the at least one rule is predefined by a communications organization (such as 3GPP) or by a network operator or service provider. This allows the at least one rule to be applied as a default setting without requiring additional signaling exchange between the terminal device and the network device 210.

[0071] Alternatively, in some embodiments, the at least one rule may be dynamically or semi-statically configured, for example, either the terminal device 220 or the network device 210 may determine the at least one rule and then notify the other device of the at least one rule.

[0072] Additionally, in some embodiments, at least one rule is common to terminal device 220 and at least one other terminal device. Alternatively, in some embodiments, at least one rule is specific to terminal device 220. This allows for more flexible and customized priority rules.

[0073] The following example illustrates the factors and rules for determining priority.

[0074] Generally speaking, for a real-time ML model, the output may immediately affect subsequent transmissions and therefore must be processed first. Taking this into account, in some embodiments, the factor for determining the priority is the real-time requirements of the ML model. Specifically, if a first ML model is a real-time ML model and a second ML model is a non-real-time ML model, the first priority of the first ML model is higher than the second priority of the second ML model. For ease of understanding, an example of a real-time ML model is an ML model for demodulation, and an example of a non-real-time ML model is an ML model for mobility.

[0075] Alternatively or additionally, in some embodiments, a factor for determining the priority may be the latency of the ML model. Specifically, if a first ML model requires lower latency compared to a second ML model, the first priority of the first ML model is higher than the second priority of the second ML model. For ease of understanding, examples of ML models with low latency include an ML model for ultra-reliable low-latency communication (URLLC) and an ML model for XR, and one example of an ML model without low latency is an ML for enhanced mobile broadband (eMBB).

[0076] Alternatively or additionally, in some embodiments, a factor for determining the priority may be the collaboration level of the ML models. Specifically, if a first ML model requires a higher collaboration level compared to a second ML model, the first priority of the first ML model is higher than the second priority of the second ML model.

[0077] In some embodiments, the collaboration level may be one of: level x: no collaboration, level y: signaling-based collaboration without model transfer, and level z: signaling-based collaboration with model transfer. Additionally, the priority from high to low may be {level z, level y, level x}.

[0078] Generally speaking, if a UE report is based on an AI / ML model output, it should be processed first. With this in mind, in some embodiments, a factor for determining priority may be the report transmission requirements of the ML models. Specifically, if a first ML model requires report transmission and a second ML model does not require report transmission, the first priority of the first ML model is higher than the second priority of the second ML model.

[0079] In summary, the factors for determining the priority may be one or more specific actions of an actor corresponding to the ML model, where the actor is a function that receives the output from the model inference function and triggers or executes a corresponding action. It is understood that in other embodiments, other actions may be used to determine the priority based on the suggestions of the exemplary embodiments described above, and such modified embodiments should also be considered within the scope of the present disclosure.

[0080] Alternatively or additionally, in some embodiments, the factor for determining the priority may be an ML type, where the ML type is one of a two-sided ML model or a one-sided ML model. Specifically, if the ML type of the first ML model is a two-sided ML model and the ML type of the second ML model is a one-sided ML model, the first priority of the first ML model is higher than the second priority of the second ML model. For ease of understanding, one example of a two-sided ML model is CSI transmission, where CSI compression is performed in terminal device 220 and CSI recovery is performed in network device 210.

[0081] Alternatively or additionally, in some embodiments, a factor for determining priority may be the number of entities involved in the ML models. Specifically, if the number of entities involved in a first ML model is greater than the number of entities involved in a second ML model, the first priority of the first ML model is higher than the second priority of the second ML model. In other words, as the number of entities involved in an ML model increases, the priority increases, and an ML model involving more entities may be processed first.

[0082] Alternatively, or in addition, in some embodiments, the factor for determining the priority may be a function of the ML model. For ease of understanding, the rule may specify that the ML model for the BM has a higher priority compared to the ML model for the CSI.

[0083] Alternatively, in some embodiments, the rules may specify that an ML model for reference signal receiving power (RSRP) has a higher priority than an ML model that is not for RSRP.

[0084] Alternatively or additionally, in some embodiments, a factor for determining priority may be the ML group to which the ML model belongs. Specifically, different ML groups / lists may be identified by different model group IDs / model list IDs, and different ML models belonging to the same model group / list may be identified by the same model group ID / model list ID. Furthermore, a model group / list may correspond to a specific function. The priority of an ML model may be determined based on the different model group IDs / model list IDs and optional model IDs.

[0085] Alternatively or additionally, in some embodiments, a factor for determining the priority may be communication correlation. Specifically, if the first ML model is a communication-related ML model and the second ML model is a communication-irrelevant ML model, the first priority of the first ML model is higher than the second priority of the second ML model. Examples of communication-irrelevant ML models include an ML model for positioning, an ML model for photos, and an ML model for videos.

[0086] Alternatively or additionally, in some embodiments, a factor for determining the priority may be the accuracy requirement of the ML model. Specifically, if a first ML model requires higher accuracy (e.g., prediction accuracy) compared to a second ML model, the first priority of the first ML model is higher than the second priority of the second ML model.

[0087] Alternatively or additionally, in some embodiments, a factor for determining the priority may be a communication protocol layer associated with the ML model. In one particular embodiment, the communication protocol layer may be identified according to the OSI model, such as a physical layer, a data link layer, a network layer, etc. Alternatively, in another particular embodiment, the communication protocol layer may be identified according to a radio access network (RAN) classification, such as RAN1-related, RAN2-related, RAN3-related, etc.

[0088] It should be understood that the above factors are presented for illustrative purposes and do not imply any limitations. Terminal device 220 may determine the priority for an ML model based on any suitable factor. Furthermore, the above factors and other suitable factors may be used separately or in combination.

[0089] Additionally, when multiple factors are used to determine the priority, different weights may be set for different factors. In one particular embodiment, the weights of the factors may be sorted in a certain order, such as {delay, collaboration level, accuracy, ...}, so that different factors have different contributions in determining the priority, and the priority rules can be defined more rationally.

[0090] In one particular embodiment, the priority value may be calculated according to the following formula (1):

number

[0091] Additionally, a smaller priority value represents a higher priority, in other words, an ML model with a smaller priority value has a higher priority than an ML model with a larger priority value.

[0092] It should be understood that the above formula (1) is provided for illustrative purposes and does not imply any limitation. In other embodiments, the priority may be calculated by any suitable formula. The present description is not limited in this regard.

[0093] In addition to or as an alternative to the priority, the terminal device 220 may determine the respective number of ML processing resources for each ML model running on the terminal device 220, and then schedule the ML processing resources based on the respective number of ML processing resources.

[0094] Similar to the determination of priority, the respective number of ML processing resources for one ML model may also be determined based on at least one factor and / or at least one rule.

[0095] Additionally, similar to the rule for determining the priority, the rule for determining the respective numbers of ML processing resources for one ML model may also be predefined or may be set dynamically or semi-statically. Furthermore, the rule for determining the respective numbers of ML processing resources for one ML model may also be common to the terminal device 220 and at least one other terminal device, or may be specific to the terminal device 220.

[0096] Details for determining the respective numbers of ML processing resources for an ML model are described below.

[0097] In some embodiments, a factor for determining the number of ML processing resources for one ML model is the real-time requirement of the ML model. Specifically, if a first ML model is a real-time ML model and a second ML model is a non-real-time ML model, a first number of ML processing resources for the first ML model is greater than a second number of ML processing resources for the second ML model. For ease of understanding, an example of a real-time ML model is an ML model for demodulation, and an example of a non-real-time ML model is an ML model for mobility.

[0098] Alternatively or additionally, in some embodiments, a factor for determining the number of ML processing resources for one ML model may be the latency of the ML model. Specifically, if a first ML model requires lower latency compared to a second ML model, a first number of ML processing resources for the first ML model is higher than a second number of ML processing resources for the second ML model. For ease of understanding, examples of ML models with low latency include an ML model for URLLC and an ML model for XR, and an example of an ML model without low latency is ML for mobility.

[0099] Alternatively or additionally, in some embodiments, a factor for determining the number of ML processing resources for one ML model may be the collaboration level of the ML models. Specifically, if a first ML model requires a higher collaboration level compared to a second ML model, a first number of ML processing resources for the first ML model is higher than a second number of ML processing resources for the second ML model.

[0100] In some embodiments, the collaboration level may be one of: level x: no collaboration, level y: signaling-based collaboration without model transfer, and level z: signaling-based collaboration with model transfer. Additionally, the number of ML processing resources may be {level z, level y, level x} from most to least.

[0101] Generally speaking, if a UE report is based on an AI / ML model output, it should be more important. In consideration of this, in some embodiments, a factor for determining the number of ML processing resources for one ML model may be the report transmission requirements of the ML model. Specifically, if a first ML model requires report transmission and a second ML model does not require report transmission, the first number of ML processing resources for the first ML model is higher than the second number of ML processing resources for the second ML model.

[0102] Alternatively or additionally, in some embodiments, a factor for determining the number of ML processing resources for one ML model may be the ML type of the ML model, where the ML type is one of a two-sided ML model or a one-sided ML model. Specifically, when the ML type of a first ML model is a two-sided ML model and the ML type of a second ML model is a one-sided ML model, the first number of ML processing resources for the first ML model is higher than the second number of ML processing resources for the second ML model. For ease of understanding, one example of a two-sided ML model is CSI transmission, in which CSI compression is performed in terminal device 220 and CSI recovery is performed in network device 210.

[0103] Alternatively or additionally, in some embodiments, a factor for determining the number of ML processing resources for one ML model may be the number of entities involved in the ML model. Specifically, if the number of entities involved in a first ML model is greater than the number of entities involved in a second ML model, the first number of ML processing resources for the first ML model is higher than the second number of ML processing resources for the second ML model. In other words, as the number of entities involved in an ML model increases, the number of ML resources increases, and ML models involving more entities can be processed first.

[0104] Alternatively or additionally, in some embodiments, the factor for determining the number of ML processing resources for an ML model may be a function of the ML model. In one particular embodiment, the rule may specify that the ML model for the BM is allocated more processing resources than the ML model for the CSI. In another particular embodiment, the rule may specify a correspondence between the function of the ML model and the number of ML processing resources. For example, the rule may specify: a first number K1 of ML processing resources corresponding to the ML model for CSI compression; a second number K2 of ML processing resources corresponding to the ML model for spatial domain beam prediction; a third number K3 of ML processing resources corresponding to the ML model for time-domain beam prediction; A fourth number K4 of ML processing resources corresponding to the ML model for LOS / NLOS discrimination; a fifth number K5 of ML processing resources corresponding to the ML model for mobility; A sixth number K6 of ML processing resources corresponding to the ML model for demodulation may be defined.

[0105] In some embodiments, terminal device 220 may report the defined number to network device 210 as a UE capability.

[0106] Alternatively or additionally, in some embodiments, a factor for determining the number of ML processing resources for one ML model may be the ML group to which the ML model belongs. Specifically, different ML groups / lists may be identified by different model group IDs / model list IDs, and different ML models belonging to the same model group / list may be identified by the same model group ID / model list ID. Furthermore, a model group / list may correspond to a specific function. The number of ML processing resources for an ML model may be determined based on the different model group IDs / model list IDs and optional model IDs.

[0107] Alternatively or additionally, in some embodiments, a factor for determining the number of ML processing resources for one ML model may be communication correlation. Specifically, if a first ML model is a communication-related ML model and a second ML model is a communication-irrelevant ML model, a first number of ML processing resources for the first ML model is greater than a second number of ML processing resources for the second ML model. Examples of communication-irrelevant ML models include an ML model for positioning, an ML model for photos, and an ML model for videos.

[0108] Alternatively or additionally, in some embodiments, a factor for determining the number of ML processing resources for one ML model may be the accuracy requirement of the ML model. Specifically, if a first ML model requires higher accuracy (e.g., prediction accuracy) compared to a second ML model, a first number of ML processing resources for the first ML model is higher than a second number of ML processing resources for the second ML model.

[0109] Alternatively or additionally, in some embodiments, a factor for determining the number of ML processing resources for an ML model may be a communication protocol layer associated with the ML model. In one particular embodiment, the communication protocol layer may be identified according to the OSI model, such as a physical layer, a data link layer, a network layer, etc. Alternatively, in another particular embodiment, the communication protocol layer may be identified according to a radio access network (RAN) classification, such as RAN1-related, RAN2-related, RAN3-related, etc.

[0110] Alternatively, or in addition, in some embodiments, a factor for determining the number of ML processing resources for an ML model may be the input size of the ML model. Specifically, the rules may specify a correspondence between the number of ML processing resources and the input size (which may be a value or a range of values). In one particular embodiment, a first number of ML processing resources corresponds to a first input size range, and a second number of ML processing resources corresponds to a second input size range.

[0111] Additionally, the correspondence between the number of ML processing resources and the input size may follow a predefined function, including but not limited to a proportional function (e.g., there is a fixed ratio between the number of ML processing resources and the input size).

[0112] Additionally, in one particular exemplary embodiment, the input size of the ML model is: The size of the precoding matrix indicator (PMI) for channel state information (CSI) as an input for the ML model; The size of the channel matrix for CSI as input to the ML model, and the number of measurement instances for the CSI as input for the ML model (e.g., the number of latest measurement instances used as input for the CSI); is associated with at least one of

[0113] In the above specific exemplary embodiment, different numbers of ML processing resources may correspond to different sizes of PMI for CSI, different sizes of channel matrices for CSI, or different numbers of measurement instances for CSI, so that when an ML model is used for CSI, the ML processing resources can be appropriately allocated to the ML model.

[0114] Alternatively, or in addition, in another embodiment, the input size of the ML model is related to the number of measurement beams or measurement instances for beam management (e.g., the number of most recent measurement instances used as input for the BM). In this particular exemplary embodiment, different numbers of ML processing resources may correspond to different numbers of measurement beams or measurement instances for beam management. This allows ML processing resources to be appropriately allocated to the ML model when using an ML model for beam management.

[0115] It should be understood that the above input size examples are provided for illustrative purposes and do not imply any limitation. In other embodiments, the ML model is used for other functions, and the input size may vary according to the associated input parameters. The disclosure is not limited in this respect.

[0116] Alternatively, or in addition, in some embodiments, a factor for determining the number of ML processing resources for an ML model may be the output size of the ML model. Specifically, the rules may specify a correspondence between the number of ML processing resources and the output size (which may be a value or a range of values). In one particular embodiment, a first number of ML processing resources corresponds to a first range of output sizes, and a second number of ML processing resources corresponds to a second range of output sizes.

[0117] Additionally, the correspondence between the number of ML processing resources and the output size may follow a predefined function, including but not limited to a proportional function (e.g., there is a fixed ratio between the number of ML processing resources and the output size).

[0118] Additionally, in one particular exemplary embodiment, the output size is: the number of compression bits for CSI as the output of the ML model, the number of prediction instances for the CSI as the output of the ML model (e.g., the number of F predictions for F future time instances); is associated with at least one of

[0119] In the above specific example embodiment, different numbers of ML processing resources may correspond to different numbers of compressed bits for the CSI or different numbers of prediction instances for the CSI, so that when an ML model is used for the CSI, the ML processing resources can be appropriately allocated to the ML model.

[0120] Alternatively or additionally, in some embodiments, the output size is related to the number of predicted beams or prediction instances (e.g., number of F predictions for F future time instances) for beam management as output of the ML model, so that when using an ML model for beam management, ML processing resources can be appropriately allocated to the ML model.

[0121] It should be understood that the above example output sizes are provided for illustrative purposes and do not imply any limitation. In other embodiments, the ML model is used for other functions, and the output size may vary according to the associated output parameters. The disclosure is not limited in this respect.

[0122] Alternatively, or in addition, in some embodiments, a factor for determining the number of ML processing resources for an ML model may be the stage of lifecycle management in which the ML model is currently operating.

[0123] In some embodiments, the stage of lifecycle management may be one of ML model deployment, performance monitoring, feedback, updates, and other stages of lifecycle management.

[0124] Alternatively, or in addition, in some embodiments, a factor for determining the number of ML processing resources for an ML model may be the phase of the ML model.

[0125] In some embodiments, the phase of the ML model is one of a training phase, a validation phase, a testing phase, and an inference phase. As one particular exemplary embodiment, the rule may specify that the number of ML processing resources increases in the order of {inference phase, validation phase, testing phase, training phase}.

[0126] Alternatively, or in addition, in some embodiments, a factor for determining the number of ML processing resources for an ML model may be the structure of the ML model. Examples of ML model structures include convolutional neural networks (CNNs), recurrent neural networks (RNNs), transformers, and inceptions. Specifically, the rule may specify a correspondence between the number of ML processing resources and the structure of the ML model.

[0127] Other factors for determining the number of ML processing resources for one ML model include the algorithm of the ML model, the number of layers / neutrons / parameters / hyperparameters of the ML model, etc.

[0128] It should be understood that the above factors are presented for illustrative purposes and do not imply any limitations. Terminal device 220 may determine the respective number of ML processing resources for an ML model based on any suitable factor. Furthermore, the above factors and other suitable factors may be used separately or in combination.

[0129] Additionally, when multiple factors are used to determine the number of ML processing resources for an ML model, different weights may be set for different factors. In one particular embodiment, the weights of the factors may be sorted in a certain order, such as {delay, collaboration level, accuracy, ...}. This allows different factors to have different contributions in determining the number of ML processing resources for an ML model, and allows rules to be defined more rationally.

[0130] In one particular exemplary embodiment, the respective number of ML processing resources for one ML model may be calculated by the following equation (2):

number

[0131] It should be understood that the above formula (2) is provided for illustrative purposes and does not imply any limitation. In other embodiments, the respective numbers of ML processing resources for one ML model may be calculated by any suitable formula. In this regard, the present description is not limited.

[0132] In some embodiments, a first number of ML processing resources is determined for a first ML model for terminal device 220, and a second number of ML processing resources is determined for the same first ML model operating in another terminal device. In one exemplary embodiment, the first number of ML processing resources is different from the second number of ML processing resources. In other words, different terminal devices may have different numbers of ML processing resources for the same ML model.

[0133] Alternatively, in another exemplary embodiment, the first number of ML processing resources is the same as the second number of ML processing resources, but the first processing timing of the first number of ML processing resources is different from the second processing timing of the second number of ML processing resources.

[0134] Additionally, one exemplary processing timing is the period during which ML processing resources are occupied. Another example processing timing is the length of time used for one or more of ML model training, ML model validation, ML model testing, and ML model inference. Another example processing timing is the time offset between the trigger / input data collection of ML model training / validation / testing / inference and the respective output of ML model training / validation / testing / inference.

[0135] It should be understood that the above processing timing examples are presented for illustrative purposes and do not imply any limitation. In some other exemplary embodiments, the processing timings may be other existing processing timings or newly defined processing timings in the future. In this regard, the present description is not limited.

[0136] In some embodiments, terminal device 220 pauses the update procedure(s) for the subset of the at least one ML model. Additionally, the number of the subset of the at least one ML model is determined based on at least one of the maximum number of ML processing resources supported by the terminal device, the number of the at least one ML model, and a respective number of ML processing resources occupied by each ML model of the at least one ML model during the period.

[0137] JPEG2025526821000007.jpg64168

[0138] It should be understood that the above specific example for calculating the number of ML models to be paused is provided for illustrative purposes and does not imply any limitation. In other embodiments, other rules may be applied to calculate the number of ML models to be paused. In this regard, the present description is not limiting.

[0139] A process for coordinating specifications For ease of understanding, an exemplary process for adjusting the specification(s) will be described with reference to FIGS. 2A and 2B.

[0140] According to the following exemplary process, the device temperature and power consumption of the terminal device 220 can be controlled by adjusting the specifications (eg, ML model specifications and MIMO specifications).

[0141] In some embodiments, the specifications of the ML model are associated with the input size of the ML model. Alternatively or additionally, in some embodiments, the specifications of the ML model are associated with the output size of the ML model. Alternatively or additionally, in some embodiments, the specifications of the ML model are associated with the processing requirements of the ML model.

[0142] In some embodiments, the MIMO specification is associated with a MIMO measurement specification. Alternatively or additionally, in some embodiments, the MIMO specification is associated with a MIMO calculation specification. Alternatively or additionally, in some embodiments, the MIMO specification is associated with a MIMO maintenance specification.

[0143] In some embodiments, terminal device 220 generates auxiliary information for adjusting the specification(s) and transmits 260 the auxiliary information to network device 210.

[0144] As a general rule, if the device temperature or power consumption is high, the terminal device 220 may lower the associated specification(s), and if the device temperature or power consumption is low / normal, the terminal device 220 may raise the associated specification(s).

[0145] Furthermore, after adjusting the specification(s), the terminal device 220 may fall back to the previous / normal settings in response to some predefined event (e.g., expiration of a timer, or detection of high / low device temperature or loss of power consumption).

[0146] In one particular example, the terminal device 220 generates the auxiliary information in response to detecting that the temperature of the terminal device has risen to a first threshold temperature (e.g., detecting internal overheating).

[0147] Alternatively, in another particular example, terminal device 220 generates the auxiliary information in response to detecting that the temperature of the terminal device has dropped to a second threshold temperature.

[0148] Alternatively, in another particular example, terminal device 220 generates the aiding information in response to detecting that terminal device power consumption has increased to a first threshold consumption.

[0149] Alternatively, in another particular example, terminal device 220 generates the aiding information in response to detecting that the terminal device's power consumption has decreased to a second threshold consumption.

[0150] In another particular example, terminal device 220 starts a timer in response to applying the aiding information at the terminal device and falls back to the previous configuration in response to expiration of the timer.

[0151] Additionally, terminal device 220 may provide additional signaling in response to switching back to the previous configuration / normal configuration.

[0152] In this way, the terminal device 220 can control device temperature and power consumption within an acceptable range.

[0153] In some embodiments, the auxiliary information is suggestive information. In this case, network device 210 can generate an adjusted configuration for terminal device 220 based on the auxiliary information. Network device 210 can then transmit (270) the adjusted configuration to terminal device 220. In this way, if the specification(s) need to be adjusted, terminal device 220 can notify network device 210 of the expected specification(s), and at the same time, network device 210 can also control the final specification(s) adjusted for terminal device 220.

[0154] For ease of understanding, an example is given below: If configured, a terminal device 220 capable of providing aiding information can start transmitting aiding information upon a specific event (e.g., for overheating: upon detecting internal overheating or upon detecting that no overheating condition exists). Furthermore, upon starting the procedure, the terminal device 220: 1> If configured to provide auxiliary information, 2> The event has already been detected and the associated timer is not running, or 2> If the current aiding information is different from that indicated in the previous transmission of aiding information and the associated timer is not running, 3> starting an associated timer with a specific value of time length; 3> To provide the assistance information, the sending of assistance information messages should be initiated.

[0155] In some embodiments, the aiding information indicates the adjusted state currently in effect at the terminal device 220. In this way, adjustments at the terminal device 220 are timely. For ease of understanding, an example is described below. If configured, a terminal device 220 capable of providing aiding information can start transmitting aiding information upon a specific event (e.g., for overheating: upon detecting internal overheating or upon detecting that no overheating condition exists). Furthermore, upon initiating the procedure, the terminal device 220: 1> If configured to perform the associated action, 2> If the event has already been detected and the associated timer is not running, 3> starting an associated timer with a specific value of time length; 3> Perform the relevant action, 3> If it is configured to provide assistance information and the current assistance information is different from that indicated in the previous transmission of assistance information, it should start transmitting an assistance information message to provide the assistance information.

[0156] In some embodiments, the auxiliary information is associated with first information used to adjust the specifications of an ML model running on terminal device 220.

[0157] In some embodiments, the first information is: > Parameters used to stop or enable the training procedure of an ML model (e.g., to stop or enable data collection for training an AI / ML model); > Parameters used to stop or enable the inference procedure of an ML model (e.g., to stop or enable data collection for inference of an AI / ML model); > parameters used to stop or enable the ML model download procedure, > parameters used to stop or enable the ML model upload procedure, > parameters used to switch an ML model to a lite ML model. A lite ML model may refer to a simpler ML model (an ML model with fewer inputs / outputs / layers / neutrons / parameters / hyperparameters).

[0158] In this way, the processing requirements of the ML model are specified.

[0159] Alternatively or additionally, the first information indicates a parameter used to relax or tighten processing timing requirements for the inference procedure. In one specific exemplary embodiment, the terminal device 220 provides information to increase / decrease the inference timing, and the inference timing can be defined as the time from when the terminal device 220 collects measurement results as ML model inputs to when the terminal device 220 obtains outputs. It should be understood that the inference timing is an example of processing timing. In other embodiments, the processing timing can be defined as training timing, validation timing, or any other appropriate timing. This description is not limited in this respect. In this manner, specifications related to ML model processing timing can be adjusted.

[0160] Alternatively, or additionally, the first information indicates parameters used to reduce or increase the number of ML processing resources of the ML model (e.g., reduce or increase the computational size / resources of the ML model). More specifically, the terminal device 220 can decrease or increase the number of APUs in the AI / ML model or the number of APUs in all AI / ML models. In this way, the ML model computational resource related specifications can be adjusted.

[0161] Alternatively, or in addition, the first information indicates parameters used to suspend or restore lifecycle management for the terminal device 220. In one exemplary embodiment, lifecycle management of the ML model can be temporarily suspended or enabled by either the terminal device 220 or the network device 210.

[0162] In one particular embodiment, performance monitoring can be stopped or enabled, e.g., a timer or counter configured for performance monitoring can be stopped or started. In another particular embodiment, the monitoring period can be shortened or extended. In yet another particular embodiment, performance feedback can be stopped or enabled. In yet another particular embodiment, the feedback period can be shortened or extended. In yet another particular embodiment, updates / tweaks to the ML model can be frozen / restored.

[0163] Additionally, in some embodiments, the control / maintenance of the lifecycle management of the ML model may also depend on the accuracy of the AI / ML model.

[0164] In some embodiments, the first information is: The size of the precoding matrix indicator (PMI) for channel state information (CSI) as input for the ML model, The size of the channel matrix for CSI as input to the ML model, and > The number of measurement instances for CSI as input for the ML model (e.g., the number of latest measurement instances used as input for CSI).

[0165] In this way, the input size(s) specifications of the ML model for CSI (e.g., the size of the input data) can be adjusted.

[0166] Alternatively or additionally, in some embodiments, the first information indicates the number of measurement beams or measurement instances for beam management as input to the ML model (e.g., the number of latest measurement instances used as input for the BM).

[0167] In this way, the input size(s) specifications of the ML model for the BM (e.g., the size of the input data) can be adjusted.

[0168] In some embodiments, the first information is: the number of compression bits for CSI as the output of the ML model, > The number of prediction instances for the CSI as the output of the ML model (e.g., the number of F predictions for F future time instances).

[0169] In this way, the output size(s) specifications of the ML model for CSI (e.g., the size of the output data) can be adjusted.

[0170] Alternatively or additionally, in some embodiments, the first information indicates the number of predicted beams or prediction instances (e.g., the number of F predictions for F future time instances) for beam management as an output of the ML model. In this way, the output size(s) of the ML model for the BM (e.g., the size of the output data) can be adjusted.

[0171] In this way, the specification(s) of either the input size of the ML model or the output size of the ML model can be adjusted.

[0172] Additionally, in some embodiments, the auxiliary information is associated with second information used to adjust the MIMO specifications, particularly the second information is associated with at least one of a MIMO measurement specification, a MIMO calculation specification, and a MIMO maintenance specification.

[0173] In some embodiments, the second information is: > the number of reference signals (RS) to be measured, > number of RS reported, > RS transmission period, > RS reporting cycle, > number of receive beams measured (over all BWP / CC or per BWP / CC), > number of transmit beams measured (over all BWP / CCs or per BWP / CC), > Number of transmit / receive beam pairs measured (over all BWP / CCs or per BWP / CC), > The number of ports of the RS.

[0174] In this way, measurement-related specification(s) can be adjusted.

[0175] In some embodiments, the second information is: > Processing time for the physical downlink shared channel (PDSCH), > preparation time for the physical uplink shared channel (PUSCH), > The time offset between any two of the reference signal (RS) trigger, RS transmission, and RS reporting; > Beam application timing, > Beam switching timing, > A period for quasi-collocation (QCL), > Channel state information (CSI) calculation time, and > The number of machine learning (ML) processing resources for the CSI.

[0176] In this way, the computation-related specification(s) can be adjusted.

[0177] In some embodiments, the second information is: > number of activated transmission configuration indicator (TCI) states, > number of beams maintained, > The number of path loss (PL) reference signals (RS) that are maintained, and > The depth of the QCL chain.

[0178] In this way, the maintenance-related specifications can be adjusted.

[0179] In some embodiments, the second information is: > UL power, > Maximum power exposure (MPE), > Maximum output power reduction (MPR), > Power management maximum power reduction (P-MPR) > Additional maximum power reduction (A-MPR) > Duty cycle for uplink transmissions, > the number of sounding reference signal (SRS) resource sets; > number of panels used for uplink transmission, > number of panels used simultaneously for uplink transmission, > An indication of the switch from supporting STxMP to stopping STxMP support, > number of panels used for uplink transmission, > Maximum supported modulation and coding scheme (MCS), > Transport block size (TBS) > Quality of Service (QoS) > Quality of Experience (QoE) > Switch BWP instruction: For example, switch back to the initial BWP or switch to the BWP with the smallest bandwidth.

[0180] Processes for forecasting and adjusting For ease of understanding, an exemplary process for adjusting the specification(s) will be described with reference to FIGS. 2A and 2B.

[0181] According to the following exemplary process, terminal device 220 may predict temperature changes in the following period and mitigate the temperature changes by using auxiliary information, thereby balancing high data rates and device temperature.

[0182] For example, if the terminal device 220 predicts that the temperature at the terminal device 220 may become high, the terminal device 220 may adopt adaptive settings for at least data transmission, MIMO processing, and AI / ML processing based on the predicted event. Alternatively, the terminal device 220 may reduce other functions based on the prediction result to avoid a temperature increase.

[0183] In some embodiments, terminal device 220 predicts a temperature change in a subsequent time period, and if the temperature change satisfies the adjustment condition, terminal device 220 determines auxiliary information (e.g., according to an example process for adjusting the specification(s) described in this disclosure) and transmits (260) the auxiliary information to network device 210.

[0184] In some embodiments, the auxiliary information includes: First information used to adjust the specifications of an ML model running on a terminal device; and and second information used to adjust multiple-input multiple-output (MIMO) specifications.

[0185] In this way, temperature variations can be mitigated by adjusting the relevant specification(s).

[0186] Alternatively, in some embodiments, the auxiliary information is: > For example, low latency is required for online games, high data rates are required for online videos, and other factors related to temperature changes, > the amount of data being sent, > At least one transmission performance requirement for the subsequent period; and and third information for adjusting performance for data transmission in a subsequent period, such as, for example, the amount of expected data to be transmitted, the required data rate, the QoS or QoE, the delay, and the duration of the high data rate transmission.

[0187] In this way, the network device 210 can properly understand the current state of the terminal device 220 .

[0188] In some embodiments, the terminal device 220 obtains the aiding information from a predefined configuration that indicates a correspondence between temperature changes and at least one parameter set. The predefined configuration can enable adaptive transmission / measurement / calculation based on different (predicted) temperatures. Furthermore, by using the predefined configuration, explicit signaling regarding MIMO configuration and AI / ML processing does not need to be provided to the network device 210.

[0189] Table 2 below shows examples of predefined settings. [Table 2]

[0190] It should be understood that the above Table 2 is presented for illustrative purposes and does not imply any limitations, and the contents, number of rows, and number of columns of the above Table 2 may be changed in other embodiments.

[0191] Alternatively, the auxiliary information is determined using an ML model (e.g., an AI / ML model for a task scheduler), where the predicted / measured temperatures (and associated causes of temperature changes) are used as inputs to the ML model, and the auxiliary information is output by the ML model.

[0192] Additionally, as explained above, device temperature is affected by multiple factors, including but not limited to environmental factors, communication-related factors, non-communication-related factors, and AI / ML-related factors. Thus, the inputs may include one or more of the following: temperature range, UE location, time, current application state, buffer size, battery state, data rate, application data rate requirement, service type, etc. Furthermore, the above inputs are obtained during a historical time, such as a past period.

[0193] In some embodiments, the aiding information indicates a reasonable combination of transmit / MIMOI / AI settings for a future time, such as a future time period.

[0194] Alternatively, if the auxiliary information is determined using an ML model, the terminal device 220 may label the auxiliary information with an AI / ML prediction indication so that the network device 210 can understand that the auxiliary information is determined by an ML model.

[0195] Similar to the process for adjusting the specification(s), the auxiliary information generated in response to temperature changes may also be suggestive or indicative of the adjusted state currently in effect at the terminal device 220. For the sake of brevity, similar content is omitted here.

[0196] Example method 3 illustrates a flowchart of an exemplary method 300 according to some embodiments of the present disclosure. For example, the method 300 may be implemented in the terminal device 220 illustrated in FIG. 2A.

[0197] In block 310, the terminal device 220 determines a respective priority or a respective number of ML processing resources for each ML model among the at least one ML model running on the terminal device 220 based on at least one factor, the at least one factor including a real-time requirement of the ML model, a delay of the ML model, a collaboration level of the ML model, a report transmission requirement of the ML model, an ML type of the ML model, which is one of a two-sided ML model or a one-sided ML model, the number of entities involved in the ML model, the function of the ML model, the ML group to which the ML model belongs, the accuracy requirement of the ML model, or a communication protocol layer associated with the ML model.

[0198] In block 320, terminal device 220 schedules ML processing resources for at least one ML model based on the respective priorities or the respective numbers of ML processing resources.

[0199] In some embodiments, different factors of the at least one factor are set with different weights.

[0200] In some embodiments, the terminal device 220 determining the respective priorities or the respective numbers of ML processing resources includes determining the respective priorities or the respective numbers of ML processing resources according to at least one rule, wherein the at least one rule is that the first ML model is a real-time ML model and the second ML model is a non-real-time ML model; that the first ML model requires a lower delay compared to the second ML model; that the first ML model requires a higher collaboration level compared to the second ML model; that the first ML model requires a higher accuracy compared to the second ML model; that the ML type of the first ML model is a two-sided ML model and the ML type of the second ML model is a one-sided ML model; the number of entities involved in the first ML model is greater than the number of entities involved in the second ML model; the first ML model requires report transmission and the second ML model does not require report transmission; the first ML model requires higher accuracy compared to the second ML model; a communication protocol layer associated with the first ML model is lower than a second communication protocol layer associated with the second ML model; and the first ML model is a communication-related ML model and the second ML model is a communication-unrelated ML model, the first priority of the first ML model is higher than the second priority of the second ML model, or the first number of ML processing resources for the first ML model is greater than the second ML processing resources for the second ML model.

[0201] In some embodiments, the at least one rule may be predefined or configured dynamically or semi-statically.

[0202] In some embodiments, at least one rule is common with respect to terminal device 220 and at least one other terminal device 220, or at least one rule is specific to terminal device 220.

[0203] In some embodiments, terminal device 220 pauses the update procedure for a subset of at least one ML model according to their respective priorities.

[0204] In some embodiments, the number of subsets of the at least one ML model is determined based on at least one of the maximum number of ML processing resources supported by the terminal device 220, the number of the at least one ML model, and the respective numbers of ML processing resources occupied by each ML model of the at least one ML model during the period.

[0205] In some embodiments, when determining the respective number of ML processing resources for each ML model of the at least one ML model, the at least one factor further includes the input size of the ML model, the output size of the ML model, the structure of the ML model, or the stage of life cycle management in which the ML model is currently operating.

[0206] In some embodiments, the stage of lifecycle management may be one of ML model deployment, performance monitoring, feedback, updates, and other stages of lifecycle management.

[0207] In some embodiments, when determining the respective number of ML processing resources for each ML model of the at least one ML model, the at least one factor further includes the phase of the ML model, which is either a training phase, a validation phase, a testing phase, or an inference phase.

[0208] In some embodiments, the phase of the lifecycle management is one of a training phase, a validation phase, a testing phase, and an inference phase.

[0209] In some embodiments, a first number of ML processing resources is determined for a first ML model of the at least one ML model, and a second number of ML processing resources is determined for the same first ML model operating in another terminal device 220, and the first number of ML processing resources is different from or the same as the second number of ML processing resources.

[0210] In some embodiments, when the first number of ML processing resources is the same as the second number of ML processing resources, a first processing timing of the first number of ML processing resources is different from a second processing timing of the second number of ML processing resources.

[0211] In some embodiments, terminal device 220 transmits to network device 210 the maximum ML processing capability that terminal device 220 supports.

[0212] In some embodiments, the ML processing resources include multiple APUs.

[0213] 4 illustrates a flowchart of an exemplary method 400 according to some embodiments of the present disclosure. For example, the method 400 may be implemented in the terminal device 220 illustrated in FIG. 2A.

[0214] In block 410, the terminal device 220 generates auxiliary information associated with at least one of first information used to adjust specifications of an ML model running on the terminal device 220 and second information used to adjust MIMO specifications, the second information being associated with at least one of MIMO measurement specifications, MIMO calculation specifications, and MIMO maintenance specifications.

[0215] In block 420 , the terminal device 220 transmits the aiding information to the network device 210 .

[0216] In some embodiments, the auxiliary information is suggestive information, and the method further includes receiving from the network device 210 an adjustment setting to be used by the terminal device 220 .

[0217] In some embodiments, the aiding information indicates the adjusted state currently in effect at terminal device 220 .

[0218] In some embodiments, the terminal device 220 generates auxiliary information in response to detecting at least one of the following: the temperature of the terminal device 220 rising to a first threshold temperature; the temperature of the terminal device 220 falling to a second threshold temperature; the power consumption of the terminal device 220 increasing to a first threshold consumption; and the power consumption of the terminal device 220 decreasing to a second threshold consumption.

[0219] In some embodiments, terminal device 220 starts a timer in response to applying the aiding information at terminal device 220 and falls back to the previous configuration in response to expiration of the timer.

[0220] In some embodiments, the specifications of the ML model are related to at least one of the input size of the ML model, the output size of the ML model, and the processing requirements of the ML model.

[0221] In some embodiments, the first information indicates at least one of a parameter used to stop or enable a training procedure for the ML model, a parameter used to stop or enable an inference procedure for the ML model, a parameter used to stop or enable a download procedure for the ML model, a parameter used to stop or enable an upload procedure for the ML model, a parameter used to relax or tighten processing timing requirements for the inference procedure, a parameter used to reduce or increase the number of ML processing resources for the ML model, a parameter used to switch the ML model to a light ML model, and a parameter used to pause or restore lifecycle management for the terminal device 220.

[0222] In some embodiments, the first information is associated with at least one of: a size of the PMI for the CSI as an input to the ML model; a size of the channel matrix for the CSI as an input to the ML model; a number of measurement instances for the CSI as an input to the ML model; a number of compression bits for the CSI as an output of the ML model; a number of prediction instances for the CSI as an output of the ML model; a number of measurement beams or measurement instances for beam management as an input to the ML model; and a number of prediction beams or prediction instances for beam management as an output of the ML model.

[0223] In some embodiments, the second information indicates at least one of the number of RSs to be measured, the number of RSs to be reported, the transmission period of the RS, the reporting period of the RS, the number of receive beams to be measured, the number of transmit beams to be measured, the number of transmit / receive beam pairs to be measured, and the number of ports of the RS.

[0224] In some embodiments, the second information indicates at least one of: a processing time for a physical downlink shared channel (PDSCH), a preparation time for a physical uplink shared channel (PUSCH), a time offset between any two of an RS trigger, an RS transmission, and an RS report, a beam application timing, a beam switching timing, a period for quasi co-location (QCL), a channel state information (CSI) calculation time, a number of ML processing resources for CSI, a number of activated transmission configuration indicator (TCI) states, a number of maintained beams, a number of maintained path loss (PL) RSs, and a depth of the QCL chain.

[0225] In some embodiments, the second information indicates at least one of: a duty cycle for uplink transmission, a number of SRS resource sets, and a number of panels used for uplink transmission.

[0226] 5 illustrates a flowchart of an exemplary method 500 according to some embodiments of the present disclosure. For example, the method 500 may be implemented in the terminal device 220 illustrated in FIG. 2A.

[0227] In block 510, the terminal device 220 determines a temperature change in a subsequent time period and, if the temperature change satisfies an adjustment condition, determines auxiliary information to be used to mitigate the temperature change in the subsequent time period.

[0228] In block 520 , the terminal device 220 transmits the aiding information to the network device 210 .

[0229] In some embodiments, the auxiliary information is associated with at least one of a temperature change, a cause related to the temperature change, an amount of data to be transmitted, at least one transmission performance requirement for a subsequent period, first information used to adjust specifications of an ML model running in the terminal device, second information used to adjust specifications of MIMO, and third information for adjusting performance for data transmission for a subsequent period.

[0230] In some embodiments, terminal device 220 obtains the aiding information from a predefined configuration that indicates a correspondence between temperature changes and at least one parameter set.

[0231] In some embodiments, terminal device 220 determines the auxiliary information using an ML model.

[0232] In some embodiments, the aiding information is suggestive information, and terminal device 220 receives from network device 210 the adjusted settings to be used by terminal device 220 .

[0233] In some embodiments, the aiding information indicates the adjusted state currently in effect at terminal device 220 .

[0234] 6 illustrates a flowchart of an example method 600 according to some embodiments of the present disclosure. For example, the method 600 may be implemented in the network device 210 illustrated in FIG. 2A.

[0235] In block 610, the network device 210 receives auxiliary information from the terminal device 220 that is associated with at least one of first information used to adjust the specifications of an ML model running in the terminal device and second information used to adjust the specifications of MIMO.

[0236] In some embodiments, the auxiliary information is suggestive information, and the method further includes generating an adjustment setting to be used by the terminal device 220 based on the auxiliary information and transmitting the adjustment setting to the terminal device 220.

[0237] In some embodiments, the aiding information indicates the adjusted state currently in effect at terminal device 220 .

[0238] In some embodiments, the specifications of the ML model are related to at least one of the input size of the ML model, the output size of the ML model, and the processing requirements of the ML model.

[0239] In some embodiments, the first information indicates at least one of a parameter used to stop or enable a training procedure for the ML model, a parameter used to stop or enable an inference procedure for the ML model, a parameter used to stop or enable a download procedure for the ML model, a parameter used to stop or enable an upload procedure for the ML model, a parameter used to relax or tighten processing timing requirements for the inference procedure, a parameter used to reduce or increase the number of ML processing resources for the ML model, a parameter used to switch the ML model to a light ML model, and a parameter used to pause or restore lifecycle management for the terminal device 220.

[0240] In some embodiments, the first information is associated with at least one of: a size of the PMI for the CSI as an input to the ML model; a size of the channel matrix for the CSI as an input to the ML model; a number of measurement instances for the CSI as an input to the ML model; a number of compression bits for the CSI as an output of the ML model; a number of prediction instances for the CSI as an output of the ML model; a number of measurement beams or measurement instances for beam management as an input to the ML model; and a number of prediction beams or prediction instances for beam management as an output of the ML model.

[0241] In some embodiments, the MIMO specification is related to at least one of a MIMO measurement specification, a MIMO calculation specification, and a MIMO maintenance specification.

[0242] In some embodiments, the second information indicates at least one of the number of RSs to be measured, the number of RSs to be reported, the transmission period of the RS, the reporting period of the RS, the number of receive beams to be measured, the number of transmit beams to be measured, the number of transmit / receive beam pairs to be measured, and the number of ports of the RS.

[0243] In some embodiments, the second information indicates at least one of: a processing time for a physical downlink shared channel (PDSCH), a preparation time for a physical uplink shared channel (PUSCH), a time offset between any two of an RS trigger, an RS transmission, and an RS report, a beam application timing, a beam switching timing, a period for quasi co-location (QCL), a channel state information (CSI) calculation time, a number of ML processing resources for CSI, a number of activated transmission configuration indicator (TCI) states, a number of maintained beams, a number of maintained path loss (PL) RSs, and a depth of the QCL chain.

[0244] In some embodiments, the second information indicates at least one of: a duty cycle for uplink transmission, a number of SRS resource sets, and a number of panels used for uplink transmission.

[0245] 7 illustrates 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 network device 210 illustrated in FIG. 2A.

[0246] In block 710, the network device 210 receives auxiliary information from the terminal device 220, the auxiliary information being transmitted by the terminal device 220 in response to detecting that the predicted temperature change satisfies the adjustment condition, and being used to mitigate the temperature change of the terminal device 220 in a subsequent period.

[0247] In some embodiments, the auxiliary information is associated with at least one of a temperature change, a cause related to the temperature change, an amount of data to be transmitted, at least one transmission performance requirement for a subsequent period, first information used to adjust specifications of an ML model running in the terminal device, second information used to adjust specifications of MIMO, and third information for adjusting performance for data transmission for a subsequent period.

[0248] In some embodiments, the auxiliary information is suggestive information, and network device 210 generates an adjustment configuration to be used by terminal device 220 based on the auxiliary information and transmits the adjustment configuration to terminal device 220 .

[0249] In some embodiments, the aiding information indicates the adjusted state currently in effect at terminal device 220 .

[0250] Device example 8 is a schematic block diagram of an apparatus 800 suitable for implementing embodiments of the present disclosure. The apparatus 800 can be considered another exemplary implementation of the terminal device 220 or the network device 210 shown in FIG. 2A. Thus, the apparatus 800 can be implemented in, or as, at least a portion of the terminal device 220 or the network device 210.

[0251] As shown, the apparatus 800 comprises a processor 810, a memory 820 coupled to the processor 810, a suitable transmitter (TX) / receiver (RX) 840 coupled to the processor 810, and a communication interface coupled to the TX / RX 840. The memory 810 stores at least a portion of a program 830. The TX / RX 840 is for bidirectional communication. The TX / RX 840 has at least one antenna to facilitate communication, although in practice the access nodes referred to in this disclosure may have multiple antennas. The communication interface may represent any interface required for communication with other network elements, such as, for example, an X2 / Xn interface for bidirectional communication between eNBs / gNBs, an S1 / NG interface for communication between a Mobility Management Entity (MME) / Access and Mobility Management Function (AMF) / SGW / UPF and an eNB / gNB, an Un interface for communication between an eNB / gNB and a relay node (RN), or a Uu interface for communication between an eNB / gNB and a terminal device.

[0252] The program 830 is assumed to include program instructions that, when executed by an associated processor 810, cause the device 800 to operate in accordance with embodiments of the present disclosure, as described herein with reference to Figures 2A-7. Embodiments of the present disclosure may be implemented by computer software executable by the processor 810 of the device 800, by hardware, or by a combination of software and hardware. The processor 810 may be configured to implement various embodiments of the present disclosure. Furthermore, the combination of the processor 810 and the memory 820 may form a processing means 880 suitable for implementing various embodiments of the present disclosure.

[0253] Memory 820 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. While only one memory 820 is shown in device 800, device 800 may have multiple physically distinct memory modules. Processor 810 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. Device 800 may have multiple processors, such as application-specific integrated circuit chips that are time-slaved to a clock that synchronizes the main processor.

[0254] In some embodiments, the terminal device comprises circuitry configured to determine a respective priority or a respective number of ML processing resources for each ML model among at least one ML model running on the terminal device 220 based on at least one factor, and to schedule the ML processing resources for the at least one ML model based on the respective priority or the respective number of ML processing resources, wherein the at least one factor includes a real-time requirement of the ML model, a delay of the ML model, a collaboration level of the ML model, a report transmission requirement of the ML model, an ML type of the ML model, which is one of a two-sided ML model or a one-sided ML model, the number of entities involved in the ML model, a function of the ML model, an ML group to which the ML model belongs, an accuracy requirement of the ML model, or a communication protocol layer associated with the ML model.

[0255] In some embodiments, different factors of the at least one factor are set with different weights.

[0256] In some embodiments, the circuitry is further configured such that determining the respective priorities or the respective numbers of ML processing resources includes determining the respective priorities or the respective numbers of ML processing resources according to at least one rule, the at least one rule being: the first ML model is a real-time ML model and the second ML model is a non-real-time ML model; the first ML model requires a lower latency compared to the second ML model; the first ML model requires a higher collaboration level compared to the second ML model; the first ML model requires a higher accuracy compared to the second ML model; the ML type of the first ML model is a two-sided ML model and the ML type of the second ML model is a one-sided ML model; The method defines that a first priority of the first ML model is higher than a second priority of the second ML model, or a first number of ML processing resources for the first ML model is greater than a second ML processing resource for the second ML model, when at least one of the following conditions is met: the number of entities involved in the ML model is greater than the number of entities involved in the second ML model; the first ML model requires report transmission and the second ML model does not require report transmission; the first ML model requires higher accuracy compared to the second ML model; the communication protocol layer associated with the first ML model is lower than the second communication protocol layer associated with the second ML model; and the first ML model is a communication-related ML model and the second ML model is a communication-unrelated ML model.

[0257] In some embodiments, the at least one rule may be predefined or configured dynamically or semi-statically.

[0258] In some embodiments, at least one rule is common with respect to terminal device 220 and at least one other terminal device 220, or at least one rule is specific to terminal device 220.

[0259] In some embodiments, the circuitry is further configured to pause the update procedure for a subset of the at least one ML model according to their respective priorities.

[0260] In some embodiments, the number of subsets of the at least one ML model is determined based on at least one of the maximum number of ML processing resources supported by the terminal device 220, the number of the at least one ML model, and the respective numbers of ML processing resources occupied by each ML model of the at least one ML model during the period.

[0261] In some embodiments, when determining the respective number of ML processing resources for each ML model of the at least one ML model, the at least one factor further includes the input size of the ML model, the output size of the ML model, the structure of the ML model, or the stage of life cycle management in which the ML model is currently operating.

[0262] In some embodiments, the stage of lifecycle management may be one of ML model deployment, performance monitoring, feedback, updates, and other stages of lifecycle management.

[0263] In some embodiments, when determining the respective number of ML processing resources for each ML model of the at least one ML model, the at least one factor further includes the phase of the ML model, which is either a training phase, a validation phase, a testing phase, or an inference phase.

[0264] In some embodiments, a first number of ML processing resources is determined for a first ML model of the at least one ML model, and a second number of ML processing resources is determined for the same first ML model operating in another terminal device 220, and the first number of ML processing resources is different from or the same as the second number of ML processing resources.

[0265] In some embodiments, when the first number of ML processing resources is the same as the second number of ML processing resources, a first processing timing of the first number of ML processing resources is different from a second processing timing of the second number of ML processing resources.

[0266] In some embodiments, the circuitry is further configured to transmit to the network device 210 the maximum ML processing capability that the terminal device 220 supports.

[0267] In some embodiments, the ML processing resources include multiple APUs.

[0268] The terminal device comprises a circuit configured to generate auxiliary information associated with at least one of first information used to adjust specifications of an ML model running on the terminal device 220 and second information used to adjust MIMO specifications, and to transmit the auxiliary information to the network device 210, the second information being associated with at least one of MIMO measurement specifications, MIMO calculation specifications, and MIMO maintenance specifications.

[0269] In some embodiments, the auxiliary information is suggestive information, and the method further includes receiving from the network device 210 an adjustment setting to be used by the terminal device 220 .

[0270] In some embodiments, the aiding information indicates the adjusted state currently in effect at terminal device 220 .

[0271] In some embodiments, the circuitry is further configured to generate auxiliary information in response to detecting at least one of the following: a temperature of the terminal device 220 rising to a first threshold temperature; a temperature of the terminal device 220 decreasing to a second threshold temperature; a power consumption of the terminal device 220 increasing to a first threshold consumption; and a power consumption of the terminal device 220 decreasing to a second threshold consumption.

[0272] In some embodiments, the circuitry is further configured to start a timer in response to applying the aiding information at terminal device 220, and to fall back to a previous setting in response to expiration of the timer.

[0273] In some embodiments, the specifications of the ML model are related to at least one of the input size of the ML model, the output size of the ML model, and the processing requirements of the ML model.

[0274] In some embodiments, the first information indicates at least one of a parameter used to stop or enable a training procedure for the ML model, a parameter used to stop or enable an inference procedure for the ML model, a parameter used to stop or enable a download procedure for the ML model, a parameter used to stop or enable an upload procedure for the ML model, a parameter used to relax or tighten processing timing requirements for the inference procedure, a parameter used to reduce or increase the number of ML processing resources for the ML model, a parameter used to switch the ML model to a light ML model, and a parameter used to pause or restore lifecycle management for the terminal device 220.

[0275] In some embodiments, the first information is associated with at least one of: a size of the PMI for the CSI as an input to the ML model; a size of the channel matrix for the CSI as an input to the ML model; a number of measurement instances for the CSI as an input to the ML model; a number of compression bits for the CSI as an output of the ML model; a number of prediction instances for the CSI as an output of the ML model; a number of measurement beams or measurement instances for beam management as an input to the ML model; and a number of prediction beams or prediction instances for beam management as an output of the ML model.

[0276] In some embodiments, the second information indicates at least one of the number of RSs to be measured, the number of RSs to be reported, the transmission period of the RS, the reporting period of the RS, the number of receive beams to be measured, the number of transmit beams to be measured, the number of transmit / receive beam pairs to be measured, and the number of ports of the RS.

[0277] In some embodiments, the second information indicates at least one of: a processing time for a physical downlink shared channel (PDSCH), a preparation time for a physical uplink shared channel (PUSCH), a time offset between any two of an RS trigger, an RS transmission, and an RS report, a beam application timing, a beam switching timing, a period for quasi co-location (QCL), a channel state information (CSI) calculation time, a number of ML processing resources for CSI, a number of transmission configuration indicator (TCI) states, a number of maintained beams, a number of maintained path loss (PL) RSs, and a depth of a QCL chain.

[0278] In some embodiments, the second information indicates at least one of: a duty cycle for uplink transmission, a number of SRS resource sets, and a number of panels used for uplink transmission.

[0279] In some embodiments, the terminal device includes circuitry configured to determine a temperature change in a subsequent time period, and if the temperature change satisfies an adjustment condition, determine auxiliary information used to mitigate the temperature change in the subsequent time period, and transmit the auxiliary information to the network device 210.

[0280] In some embodiments, the auxiliary information is associated with at least one of a temperature change, a cause related to the temperature change, an amount of data to be transmitted, at least one transmission performance requirement for a subsequent period, first information used to adjust specifications of an ML model running in the terminal device, second information used to adjust specifications of MIMO, and third information for adjusting performance for data transmission for a subsequent period.

[0281] In some embodiments, the circuitry is further configured to obtain auxiliary information from a predefined configuration that indicates a correspondence between temperature changes and at least one parameter set.

[0282] In some embodiments, the circuitry is further configured to determine the auxiliary information using an ML model.

[0283] In some embodiments, the auxiliary information is suggestive information, and the circuitry is further configured to receive from the network device 210 an adjustment setting to be used by the terminal device 220 .

[0284] In some embodiments, the aiding information indicates the adjusted state currently in effect at terminal device 220 .

[0285] In some embodiments, the network device comprises circuitry configured to receive, from the terminal device 220, auxiliary information associated with at least one of first information used to adjust specifications of an ML model operating in the terminal device and second information used to adjust specifications of the MIMO.

[0286] In some embodiments, the auxiliary information is suggestive information, and the method further includes generating an adjustment setting to be used by the terminal device 220 based on the auxiliary information and transmitting the adjustment setting to the terminal device 220.

[0287] In some embodiments, the aiding information indicates the adjusted state currently in effect at terminal device 220 .

[0288] In some embodiments, the specifications of the ML model are related to at least one of the input size of the ML model, the output size of the ML model, and the processing requirements of the ML model.

[0289] In some embodiments, the first information indicates at least one of a parameter used to stop or enable a training procedure for the ML model, a parameter used to stop or enable an inference procedure for the ML model, a parameter used to stop or enable a download procedure for the ML model, a parameter used to stop or enable an upload procedure for the ML model, a parameter used to relax or tighten processing timing requirements for the inference procedure, a parameter used to reduce or increase the number of ML processing resources for the ML model, a parameter used to switch the ML model to a light ML model, and a parameter used to pause or restore lifecycle management for the terminal device 220.

[0290] In some embodiments, the first information is associated with at least one of: a size of the PMI for the CSI as an input to the ML model; a size of the channel matrix for the CSI as an input to the ML model; a number of measurement instances for the CSI as an input to the ML model; a number of compression bits for the CSI as an output of the ML model; a number of prediction instances for the CSI as an output of the ML model; a number of measurement beams or measurement instances for beam management as an input to the ML model; and a number of prediction beams or prediction instances for beam management as an output of the ML model.

[0291] In some embodiments, the MIMO specification is related to at least one of a MIMO measurement specification, a MIMO calculation specification, and a MIMO maintenance specification.

[0292] In some embodiments, the second information indicates at least one of the number of RSs to be measured, the number of RSs to be reported, the transmission period of the RS, the reporting period of the RS, the number of receive beams to be measured, the number of transmit beams to be measured, the number of transmit / receive beam pairs to be measured, and the number of ports of the RS.

[0293] In some embodiments, the second information indicates at least one of: a processing time for a physical downlink shared channel (PDSCH), a preparation time for a physical uplink shared channel (PUSCH), a time offset between any two of an RS trigger, an RS transmission, and an RS report, a beam application timing, a beam switching timing, a period for quasi co-location (QCL), a channel state information (CSI) calculation time, a number of ML processing resources for CSI, a number of transmission configuration indicator (TCI) states, a number of maintained beams, a number of maintained path loss (PL) RSs, and a depth of a QCL chain.

[0294] In some embodiments, the second information indicates at least one of: a duty cycle for uplink transmission, a number of SRS resource sets, and a number of panels used for uplink transmission.

[0295] In some embodiments, the network device comprises circuitry configured to receive auxiliary information from the terminal device 220, the auxiliary information being transmitted by the terminal device 220 in response to detecting that a predicted temperature change satisfies an adjustment condition, and used to mitigate a temperature change of the terminal device 220 in a subsequent period of time.

[0296] In some embodiments, the auxiliary information is associated with at least one of a temperature change, a cause related to the temperature change, an amount of data to be transmitted, at least one transmission performance requirement for a subsequent period, first information used to adjust specifications of an ML model running in the terminal device, second information used to adjust specifications of MIMO, and third information for adjusting performance for data transmission for a subsequent period.

[0297] In some embodiments, the auxiliary information is suggestive information, and the circuitry is further configured to generate adjustment settings to be used by the terminal device 220 based on the auxiliary information and to transmit the adjustment settings to the terminal device 220.

[0298] In some embodiments, the auxiliary information indicates the adjusted state currently in effect at the terminal device 220. As used in this disclosure, the term “circuitry” 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 another example, a circuit may be any portion of a hardware processor with software, where the hardware processor includes digital signal processor(s), software, and memory(s) that work together to cause a device, such as a terminal device or network device, to perform various functions. As yet another example, a circuit may be a hardware circuit and / or processor, such as a microprocessor or portion of a microprocessor, that requires software / firmware for operation, but the software may be absent when not necessary for operation. As used in this disclosure, the term circuitry also includes simply a hardware circuit or processor(s) or portion of a hardware circuit or processor(s) and its associated software and / or firmware implementation.

[0299] In summary, the embodiments of the present disclosure provide the following solutions:

[0300] In one solution, a communication method includes, in a terminal device supporting multiple machine learning (ML) models, determining a respective priority or a respective number of ML processing resources for each ML model among at least one ML model operating on the terminal device based on at least one factor including a real-time requirement of the ML model, a delay of the ML model, a collaboration level of the ML model, a report transmission requirement of the ML model, an ML type of the ML model being one of a two-sided ML model or a one-sided ML model, the number of entities involved in the ML model, a function of the ML model, an ML group to which the ML model belongs, an accuracy requirement of the ML model, or a communication protocol layer associated with the ML model, and scheduling the ML processing resources for the at least one ML model based on the respective priority or the respective number of ML processing resources.

[0301] In some embodiments, different factors of the at least one factor are set with different weights.

[0302] In some embodiments, determining the respective priorities or the respective numbers of ML processing resources includes determining the respective priorities or the respective numbers of ML processing resources according to at least one rule, wherein the at least one rule is: the first ML model is a real-time ML model and the second ML model is a non-real-time ML model; the first ML model requires a lower latency compared to the second ML model; the first ML model requires a higher collaboration level compared to the second ML model; the first ML model requires a higher accuracy compared to the second ML model; the ML type of the ML model of the first ML model is a two-sided ML model and the ML type of the ML model of the second ML model is a one-sided ML model; the first ML model requires higher accuracy compared to the second ML model; the communication protocol layer associated with the first ML model is lower than the second communication protocol layer associated with the second ML model; and the first ML model is a communication-related ML model and the second ML model is a communication-unrelated ML model, the first priority of the first ML model is higher than the second priority of the second ML model, or the first number of ML processing resources for the first ML model is greater than the second ML processing resources for the second ML model, when at least one of the following conditions is met: the number of entities involved in the first ML model is greater than the number of entities involved in the second ML model; the first ML model requires report transmission and the second ML model does not require report transmission; the first ML model requires higher accuracy compared to the second ML model; the communication protocol layer associated with the first ML model is lower than the second communication protocol layer associated with the second ML model; and the first ML model is a communication-related ML model and the second ML model is a communication-unrelated ML model.

[0303] In some embodiments, the at least one rule may be predefined or configured dynamically or semi-statically.

[0304] In some embodiments, at least one rule is common with respect to the terminal device and at least one other terminal device, or at least one rule is terminal device specific.

[0305] In some embodiments, the method further includes pausing the update procedure for a subset of the at least one ML model according to their respective priorities.

[0306] In some embodiments, the number of subsets of the at least one ML model is determined based on at least one of the maximum number of ML processing resources supported by the terminal device, the number of the at least one ML model, and the respective numbers of ML processing resources occupied by each ML model of the at least one ML model during a period of time.

[0307] In some embodiments, when determining the respective number of ML processing resources for each ML model of the at least one ML model, the at least one factor further includes the input size of the ML model, the output size of the ML model, the structure of the ML model, or the stage of life cycle management in which the ML model is currently operating.

[0308] In some embodiments, the stage of lifecycle management may be one of ML model deployment, performance monitoring, feedback, updates, and other stages of lifecycle management.

[0309] In some embodiments, when determining the respective number of ML processing resources for each ML model of the at least one ML model, the at least one factor further includes the phase of the ML model, which is either a training phase, a validation phase, a testing phase, or an inference phase.

[0310] In some embodiments, a first number of ML processing resources is determined for a first ML model of the at least one ML model, and a second number of ML processing resources is determined for the same first ML model operating in another terminal device, and the first number of ML processing resources is different from or the same as the second number of ML processing resources.

[0311] In some embodiments, when the first number of ML processing resources is the same as the second number of ML processing resources, a first processing timing of the first number of ML processing resources is different from a second processing timing of the second number of ML processing resources.

[0312] In some embodiments, the method further includes transmitting to the network device a maximum ML processing capability supported by the terminal device.

[0313] In some embodiments, the ML processing resources include multiple artificial intelligence (AI) processing units (APUs).

[0314] In one solution, a communication method includes generating, in a terminal device, auxiliary information associated with at least one of first information used to adjust specifications of an ML model running in the terminal device and second information used to adjust multiple input multiple output (MIMO) specifications, and transmitting the auxiliary information to a network device, wherein the second information is associated with at least one of MIMO measurement specifications, MIMO calculation specifications, and MIMO maintenance specifications.

[0315] In some embodiments, the auxiliary information is suggestive information, and the method further includes receiving, from the network device, an adjustment setting to be used by the terminal device.

[0316] In some embodiments, the auxiliary information indicates the adjusted state currently in effect at the terminal device.

[0317] In some embodiments, generating the auxiliary information includes generating the auxiliary information in response to detecting at least one of: a temperature of the terminal device rising to a first threshold temperature; a temperature of the terminal device decreasing to a second threshold temperature; a power consumption of the terminal device increasing to a first threshold consumption; and a power consumption of the terminal device decreasing to a second threshold consumption.

[0318] In some embodiments, the method further includes starting a timer in response to applying the aiding information at the terminal device, and falling back to a previous configuration in response to expiration of the timer.

[0319] In some embodiments, the specifications of the ML model are related to at least one of the input size of the ML model, the output size of the ML model, and the processing requirements of the ML model.

[0320] In some embodiments, the first information indicates at least one of a parameter used to stop or enable a training procedure for the ML model, a parameter used to stop or enable an inference procedure for the ML model, a parameter used to stop or enable a download procedure for the ML model, a parameter used to stop or enable an upload procedure for the ML model, a parameter used to relax or tighten processing timing requirements for the inference procedure, a parameter used to reduce or increase the number of ML processing resources for the ML model, a parameter used to switch the ML model to a light ML model, and a parameter used to pause or restore lifecycle management for the terminal device.

[0321] In some embodiments, the first information is associated with at least one of: a size of a precoding matrix indicator (PMI) for channel state information (CSI) as an input to the ML model; a size of a channel matrix for CSI as an input to the ML model; a number of measurement instances for CSI as an input to the ML model; a number of compression bits for CSI as an output of the ML model; a number of prediction instances for CSI as an output of the ML model; a number of measurement beams or measurement instances for beam management as an input to the ML model; and a number of prediction beams or prediction instances for beam management as an output of the ML model.

[0322] In some embodiments, the second information indicates at least one of the number of reference signals (RSs) to be measured, the number of RSs to be reported, the transmission period of the RSs, the reporting period of the RSs, the number of receive beams to be measured, the number of transmit beams to be measured, the number of transmit / receive beam pairs to be measured, and the number of ports of the RSs.

[0323] In some embodiments, the second information indicates at least one of a processing time for a physical downlink shared channel (PDSCH), a preparation time for a physical uplink shared channel (PUSCH), a time offset between any two of a reference signal (RS) trigger, an RS transmission, and an RS report, a beam application timing, a beam switching timing, a period for quasi co-location (QCL), a channel state information (CSI) calculation time, a number of machine learning (ML) processing resources for CSI, a number of transmission configuration indicator (TCI) states, a number of maintained beams, a number of maintained path loss (PL) reference signals (RS), and a depth of a QCL chain.

[0324] In some embodiments, the second information indicates at least one of a duty cycle for uplink transmission, a number of sounding reference signal (SRS) resource sets, and a number of panels used for uplink transmission.

[0325] In one solution, a communication method includes, in a terminal device, predicting a temperature change in a subsequent period; if the temperature change satisfies an adjustment condition, determining auxiliary information used to mitigate the temperature change in the subsequent period; and transmitting the auxiliary information to a network device.

[0326] In some embodiments, the auxiliary information is associated with at least one of a temperature change, a cause related to the temperature change, an amount of data to be transmitted, at least one transmission performance requirement for a subsequent period, first information used to adjust specifications of an ML model running in the terminal device, second information used to adjust specifications of multiple-input multiple-output (MIMO), and third information for adjusting performance for data transmission for a subsequent period.

[0327] In some embodiments, determining the auxiliary information includes obtaining the auxiliary information from a predefined configuration that indicates a correspondence between temperature changes and at least one parameter set.

[0328] In some embodiments, determining the auxiliary information includes determining the auxiliary information using an ML model.

[0329] In some embodiments, the auxiliary information is suggestive information, and the method further includes receiving, from the network device, an adjustment setting to be used by the terminal device.

[0330] In some embodiments, the auxiliary information indicates the adjusted state currently in effect at the terminal device.

[0331] In one solution, a communication method includes receiving, at a network device, auxiliary information from a terminal device that is associated with at least one of first information used to adjust specifications of an ML model running in the terminal device and second information used to adjust specifications of a multiple-input multiple-output (MIMO).

[0332] In some embodiments, the auxiliary information is suggestive information, and the method further includes generating an adjustment setting to be used by the terminal device based on the auxiliary information, and transmitting the adjustment setting to the terminal device.

[0333] In some embodiments, the auxiliary information indicates the adjusted state currently in effect at the terminal device.

[0334] In some embodiments, the specifications of the ML model are related to at least one of the input size of the ML model, the output size of the ML model, and the processing requirements of the ML model.

[0335] In some embodiments, the first information indicates at least one of a parameter used to stop or enable a training procedure for the ML model, a parameter used to stop or enable an inference procedure for the ML model, a parameter used to stop or enable a download procedure for the ML model, a parameter used to stop or enable an upload procedure for the ML model, a parameter used to relax or tighten processing timing requirements for the inference procedure, a parameter used to reduce or increase the number of ML processing resources for the ML model, a parameter used to switch the ML model to a light ML model, and a parameter used to pause or restore lifecycle management for the terminal device.

[0336] In some embodiments, the first information is associated with at least one of: a size of a precoding matrix indicator (PMI) for channel state information (CSI) as an input to the ML model; a size of a channel matrix for CSI as an input to the ML model; a number of measurement instances for CSI as an input to the ML model; a number of compression bits for CSI as an output of the ML model; a number of prediction instances for CSI as an output of the ML model; a number of measurement beams or measurement instances for beam management as an input to the ML model; and a number of prediction beams or prediction instances for beam management as an output of the ML model.

[0337] In some embodiments, the MIMO specification is related to at least one of a MIMO measurement specification, a MIMO calculation specification, and a MIMO maintenance specification.

[0338] In some embodiments, the second information indicates at least one of the number of reference signals (RSs) to be measured, the number of RSs to be reported, the transmission period of the RSs, the reporting period of the RSs, the number of receive beams to be measured, the number of transmit beams to be measured, the number of transmit / receive beam pairs to be measured, and the number of ports of the RSs.

[0339] In some embodiments, the second information indicates at least one of a processing time for a physical downlink shared channel (PDSCH), a preparation time for a physical uplink shared channel (PUSCH), a time offset between any two of a reference signal (RS) trigger, an RS transmission, and an RS report, a beam application timing, a beam switching timing, a period for quasi co-location (QCL), a channel state information (CSI) calculation time, a number of machine learning (ML) processing resources for CSI, a number of activated transmission configuration indicator (TCI) states, a number of maintained beams, a number of maintained path loss (PL) reference signals (RS), and a QCL chain depth.

[0340] In some embodiments, the second information indicates at least one of a duty cycle for uplink transmission, a number of sounding reference signal (SRS) resource sets, and a number of panels used for uplink transmission.

[0341] In one solution, the communication method includes receiving, in the network device, auxiliary information from the terminal device, the auxiliary information being transmitted by the terminal device in response to detecting that the predicted temperature change satisfies the adjustment condition, and being used to mitigate the temperature change of the terminal device in a subsequent period.

[0342] In some embodiments, the auxiliary information is associated with at least one of a temperature change, a cause related to the temperature change, an amount of data to be transmitted, at least one transmission performance requirement for a subsequent period, first information used to adjust specifications of an ML model running in the terminal device, second information used to adjust specifications of multiple-input multiple-output (MIMO), and third information for adjusting performance for data transmission for a subsequent period.

[0343] In some embodiments, the auxiliary information is suggestive information, and the method further includes generating an adjustment setting to be used by the terminal device based on the auxiliary information, and transmitting the adjustment setting to the terminal device.

[0344] In some embodiments, the auxiliary information indicates the adjusted state currently in effect at the terminal device.

[0345] In another solution, a communications device comprises a processor configured to cause the device to perform any of the above methods.

[0346] In general, 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. While various aspects of embodiments of the present disclosure are illustrated and described using block diagrams, flowcharts, or some other pictorial representations, it should be understood that the blocks, devices, systems, techniques, or methods described in this disclosure may be implemented in, by way of non-limiting example, hardware, software, firmware, special purpose circuits or logic, general purpose hardware or controller, or other computing device, or some combination thereof.

[0347] 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 (e.g., computer-executable instructions included in program modules) that execute on a device by a target real or virtual processor to perform the processes or methods described above with reference to FIGS. 2A-7. Generally, program modules include routines, programs, libraries, objects, classes, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The functionality of the program modules may be combined or split among program modules as desired in various embodiments. The machine-executable instructions for 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.

[0348] Program code for implementing 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 apparatus, so that when executed by the processor or controller, the program code performs the functions / acts specified in the flowcharts and / or block diagrams. The program code may be executed entirely on a machine, partially on a machine, as a stand-alone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0349] The program code may be embodied in a machine-readable medium, which may be any tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium includes, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination thereof. More specific examples of machine-readable storage media 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 disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0350] Furthermore, although operations are described in a particular order, this should not be understood as requiring such operations to be performed in the particular order shown, or sequentially, or that all of the operations shown be performed, to achieve desirable results. In certain situations, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above description, these should not be construed as limiting the scope of the disclosure, but rather as descriptions of features that may be specific to particular embodiments. Certain features that are described in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination.

[0351] Although the present disclosure has been described in language specific to structural features and / or methodological operations, it is to be understood that the present disclosure, which is limited to the appended claims, is not necessarily limited to the specific features or operations described above. Rather, the specific features and operations described above are disclosed as example forms of implementing the claims.

Claims

1. A method for a terminal device, comprising: generating auxiliary information used to indicate a specification of a machine learning (ML) model to be run on the terminal device; transmitting the auxiliary information to a network device; The auxiliary information is parameters used to stop or enable data collection for training the ML model; parameters used to stop or enable the inference procedure of the ML model; parameters used to stop or enable the download procedure of the ML model; parameters used to stop or enable the ML model upload procedure; parameters used to relax or tighten processing timing requirements for the inference procedure; A parameter used to reduce or increase the number of ML processing resources of the ML model; A parameter used to switch the ML model to a light ML model; and parameters used to suspend or restore life cycle management for the terminal device; Indicating at least one of Terminal device method.

2. The specifications of the ML model are: Measurements of the beam as input to the ML model, The predicted results of the beam as an output of the ML model, or the timing of the predictions; associated with at least one of The method of claim 1.

3. The auxiliary information is The size of the precoding matrix indicator (PMI) for channel state information (CSI) as an input of the ML model; the size of the channel matrix for the CSI as input to the ML model; the number of measurement instances for the CSI as input of the ML model; the number of compression bits for the CSI as output of the ML model; the number of prediction instances for the CSI as the output of the ML model; the number of measurement beams or measurement instances for beam management as input to the ML model, and the number of predicted beams or predicted instances for the beam management as output of the ML model; associated with at least one of The method of claim 1.

4. A network device including means for receiving, from a terminal device, auxiliary information used to indicate a specification of a machine learning (ML) model operated on the terminal device, The auxiliary information is parameters used to stop or enable data collection for training the ML model; parameters used to stop or enable the inference procedure of the ML model; parameters used to stop or enable the download procedure of the ML model; parameters used to stop or enable the ML model upload procedure; parameters used to relax or tighten processing timing requirements for the inference procedure; A parameter used to reduce or increase the number of ML processing resources of the ML model; A parameter used to switch the ML model to a light ML model; and parameters used to suspend or restore life cycle management for the terminal device; Indicating at least one of Network equipment.

5. The specifications of the ML model are: Measurements of the beam as input to the ML model, beam predictions as output of the ML model, or The timing of the prediction, associated with at least one of The network device according to claim 4 .

6. The auxiliary information is the size of a precoding matrix indicator (PMI) for channel state information (CSI) as an input of the ML model; the size of the channel matrix for the CSI as input to the ML model; the number of measurement instances for the CSI as input of the ML model; the number of compression bits for the CSI as output of the ML model; the number of prediction instances for the CSI as the output of the ML model; the number of measurement beams or measurement instances for beam management as input to the ML model, and the number of predicted beams or predicted instances for the beam management as output of the ML model; associated with at least one of The network device according to claim 4 .

7. A means for generating auxiliary information used to indicate the specifications of a machine learning (ML) model operated on a terminal device; means for transmitting the auxiliary information to a network device; A terminal device comprising: The auxiliary information is parameters used to stop or enable data collection for training the ML model; parameters used to stop or enable the inference procedure of the ML model; parameters used to stop or enable the download procedure of the ML model; parameters used to stop or enable the ML model upload procedure; parameters used to relax or tighten processing timing requirements for the inference procedure; A parameter used to reduce or increase the number of ML processing resources of the ML model; A parameter used to switch the ML model to a light ML model; and parameters used to suspend or restore life cycle management for the terminal device; Indicating at least one of Terminal device.

8. The specifications of the ML model are: Measurements of the beam as input to the ML model, beam predictions as output of the ML model, or The timing of the prediction, associated with at least one of The terminal device according to claim 7.

9. The auxiliary information is the size of a precoding matrix indicator (PMI) for channel state information (CSI) as an input of the ML model; the size of the channel matrix for the CSI as input to the ML model; the number of measurement instances for the CSI as input of the ML model; the number of compression bits for the CSI as output of the ML model; the number of prediction instances for the CSI as the output of the ML model; the number of measurement beams or measurement instances for beam management as input to the ML model, and the number of predicted beams or predicted instances for the beam management as output of the ML model; associated with at least one of The terminal device according to claim 7.

Images