Apparatus, method, and computer program

Abstract

The present disclosure relates to an apparatus configured to: receive (400), from another apparatus, an indication of performing radio beam prediction for cellular communication using a machine learning model, where the machine learning model is associated with a complexity value; determining (402) whether to perform radio beam prediction for cellular communications using a machine learning model based on the complexity value; and sending (404) an indication to other devices, the indication indicating whether the device has determined to perform radio beam prediction for cellular communications using the machine learning model.

Description

Technical Field

[0001] This disclosure relates to apparatus, methods, and computer programs for performing radio beam prediction for cellular communications in a communication system using machine learning models. Background Technology

[0002] A communication system can be viewed as a facility that enables a communication session between two or more entities (such as communication equipment, base stations, and / or other nodes) by providing carriers between various entities involved in the communication path.

[0003] A communication system can be a wireless communication system. Examples of wireless systems include Public Land Mobile Networks (PLMNs) that operate based on radio standards (such as those provided by 3GPP), satellite-based communication systems, and various wireless local area networks (WLANs). Wireless systems are typically divided into cells and are therefore often referred to as cellular systems.

[0004] Communication systems and associated equipment typically operate according to a given standard or specification that defines what the various entities associated with the system are allowed to do and how they should be implemented. The communication protocols and / or parameters used for the connection are also usually defined. An example of such a standard is the so-called 5G standard. Summary of the Invention

[0005] According to one aspect, an apparatus is provided, the apparatus comprising: at least one processor; and at least one memory storing instructions, which, when executed by the at least one processor, cause the apparatus to at least: receive from another apparatus an instruction to perform radio beam prediction for cellular communication using a machine learning model, wherein the machine learning model is associated with a complexity value; determine, based on the complexity value, whether to use the machine learning model to perform radio beam prediction for cellular communication; and send an instruction to another apparatus indicating whether the apparatus has determined to use the machine learning model to perform radio beam prediction for cellular communication.

[0006] The device can determine whether to use a machine learning model or switch to a different machine learning model.

[0007] The device can determine to use a machine learning model based on the fact that the complexity value of the machine learning model is lower than the threshold complexity value supported by the device.

[0008] The device can determine not to use a machine learning model if the complexity value of the machine learning model is higher than a threshold complexity value supported by the device.

[0009] The device can determine whether to not perform radio beam prediction, or to perform some or all of the radio beam prediction.

[0010] At least one memory storage instruction, when executed by at least one processor, causes the device to: determine, based on a complexity value, at least one of the following: a preference, capability, or lack thereof for using a machine learning model; and send an indication to other devices of a preference, capability, or lack thereof for using a machine learning model.

[0011] Indications of preference for using a machine learning model may specify at least one of the following: preference for using a different machine learning model with a complexity value different from that of the machine learning model; preference for using a different machine learning model with a specific complexity value; preference for using a specific machine learning model; preference for performing a specific radio beam prediction; preference for using a machine learning model with a different number of parameters; preference for using a machine learning model with a specific prediction period; preference for using a machine learning model with a different prediction period; preference for operating in an operating mode different from the current operating mode, which requires the use of a different machine learning model; or preference for operating in a specific operating mode that requires the use of a different machine learning model.

[0012] Indications for using a machine learning model preference may include: a preference for using a different machine learning model with a complexity value lower than that of the machine learning model, based on at least one of the following: the device is operating in a low-complexity operating mode, a low-computation operating mode, or a low-power operating mode; the device is operating in a low-mobility state; the device has its panel turned off; the device has computational consumption above a threshold; the device has power consumption above a threshold; the device is performing another task; or the device has a temperature above a threshold.

[0013] An indication of the ability or inability to use a machine learning model can specify at least one of the following: the ability to use a machine learning model; the inability to use a machine learning model; or the inability to perform radio beam prediction.

[0014] At least one memory may store instructions that, when executed by at least one processor, cause the device to: receive a machine learning model from another device, or information instructing the machine learning model.

[0015] At least one memory may store instructions that, when executed by at least one processor, cause the device to: receive instructions from other devices to perform radio beam prediction using a machine learning model or a different machine learning model based on preferences, capabilities, or lack thereof indicated by the device; perform radio beam prediction using a machine learning model or a different machine learning model; and report the radio beam prediction to other devices.

[0016] At least one memory may store instructions that, when executed by at least one processor, cause the apparatus to: receive from another apparatus an instruction to suspend or interrupt the use of a machine learning model, or a different machine learning model, to perform radio beam prediction; and to suspend or interrupt the use of a machine learning model, or a different machine learning model, to perform radio beam prediction.

[0017] At least one memory stores an instruction that, when executed by at least one processor, causes the device to: receive from another device an instruction to suspend or interrupt reporting radio beam predictions to the other device; and to suspend or interrupt reporting radio beam predictions to the other device.

[0018] A machine learning model can have a complexity value based on at least one of the following: the number of parameters of the machine learning model; the number of layers of the machine learning model; the number of rounds of the machine learning model; the power consumption of the machine learning model; or the computational cost of the machine learning model.

[0019] The complexity value can be an absolute complexity value or a relative complexity value.

[0020] Radio beam prediction may include at least one of the following: prediction of radio beams used for communication over a channel; or prediction of the quality of radio beams used for communication over a channel.

[0021] Indication of whether the device has determined to use a machine learning model to perform radio beam prediction for cellular communication can be transmitted via at least one of the following: Layer 1 signaling; Layer 2 signaling; or Layer 3 signaling.

[0022] The device may be a user equipment, and other devices may be a base station; or the device may be a base station distributed unit, and other devices may be a base station central unit.

[0023] According to one aspect, an apparatus is provided, the apparatus including components for: receiving from another device an instruction to perform radio beam prediction for cellular communication using a machine learning model, wherein the machine learning model is associated with a complexity value; determining, based on the complexity value, whether to use the machine learning model to perform radio beam prediction for cellular communication; and sending an instruction to another device indicating whether the device has determined to use the machine learning model to perform radio beam prediction for cellular communication.

[0024] According to one aspect, an apparatus is provided, the apparatus including a circuit system configured to: receive from another apparatus an instruction to perform radio beam prediction for cellular communication using a machine learning model, wherein the machine learning model is associated with a complexity value; determine, based on the complexity value, whether to use the machine learning model to perform radio beam prediction for cellular communication; and send an instruction to another apparatus indicating whether the apparatus has determined to use the machine learning model to perform radio beam prediction for cellular communication.

[0025] According to one aspect, a method is provided, the method comprising: receiving from another device an instruction by a device to perform radio beam prediction for cellular communication using a machine learning model, wherein the machine learning model is associated with a complexity value; determining, based on the complexity value, whether to use the machine learning model to perform radio beam prediction for cellular communication; and sending an instruction from the device to another device indicating whether the device has determined to use the machine learning model to perform radio beam prediction for cellular communication.

[0026] According to one aspect, a computer program is provided, the computer program including computer-executable instructions, which, when executed on at least one processor, are configured to: receive from another device an instruction to perform radio beam prediction for cellular communication using a machine learning model, wherein the machine learning model is associated with a complexity value; determine, based on the complexity value, whether to use the machine learning model to perform radio beam prediction for cellular communication; and send an instruction to another device indicating whether the device has determined to use the machine learning model to perform radio beam prediction for cellular communication.

[0027] According to one aspect, an apparatus is provided, the apparatus including at least one processor and at least one memory, the at least one memory including computer code for one or more programs, the at least one memory and the computer code being configured together with the at least one processor such that the apparatus at least: sends an instruction to another apparatus to perform radio beam prediction for cellular communication using a machine learning model, wherein the machine learning model is associated with a complexity value; and receives an instruction from the other apparatus indicating whether the other apparatus has determined, based on the complexity value, to perform radio beam prediction for cellular communication using a machine learning model.

[0028] At least one memory may store instructions that, when executed by at least one processor, cause the device to: determine, based on the instructions, whether another device uses a machine learning model, or a different machine learning model, to perform radio beam prediction; and send instructions to the other device to use a machine learning model, or a different machine learning model, to perform radio beam prediction.

[0029] At least one memory may store instructions that, when executed by at least one processor, cause the device to: receive from other devices an indication of preference, capability, or inability to perform radio beam prediction using a machine learning model.

[0030] At least one memory may store instructions that, when executed by at least one processor, cause the device to: send a machine learning model or instructions for a machine learning model to other devices.

[0031] At least one memory may store instructions that, when executed by at least one processor, cause the device to: receive from another device an instruction to suspend or interrupt the use of a machine learning model to perform radio beam prediction; or receive from another device an instruction to suspend or interrupt reports of radio beam prediction to other devices.

[0032] The device may be a base station, and other devices may be user equipment; or the device may be a central unit of a base station, and other devices may be distributed units of a base station.

[0033] According to one aspect, an apparatus is provided, the apparatus including components for: sending an instruction to another device to perform radio beam prediction for cellular communication using a machine learning model, wherein the machine learning model is associated with a complexity value; and receiving an instruction from the other device indicating whether the other device has determined, based on the complexity value, to perform radio beam prediction for cellular communication using a machine learning model.

[0034] According to one aspect, an apparatus is provided, the apparatus including a circuit system configured to: send an instruction to another device to perform radio beam prediction for cellular communication using a machine learning model, wherein the machine learning model is associated with a complexity value; and receive an instruction from the other device indicating whether the other device has determined, based on the complexity value, to perform radio beam prediction for cellular communication using a machine learning model.

[0035] According to one aspect, a method is provided, the method comprising: sending an instruction from a device to another device to perform radio beam prediction for cellular communication using a machine learning model, wherein the machine learning model is associated with a complexity value; and receiving an instruction from the other device indicating whether the other device has determined, based on the complexity value, to perform radio beam prediction for cellular communication using a machine learning model.

[0036] According to one aspect, a computer program is provided, the computer program including computer-executable instructions, which, when executed on at least one processor, are configured to: send an instruction from a device to another device to perform radio beam prediction for cellular communication using a machine learning model, wherein the machine learning model is associated with a complexity value; and receive an instruction from the other device indicating whether the other device has determined, based on the complexity value, to perform radio beam prediction for cellular communication using a machine learning model.

[0037] According to one aspect, a computer-readable medium is provided that stores program instructions for performing at least one of the methods described above.

[0038] According to one aspect, a non-transitory computer-readable medium is provided, on which program instructions are stored for performing at least one of the methods described above.

[0039] According to one aspect, a non-volatile tangible storage medium is provided, on which program instructions are stored for performing at least one of the methods described above.

[0040] Many different aspects have been described above. It should be understood that other aspects can be provided through any combination of two or more of the aspects mentioned above.

[0041] Various other aspects are also described in the following detailed description and the appended claims. List of abbreviations AF: Application Functions AI: Artificial Intelligence AMF: Access and Mobility Management Function API: Application Programming Interface BS: Base Station CSI: Channel State Information CU: Centralized Unit DL: Downlink DU: Distributed Unit gNB: gNodeB GSM: Global System for Mobile Communications HSS: Home Subscriber Server IoT: Internet of Things LTE: Long Term Evolution MAC: Media Access Control ML: Machine Learning MS: Mobile Station MTC: Machine Type Communication NEF: Network Open Functionality NF: Network Functions NN: Neural Network NR: New Radio NRF: Network Repository Functionality PDU: Packet Data Unit RAM: Random Access Memory (R)AN: (Radio) Access Network ROM: Read-Only Memory SMF: Session Management Function TR: Technical Report TS: Technical Specifications UE: User Equipment UMTS: Universal Mobile Telecommunications System 3GPP: Third Generation Partnership Project 5G: Fifth Generation 5GC: 5G Core Network 5GS: 5G system Attached Figure Description

[0042] The embodiments will now be described by way of example only with reference to the accompanying drawings, in which:

[0043] Figure 1 A schematic diagram of a 5G system is shown;

[0044] Figure 2 A schematic diagram of the control device is shown;

[0045] Figure 3 A schematic diagram of the user equipment is shown;

[0046] Figure 4 A block diagram is shown for a method of performing radio beam prediction for cellular communications in a communication system using a machine learning model;

[0047] Figure 5 A block diagram is shown for another method for performing radio beam prediction for cellular communications in a communication system using a machine learning model;

[0048] Figure 6 A schematic diagram of a non-volatile storage medium for storing instructions is shown. These instructions, when executed by the processor, allow the processor to perform... Figure 4 and Figure 5 One or more steps in the method. Detailed Implementation

[0049] In the following explanation, certain embodiments are described with reference to mobile communication devices capable of communicating via wireless cellular systems and mobile communication systems serving such mobile communication devices. Before explaining the exemplary embodiments in detail, refer to... Figure 1 , Figure 2 ,as well as Figure 3 Briefly explain some general principles of wireless communication systems, their access systems, and mobile communication devices to help understand the technology behind the described examples.

[0050] Figure 1 A schematic diagram of a 5G system (5GS) is shown. 5GS may include user equipment (UE), (radio) access network ((R)AN), 5G core network (5GC), one or more application functions (AF), and one or more data networks (DN).

[0051] 5G(R)AN may include one or more gNodeB (gNB) distributed unit functions connected to one or more gNodeB (gNB) centralized unit functions.

[0052] 5GC may include: Access and Mobility Management Function (AMF), Session Management Function (SMF), Authentication Server Function (AUSF), User Data Management (UDM), User Plane Function (UPF), and / or Network Open Function (NEF).

[0053] Figure 2 The diagram illustrates the control Figure 1 An example of a control device 200 for the functions of (R)AN or 5GC is shown. The control device may include at least one random access memory (RAM) 211a, at least one read-only memory (ROM) 211b, at least one processor 212, 213, and an input / output interface 214. At least one processor 212, 213 may be coupled to RAM 211a and ROM 211b. At least one processor 212, 213 may be configured to execute appropriate software code 215. Software code 215 may, for example, allow the execution of one or more steps to perform one or more aspects of this aspect. Software code 215 may be stored in ROM 211b. Control device 200 may be interconnected with another control device 200 that controls another function of the 5G(R)AN or 5GC. In some embodiments, each function of (R)AN or 5GC includes control device 200. In alternative embodiments, two or more functions of (R)AN or 5GC may share a control device.

[0054] Figure 3 An example of UE 300 is illustrated, such as Figure 1The UE 300 shown is an example of a device capable of transmitting and receiving radio signals. Non-limiting examples include user equipment, mobile station (MS) or mobile device (such as a mobile phone or so-called "smartphone"), computer equipped with a wireless interface card or other wireless interface facility (such as a USB dongle), personal data assistant (PDA) or tablet computer equipped with wireless communication capabilities, machine-type communication (MTC) device, cellular Internet of Things (CIoT) device, or any combination of these devices. The UE 300 can provide, for example, communication for carrying data. Communication can be one or more of voice, email, text messages, multimedia, data, machine data, etc.

[0055] UE 300 can receive signals via air or radio interface 307 through appropriate means for receiving, and can transmit signals via appropriate means for transmitting radio signals. Figure 3 In the diagram, the transceiver device is schematically represented by block 306. The transceiver device 306 can be provided, for example, by means of radio components and an associated antenna arrangement. The antenna arrangement can be located inside or outside the mobile device.

[0056] UE 300 may be equipped with at least one processor 301, at least one memory ROM 302a, at least one RAM 302b, and other possible components 303 for software and hardware-assisted execution of the tasks it is designed to perform, including controlling access to and communication with access systems and other communication devices. At least one processor 301 is coupled to RAM 302b and ROM 302a. At least one processor 301 may be configured to execute appropriate software code 308. The software code 308 may, for example, allow the execution of one or more aspects of this aspect. The software code 308 may be stored in ROM 302a.

[0057] The processor, storage device, and other related control devices can be mounted on a suitable circuit board and / or chipset. This feature is indicated by reference numeral 304. The device may optionally have a user interface, such as a keyboard 305, a touch-sensitive screen or touchpad, or a combination thereof. Depending on the type of device, one or more of a display, speaker, and microphone may optionally be provided.

[0058] One or more aspects of this disclosure relate to using machine learning (ML) models to perform predictions in communication systems.

[0059] One or more aspects of this disclosure relate to “Research on Artificial Intelligence (AI) / Machine Learning (ML) for NR Air Interface” (RP-213599), the contents of which are as follows. "Target 4.1 Target for SI or core part WI or test part WI The study targets 3GPP framework for AI / ML for air interface corresponding to each target use case (such as performance, complexity, and potential specification impact, etc.). Use cases to be focused: Initial set of use cases includes: - CSI feedback enhancements, e.g., overhead reduction, accuracy improvement, prediction [RAN1] the Beam management, e.g., beam prediction in time and / or spatial domain for overhead and latency reduction, beam selection accuracy improvement [RAN1] the Positioning accuracy enhancements for different scenarios, including e.g., scenarios with severe NLOS conditions [RAN1] Finalize representative sub-use cases for each use case for characterization and baseline performance evaluation by RAN#98 the Evaluation - AI / ML methods for selected sub-use cases need to be diverse enough to support various requirements on gNB-UE collaboration levels Note: The selection of use cases for this study is only intended to develop a framework for AI / ML application to air interface for these and other use cases. The selection itself is not intended to provide any indication of the prospects of any future specification project. the AI / ML models, terminology, and description, to identify generic and specific features for framework study: Definition phase to characterize AI / ML related algorithms and associated complexity: Model generation, e.g., model training (including input / output, pre-processing / post-processing, online / offline (as applicable)), model validation, model testing (as applicable) Inference operation, e.g., input / output, pre-processing / post-processing (as applicable) Identify various collaboration levels between UE and gNB related to selected use cases, e.g., - No collaboration: based only on implemented AI / ML algorithms without information exchange [for comparison purpose] the Various UE / gNB collaboration levels aiming at separate or joint ML operation. Characterize life cycle management of AI / ML models: e.g., model training, model deployment, model inference, model monitoring, model updating the Dataset(s) for training, validation, testing, and inference - Identify generic notation and terminology for AI / ML related functions, procedures, and interfaces the Note: Consider the work done for FS_NR_ENDC_data_collect when appropriate the For considered use cases: - Evaluate performance benefits of AI / ML based algorithms for agreed use cases in final representative set: Methods based on statistical models (from TR 38.901 and TR 38.857 [positioning]) for link level and system level simulations. - Extensions of 3GPP evaluation methodology should be considered as needed for better suitability for AI / ML based techniques. - Whether field data is optionally needed to further evaluate performance and robustness in real-world environment should be discussed as part of the study. - Common assumptions are needed in dataset construction for training, validation, and testing for selected use cases. ​ 1) ​ the ​ ​ ​ ​ ​ ​ Consider appropriate model training strategies, collaboration levels, and associated impacts. Consider the agreed-upon base AI models (multiple) used for calibration. The descriptions of the AI ​​models used for evaluation and the training methods should be reported for information and cross-checking purposes. o KPIs: Define the general KPIs and corresponding requirements for AI / ML operations. Determine the use case specificity of the selected use cases. KPIs and benchmarks. The performance, inference latency, and computational complexity of AI / ML-based algorithms should be compared with state-of-the-art baselines. The overhead, power consumption (including computation), memory storage, and other factors associated with enabling the corresponding AI / ML solutions should be considered. And hardware requirements (including for a given processing latency) and generalization capabilities. 2) Assess potential specification impacts, particularly for the agreed-upon use cases and general framework in the final representative set: o Regarding the PHY layer, for example, (RAN1) Considering various aspects, such as potential specifications for AI model lifecycle management, and the selection of... Related to the training, validation, and test dataset construction of use cases. Use case and collaboration level-specific specification impacts, such as new signaling, components used for training and validation data assistance, Auxiliary information, measurement and feedback o Regarding protocols, for example (RAN2) - work on RAN2 only begins after sufficient progress has been made in use case studies in RAN1. Based on the RAN1 input, considering various aspects such as capability indications, configuration, and control processes (training / push), this aspect is related to, for example, capability indications, configuration, and control processes. (Theory), and related to data and AI / ML model management. The specific specification impact based on the collaboration level for each use case. o Regarding interoperability and testability, for example, (RAN4) - RAN4 use case studies are only available in RAN1 and RAN2. Work begins after full progress. Verify AI / ML-based performance enhancements and ensure that UEs and gNBs with AI / ML meet or exceed existing minimum requirements. Requirements (where applicable) and testing framework Consider the needs and impacts of defining AI / ML processing capabilities. Note 1: Specific AI / ML models are not expected to be specified and are left to be implemented. User data privacy must be preserved. Note 2: Research on AI / ML for the air interface is based on the current RAN architecture and should not introduce new interfaces. mouth."

[0060] The agreement has been reached in the "Final Report of 3GPP TSG RAN WG1 #110bis-e v1.0.0" (R1-2210801), the contents of which are as follows. "protocol For BM-Case 1 with UE-side AI / ML models, investigate the potential canonical impacts of L1 signaling to report AI / ML to the NW. The following information about ML model inference · Multiple beams based on the output of AI / ML model inference · FFS: Predicted L1-RSRP corresponding to (multiple) beams · FFS: Other Information protocol For BM-Case 2 with UE-side AI / ML models, investigate the potential canonical impacts of L1 signaling to report AI / ML to the NW. The following information about ML model inference · Multiple beams of N (or more) future time instances based on AI / ML model inference output. o FFS: The value of N · FFS: Predicted L1-RSRP corresponding to (multiple) beams · Information regarding the timestamps corresponding to the beams in (multiple) reports. o FFS: Explicit or Implicit · FFS: Other Information protocol For BM-Case1 and BM-Case2 with UE-side AI / ML models, the study focuses on potential downward selection. The following are alternatives for model monitoring: · Option 1. UE-side model monitoring o UE monitoring (multiple) performance metrics o The UE makes (multiple) decisions regarding model selection, activation, deactivation, switching, and rollback. · Alternative 2. NW-side model monitoring o NW monitors (multiple) performance metrics o NW makes (multiple) decisions regarding model selection, activation, deactivation, switching, and rollback. · Option 3. Hybrid model monitoring o UE monitoring (multiple) performance metrics o NW makes (multiple) decisions regarding model selection, activation, deactivation, switching, and rollback. protocol Regarding NW-side model monitoring for network-side AI / ML models of BM-Case1 and BM-Case2, the following aspects are considered: Necessity and potential normative impact of the study: · UE report based on beam(s) measurements of the beam set indicated by the gNB. · Signaling, such as RRC-based, L1-based · Note: Performance, UE complexity, and power consumption should be considered.

[0061] The UE can be configured to perform radio beam prediction using an ML model. To do this, the UE can receive the ML model from a BS (or some other network node—this article uses a BS as an example, but it should be noted that other types of network nodes can be used instead) and can subsequently receive an instruction to use (i.e., activate) the ML model to perform radio beam prediction. However, conditions at the UE may change over time, and using the ML model or performing radio beam prediction may not always be possible or may be inefficient.

[0062] In the example, the UE may have limited computing power. The UE may need to perform another task that requires medium to large computing power (e.g., receiving DL communication or sending UL communication) while using an ML model to perform radio beam prediction.

[0063] In another example, the UE may have limited power capacity. The UE may need to perform another task that requires medium to high power capacity (e.g., carrier aggregation) while using an ML model to perform radio beam prediction.

[0064] In another example, the ML model may rely on measurements of a reference signal from the BS, and the configuration of the reference signal (e.g., the number and periodicity of the reference signal) may vary over time.

[0065] In another example, the UE can be configured to use an ML model to perform multiple different radio beam predictions for different reference signals, and higher-priority radio beam predictions may have already been performed.

[0066] In another example, the UE can be configured to use an ML model to perform radio beam prediction for one or more serving cells or carriers. The same radio beam prediction can be performed for the primary cell and one or more secondary cells. Separate radio beam predictions can be performed for the primary cell and one or more secondary cells. Radio beam predictions can exist for the primary cell and one or more secondary cells, as well as additional radio beam predictions performed for one or more other secondary cells.

[0067] In another example, the UE can operate in low-power modes (e.g., power-saving modes), medium-power modes, and high-power modes. The UE may tend to enter a low-power mode or a power mode with lower power consumption than the current power mode.

[0068] One or more aspects of this disclosure provide a mechanism that allows a UE to perform radio beam prediction in a more efficient manner by taking into account the fact that conditions at the UE may change over time.

[0069] The UE can operate in different operating modes.

[0070] The UE can operate in different complexity operation modes (e.g., low complexity operation mode, medium complexity operation mode, and high complexity operation mode).

[0071] The UE can operate in different computationally intensive operating modes (e.g., low computational consumption operating mode, medium computational consumption operating mode, and high computational consumption operating mode).

[0072] The UE can operate in different power consumption modes (e.g., low power consumption mode, medium power consumption mode, and high power consumption mode).

[0073] The UE can operate in different mobility states. For example, the UE can operate in low mobility state, medium mobility state, or high mobility state.

[0074] The UE can store ML models to perform radio beam prediction. Radio beam prediction in this paper can be prediction for cellular communication (i.e., radio communication in a cellular system). For example, radio beam prediction can include prediction of beams (e.g., beam identifiers) communicating over a channel. The channel can be a physical downlink control channel, a physical downlink shared channel, a physical uplink control channel, or a physical uplink shared channel. Radio beam prediction can include prediction of the quality of the beams communicating over the channel (e.g., the quality of channel state information or layer 1 measurements). Radio beam prediction can include prediction of channel quality information (e.g., channel state information (CSI)). Radio beam prediction can include prediction of mobility (e.g., trajectory, target cells, etc.) and channel quality information (e.g., CSI).

[0075] The UE can send an indication of the ML model to the BS. This indication can refer to the ML model using the ML model identifier.

[0076] The UE can send the set of complexity values ​​supported by the UE to the BS. The UE can also send the maximum complexity value supported by the UE to the BS. The BS can use either the set of complexity values ​​supported by the UE or the maximum complexity value supported by the UE to determine which ML model to provide to the UE.

[0077] In this specification, the terms “complexity value”, “complexity level”, “complexity indicator”, “complexity value indicator”, or “complexity index” are used interchangeably.

[0078] The complexity value can depend on multiple parameters of the ML model. ML model parameters can include: observation space input parameters, action space input parameters, and trainable parameters of the neural network, including weights and / or biases. ML models with high complexity values ​​can include a large number of parameters, while ML models with low complexity values ​​can include a small number of parameters. The listed parameters can increase the computational / deterministic / measured complexity value.

[0079] The complexity value can depend on the number of layers in the ML model. For example, a higher complexity value may indicate that the ML model has a higher number of layers than the ML model with a lower complexity value.

[0080] The complexity value can depend on the number of rounds of the ML model (i.e., the number of times the ML model's learning algorithm works across the entire training dataset). For example, for prediction, the dataset (the input dataset used for prediction) can consist of N input values ​​at earlier time steps tN…t-1, where t is the current time step. For example, in AI / ML spatial domain beam prediction, the dataset (the input dataset used for prediction) can consist of N L1-RSRP measurements and beam IDs at earlier time steps tN…t-1, where t is the current time step. The dataset can also include UE locations at time steps tN…t-1. In AI / ML temporal domain prediction, the dataset can consist of a sequence of measurements (e.g., L1-RSRP) and a sequence of identifiers (e.g., beam IDs associated with measurements) at earlier time steps tN, …, t-2, t-1, where t is the current time step (in some cases, the current time step is included). Furthermore, as an example of other parameters, the UE positions at time steps tN, ..., t-2, t-1 can be included in the dataset. Larger datasets for ML models can increase complexity, thus allowing ML models to have higher complexity values ​​(higher number of rounds).

[0081] The complexity value can depend on the estimated power consumption of the ML model. As an example, higher power consumption may imply a higher complexity value associated with the ML model.

[0082] The complexity value can depend on the estimated computational cost of the ML model.

[0083] The complexity value can be an absolute value or a relative value.

[0084] Complexity values ​​can be labeled or annotated for a specific parameter or set of parameters. For example, an ML model can be associated with a complexity value labeled "Power Consumption," indicating that the complexity value represents the power consumption of the ML model. Similarly, an ML model can be associated with a complexity value labeled "Parameter #x," indicating that the complexity value represents parameter #x.

[0085] In one example, the complexity value can be defined in a way that the UE can determine the joint complexity value for one or more ML models. The joint complexity value can indicate the joint complexity value when one or more ML models are used. As an example, the UE can have a complexity value constraint (or budget), and if the joint complexity value is within the UE's complexity value constraint, the UE can determine that it can use one or more ML models. In the example, the UE can indicate the complexity value constraint to the BS.

[0086] In one example, the complexity value can be determined as the sum of complexity values ​​(or the contribution of the complexity value of each parameter in the parameters associated with the ML model). In another example, a (single) complexity value can be used for the ML model.

[0087] In one example, the complexity value can depend on the ML model for the ML function. In the AI / ML beam management use case, when (multiple) ML models are deployed at the BS or UE, the BS or UE can predict the top K beams in the spatial or temporal domain. For different AI / ML functions (different K values ​​for the prediction of the top K beams), the BS or UE will use different models. The complexity value can be defined as the complexity associated with the ML model used for a specific AI / ML function.

[0088] The complexity value can be selected from a predefined set of values ​​known to both the UE and the BS. In this way, the UE and the BS can understand the complexity value.

[0089] Complexity values ​​can be determined, measured, assumed, or estimated through a reference. The reference can be another device (e.g., another UE). The reference can have specific capabilities. For example, the reference can have specific power consumption limits and / or computational cost limits. Higher computational cost can be associated with higher power consumption. The reference can have specific hardware.

[0090] A complexity value can be an absolute value determined, measured, assumed, or estimated by reference. A complexity value can also be a relative value relative to a specific complexity value determined, measured, assumed, or estimated by reference. A higher complexity value generally indicates a higher complexity of the ML model.

[0091] If a complexity value falls within a specific range (e.g., between X and Y), the complexity can be assumed to be at a specific complexity level. The complexity value can be quantized so that the (quantized) complexity value refers to a specific range of complexity values ​​(e.g., between X and Y). The (quantized) complexity value can be used for signaling between the UE and the BS.

[0092] An ML model can be identified by an ML model ID. An ML model ID can be specific to an ML model. When referencing a specific ML model, the ML model ID can be used by the UE and / or BS.

[0093] The ML model ID can be associated with a complexity value. The complexity value can be signaled together with the ML model ID. For example, if the ML model ID (e.g., 1) is signaled, the model ID can be signaled along with the complexity value. Alternatively, if the ML model ID (e.g., 1) is signaled, the association with the complexity value can be determined by the UE based on additional information. This additional information can be pre-configured. The UE can request this additional information from the BS. The BS can provide this additional information to the UE. This additional information can be provided as part of broadcast signaling (e.g., system information), non-access stratum signaling (i.e., signaling of higher-layer signaling of 3GPP RAN signaling), or in dedicated signaling.

[0094] In mobility use cases, when an ML model is deployed on the UE side, the complexity value can be associated with the ML model regarding when the UE is in the serving cell or when the UE is served by the target cell. When the UE is served by the serving cell, the UE can use the ML model ID#A to predict the candidate target cell and the top K beams of the candidate target cell. After the BS switches the UE to the candidate target cell, the UE can use different ML models with different complexity values.

[0095] The complexity value can be associated with the periodicity of the prediction. As an example, the complexity value X can be associated with the periodicity of the prediction for N predictions in each time period.

[0096] Complexity values ​​can indicate the periodicity of predictions.

[0097] Based on complexity values ​​and / or periodicity information, the UE can indicate the supported complexity values, preferred complexity values, and / or ML model IDs.

[0098] The complexity of radio beam prediction (e.g., determined by the UE) can depend on the number of cells or carriers (serving cells or carriers in carrier aggregation) that the UE performs radio beam prediction on. In one example, the UE can be configured with K cells or carriers and perform a total of M radio beam predictions (one radio beam prediction per cell or carrier, or one radio beam prediction for multiple cells or carriers). The UE can send an indication to the BS of the preferred number of cells or carriers, or the preferred number of radio beam predictions to operate for multiple cells or carriers (e.g., for complexity reduction).

[0099] The UE can request or indicate a specific prediction periodicity to the BS (i.e., a shorter prediction periodicity can mean more computation at the UE because the UE performs more predictions within the time period). The UE can request or indicate a prediction periodicity (absolute or relative to the "default" prediction periodicity) that supports or prefers a specific ML model ID. The preferred prediction periodicity can be referred to as a periodicity indication. The UE can indicate a preferred prediction periodicity per cell or carrier (or cell set or carrier set).

[0100] Periodicity indications can be used to allow the UE to indicate preferences that reduce the complexity of the current ML model. Periodicity indications can include: indications or requests for a specific prediction periodicity; indications or requests for changing from the current prediction periodicity to a new prediction periodicity N; and indications or requests for increasing or decreasing the current prediction periodicity by a factor N. The UE can indicate a preferred N value (e.g., a default value).

[0101] Periodicity indications can be used to reduce the complexity of the current ML model by indicating UE preferences. Periodicity indications can include: indications or requests for preferred prediction periodicity. Periodicity indications can also include: indications or requests for longer or shorter prediction periodicity.

[0102] Periodic indications can lead to a decrease or increase in complexity (e.g., in terms of reduced computation or power consumption).

[0103] Prediction indications can cause the BS to change the prediction periodicity for the UE. The BS can reject prediction indications. The BS can configure the UE to maintain prediction periodicity. The BS can configure the UE to suspend or interrupt radio beam prediction and report predictions.

[0104] The UE can receive from the BS one or more ML models with associated complexity values ​​that match the complexity values ​​supported by the UE, or information indicating one or more ML models with associated complexity values ​​that match the complexity values ​​supported by the UE.

[0105] Alternatively, the UE may receive from the BS one or more ML models with associated complexity values, or information indicating one or more ML models with associated complexity values, to allow the UE to determine one or more ML models with associated complexity values ​​that match the complexity values ​​supported by the UE.

[0106] For example, the ML model described in this paper can be a supervised learning model, a deep reinforcement learning model, or a continuous learning model.

[0107] The ML models described in this article can have different complexity values. That is, some ML models can have higher complexity values ​​than others.

[0108] The UE can receive instructions from the BS to perform predictions using (i.e., activate) the selected ML model.

[0109] The UE can receive from the BS an instruction to determine whether to use the selected ML model to perform radio beam prediction for cellular communication, to perform radio beam prediction based on the complexity value of the selected ML model and / or conditions at the UE, and send an instruction of preference, capability, or inability to use the selected ML model to perform radio beam prediction.

[0110] The UE can determine whether to use the selected ML model to perform prediction based on the complexity value of the selected ML model and / or conditions at the UE. The UE can also determine whether to use the selected ML model or switch to a different ML model based on the complexity value of the selected ML model and / or conditions at the UE. Furthermore, the UE can determine whether to not perform radio beam prediction, perform some or all of the radio beam prediction, based on the complexity value of the selected ML model and / or conditions at the UE. Conditions at the UE may include at least one of the following: operating mode, mobility status, number of on / off panels, computational consumption, need to perform another task, or temperature.

[0111] In implementation, the UE may determine at least one of the following based on the complexity value of the selected ML model and / or the conditions at the UE: preference, capability, or lack thereof for performing radio beam prediction.

[0112] The UE can send an indication to the BS of its preference, capability, or inability to perform radio beam prediction using the selected ML model.

[0113] Sending an indication to the BS or other network nodes, specifying whether the UE has determined to use the ML model for radio beam prediction, can directly or explicitly indicate whether the UE has determined to use the ML model or whether to perform radio beam prediction for cellular communication. Alternatively, the indication can indicate whether the UE has determined to perform radio beam prediction, which can implicitly indicate whether the UE has determined to use the selected ML model. That is, if the UE determines to perform radio beam prediction and therefore indicates this, it can also mean that the UE has determined to use the selected ML model. On the other hand, if the UE has determined not to perform radio beam prediction and therefore indicates this, it can mean that the UE has determined not to use the selected ML model.

[0114] Depending on the implementation, the instruction can indicate that the UE has determined to use the ML model based on the determination to use the ML model. Alternatively, the instruction can indicate that the UE has determined not to use the ML model based on the determination not to use the ML model.

[0115] The instruction can indicate a preference for using (i.e., switching to) a different ML model with a complexity value different from the selected ML model. For example, the instruction can indicate a preference for using a different ML model with a complexity value that is lower or higher than the selected ML model. The instruction can indicate a request to receive a different ML model. The instruction can indicate a request to use a different ML model instead of the selected ML model. The instruction can indicate a request to use a different ML model instead of the selected ML model for a specific time period (predefined or non-predefined). The instruction can instruct the UE to use a different ML model instead of the selected ML model. The instruction can instruct the UE to use a different ML model instead of the selected ML model for a specific time period (predefined or non-predefined).

[0116] In implementation, the indication may be based on at least one of the following, indicating a preference for using different machine learning models with a complexity value lower than that of the machine learning model: UE operating in a low-complexity operation mode, a low-computation operation mode, or a low-power operation mode; UE operating in a low-mobility state; UE with a closed panel; UE with computational consumption above a threshold; UE performing another other task; or UE with a temperature above a threshold.

[0117] The instruction can indicate a preference for using (i.e., switching to) a different ML model with a specific complexity value that is different from the complexity value of the selected ML model.

[0118] Indicators can indicate a preference for using a specific ML model that is different from the selected ML model.

[0119] The indication can specify a preference for operating in a different mode than the current operating mode, which requires the use of different ML models to perform radio beam prediction. For example, the indication could specify operating in a lower complexity mode, a lower computational cost mode, or a lower power consumption mode.

[0120] The indication can specify a preference for operating in a particular operating mode that requires the use of different ML models to perform predictions. For example, the indication can specify a preference for operating in a low-complexity operating mode, a low-computational-cost operating mode, or a low-power operating mode. This can indicate to the UE that it has the capability to perform radio beam prediction using the selected ML model, but may not have the capability to perform ML model monitoring.

[0121] The instruction can specify a preference for performing particular radio beam prediction. For example, the instruction can specify a preference for performing prediction for radio beams used in channel communication, but not for predicting the quality of radio beams used in channel communication.

[0122] An indication can specify a preference not to perform radio beam prediction. For example, such an indication could take the form of a specific complexity value not associated with any ML model. If such an indication is successfully provided, the UE may not be required to perform radio beam prediction.

[0123] The indicator can specify a preference for using a selected ML model with different numbers of parameters. For example, it can specify a preference for using a selected ML model with more or fewer parameters. Using more parameters can increase processing consumption and power consumption at the UE, but it can also increase prediction accuracy. Using fewer parameters can reduce processing consumption and power consumption at the UE, but it can also reduce the accuracy of radio beam prediction.

[0124] Instructions can indicate the ability to use ML models. Instructions can indicate the ability to use the selected ML model, while also indicating a preference for using different ML models with varying degrees of complexity.

[0125] The instruction can indicate the inability to use the selected ML model.

[0126] An indication can indicate the inability to perform prediction of a specific radio beam. For example, an indication can indicate the capability to perform prediction of a radio beam for communication over a channel, and the inability to perform prediction of the quality of a radio beam for communication over a channel.

[0127] An indication can indicate the inability to perform radio beam prediction. For example, an indication can take the form of a specific complexity value that is not associated with any ML model. When such an indication is successfully provided, the UE may not need to perform prediction.

[0128] Instructions can be part of supplementary information sent to the BS.

[0129] Instructions can be sent via Layer 1 signaling (e.g., random access channel, scheduling request, physical uplink control channel, or physical uplink shared channel).

[0130] Instructions can be sent via Layer 2 signaling (e.g., media access control control elements).

[0131] Instructions can be sent via Layer 3 signaling (e.g., radio resource control).

[0132] Instructions can be sent based on a timer. For example, a UE can start a timer when sending an instruction and can be prevented from sending another instruction until the timer expires.

[0133] The UE can receive instructions from the BS to use either the selected model or a newly selected ML model to perform radio beam prediction based on the UE's preferences, capabilities, or lack thereof as indicated to the BS. The UE can use either the selected ML model or a newly selected ML model to perform prediction.

[0134] Alternatively, the UE may receive an instruction from the BS to perform radio beam prediction using a selected model or a newly selected ML model, the radio beam prediction not based on the UE’s preference, capability, or lack thereof as indicated to the BS, and the UE may use the selected ML model or the newly selected ML model to perform radio beam prediction.

[0135] The UE can perform radio beam prediction using the selected ML model or a newly selected ML model, and can perform ML model monitoring (e.g., the UE operates in a low-complexity operating mode, a low-computational-cost operating mode, or a low-power operating mode). The UE can report the radio beam prediction to the BS.

[0136] The UE can perform radio beam prediction using the selected ML model or a newly selected ML model, and can also choose not to perform ML model monitoring (e.g., when the UE is operating in a medium or high complexity operating mode, a medium or high computational consumption operating mode, or a medium or high power consumption operating mode). The UE can report the radio beam prediction to the BS.

[0137] The UE can send an instruction to the BS requesting to suspend or interrupt (i.e., stop) the execution of radio beam prediction using the selected ML model or a newly selected ML model. The UE can suspend the execution of radio beam prediction using the selected ML model or a newly selected ML model for a period of time (e.g., X milliseconds), either semi-persistently or continuously. The instruction can be explicit or implicit (e.g., a specific complexity value sent to the BS can imply a request to suspend or interrupt the execution of radio beam prediction using the selected ML model or a newly selected ML model).

[0138] The UE may send an indication to the BS requesting a suspension or interruption (i.e., a halt) of radio beam prediction reporting to the BS. The UE may suspend or interrupt radio beam prediction reporting to the BS. The UE may suspend the prediction reporting to the BS semi-persistently or continuously for a period of time (e.g., X milliseconds). The indication may be explicit or implicit (e.g., a specific complexity value sent to the BS may imply a request to suspend or interrupt radio beam prediction reporting to the BS).

[0139] The UE can receive an instruction from the BS to suspend or interrupt (i.e., stop) the use of the selected ML model or a newly selected ML model in performing radio beam prediction. The UE can suspend or interrupt the use of the selected ML model or a newly selected ML model in performing radio beam prediction for a period of time (e.g., X milliseconds), either semi-persistently or continuously.

[0140] The UE can receive an instruction from the BS to suspend or interrupt reporting radio beam prediction to the BS. The UE can suspend or interrupt reporting radio beam prediction to the BS. The UE can suspend reporting radio beam prediction to the BS semi-persistently or continuously for a period of time (e.g., X milliseconds).

[0141] Although UE and BS have been considered above, it is understood that similar concepts can be implemented using alternative devices. For example, UE can be replaced by BS-DU (e.g., gNB-DU), and BS can be replaced by BS-CU (gNB-CU).

[0142] One or more aspects of this disclosure have the advantage of providing a mechanism that allows the UE to perform radio beam prediction using an ML model tailored to the conditions at the UE.

[0143] In appropriate circumstances (e.g., UEs operating in low-complexity, low-computation, or low-power modes; UEs operating in low-mobility states; UEs with their panels turned off; UEs with computational consumption exceeding a threshold; UEs with power consumption exceeding a threshold; UEs performing another task; or UEs with temperatures exceeding a threshold), UEs can use ML models with low complexity values ​​to perform radio beam prediction. In some cases, such as when the dataset size is small, this can have minimal impact on prediction accuracy while improving conditions at the UE.

[0144] Figure 4 A block diagram is shown for a method of using a machine learning model to perform radio beam prediction for cellular communications in a communication system.

[0145] In step 400, the device may receive from another device an instruction to perform radio beam prediction for cellular communication using an ML model, wherein the machine learning model is associated with a complexity value.

[0146] In step 402, the device may determine, based on the complexity value, whether to use an ML model to perform radio beam prediction for cellular communication.

[0147] In step 404, the device may send an instruction to other devices indicating whether the device has determined to use an ML model to perform radio beam prediction for cellular communications.

[0148] The device can determine whether to use an ML model or switch to a different ML model.

[0149] The device can determine to use an ML model based on the fact that the complexity value of the ML model is lower than the threshold complexity value supported by the device.

[0150] The device can determine not to use an ML model if the complexity value of the ML model is greater than the threshold complexity value supported by the device.

[0151] The device can determine whether to not perform radio beam prediction, or to perform some or all of the radio beam prediction.

[0152] The device can determine at least one of the following based on the complexity value: preference for using ML models, capability, or lack thereof.

[0153] The indication of preference for using an ML model specifies at least one of the following: preference for using a different ML model with a different complexity value than the ML model's complexity value; preference for using a different ML model with a specific complexity value; preference for using a specific ML model; preference for performing a specific radio beam prediction; preference for using an ML model with a different number of parameters; preference for using an ML model with a specific prediction period; preference for using an ML model with a different prediction period; preference for operating in an operating mode different from the current operating mode, which requires the use of a different ML model; or preference for operating in a specific operating mode that requires the use of a different ML model.

[0154] Indications for using a preference for an ML model may include: a preference for using a different ML model with a complexity value lower than that of the ML model, based on at least one of the following: the device is operating in a low-complexity operating mode, a low-computation operating mode, or a low-power operating mode; the device is operating in a low-mobility state; the device has its panel turned off; the device has computational consumption above a threshold; the device has power consumption above a threshold; the device is performing another task; or the device has a temperature above a threshold.

[0155] An indication of the ability or inability to use an ML model can specify at least one of the following: the ability to use an ML model; the inability to use an ML model; or the inability to perform radio beam prediction.

[0156] The device can receive ML models or information instructing ML models from other devices.

[0157] The device can receive instructions from other devices to perform radio beam prediction using an ML model or different ML models, based on preferences, capabilities, or lack thereof indicated by the device. The device can use an ML model or different ML models to perform radio beam prediction. The device can report radio beam predictions to other devices.

[0158] The device can receive instructions from other devices to pause or interrupt the use of an ML model or a different ML model to perform radio beam prediction. The device can pause or interrupt the use of a machine learning model or a different machine learning model to perform radio beam prediction.

[0159] The device can receive instructions from other devices to suspend or interrupt reporting radio beam predictions to other devices. The device can suspend or interrupt reporting radio beam predictions to other devices.

[0160] ML models can have a complexity value based on at least one of the following: the number of parameters of the ML model; the number of layers of the ML model; the number of rounds of the ML model; the power consumption of the ML model; or the computational cost of the ML model.

[0161] The complexity value can be an absolute complexity value or a relative complexity value.

[0162] Radio beam prediction may include at least one of the following: prediction of radio beams used for communication over a channel; or prediction of the quality of radio beams used for communication over a channel.

[0163] Indication of whether the device has determined to use a machine learning model to perform radio beam prediction for cellular communication can be transmitted via at least one of the following: Layer 1 signaling; Layer 2 signaling; or Layer 3 signaling.

[0164] The device can be a user equipment, and other devices can be a BS. The device can be a BS-DU, and other devices can be a BS-CU.

[0165] Figure 5 A block diagram is shown for another approach to performing radio beam prediction for cellular communications in a communication system using a machine learning model.

[0166] In step 500, the device may send an instruction to another device to perform radio beam prediction for cellular communication using an ML model, wherein the ML model is associated with a complexity value.

[0167] In step 502, the device may receive an instruction from another device indicating whether the other device has determined, based on a complexity value, to use an ML model to perform radio beam prediction for cellular communication.

[0168] The device can determine, based on an instruction, whether other devices use an ML model or a different ML model to perform radio beam prediction. The device can send instructions to other devices regarding whether to use an ML model or a different ML model to perform radio beam prediction.

[0169] The device can receive indications from other devices regarding preferences, capabilities, or inabilities for performing radio beam prediction using ML models.

[0170] The device can send ML models or instructions for ML models to other devices.

[0171] The device can receive instructions from other devices to suspend or interrupt the use of ML models to perform radio beam prediction. The device can also receive instructions from other devices to suspend or interrupt reports of radio beam prediction to other devices.

[0172] The device can be a BS, and other devices can be UEs. The device can be a BS CU, and other devices can be BS DUs.

[0173] Figure 6 A schematic diagram of a non-volatile storage medium 600 is shown, which stores instructions and / or parameters that, when executed by a processor, cause the processor to perform... Figure 4 and Figure 5 One or more steps in the method.

[0174] It should be noted that although exemplary embodiments have been described above, various changes and modifications can be made to the disclosed solutions without departing from the scope of the invention.

[0175] It is understandable that, although the above concepts are discussed in the context of 5GS, one or more of these concepts can be applied to other cellular systems.

[0176] Therefore, these embodiments can vary within the scope of the appended claims. Generally, some embodiments can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. For example, some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device, but the embodiments are not limited thereto. Although various embodiments may be illustrated and described as block diagrams, flowcharts, or using some other graphical representation, it is well understood that, by way of non-limiting example, the blocks, apparatuses, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0177] These embodiments can be implemented by computer software stored in memory and executable by at least one data processor of the entity involved, or by hardware, or by a combination of software and hardware. Furthermore, it should be noted in this regard that any process (e.g., Figure 4 and Figure 5 The process in the program can represent program steps, or interconnected logic circuits, blocks and functions, or a combination of program steps and logic circuits, blocks and functions. Software can be stored on physical media, such as memory chips or memory blocks implemented within a processor, magnetic media such as hard disks or floppy disks, and optical media such as DVDs and their data variants CDs.

[0178] The memory can be of any type suitable for the local technical environment and can be implemented using any suitable data storage technology, such as semiconductor-based memory devices, magnetic memory devices and systems, optical memory devices and systems, fixed memory, and removable memory. The data processor can be of any type suitable for the local technical environment and, by way of non-limiting example, can include one or more of the following: general-purpose computers, special-purpose computers, microprocessors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), gate-level circuits, and processors based on multi-core processor architectures.

[0179] Alternatively or additionally, some embodiments may be implemented using a circuit system. This circuit system may be configured to perform one or more of the functions and / or method steps previously described. This circuit system may be located in a base station and / or communication equipment.

[0180] As used in this application, the term "circuit system" may refer to one or more of the following: (a) Pure hardware circuit implementation (such as implementation using only analog and / or digital circuit systems); (b) A combination of hardware circuitry and software, for example: (i) A combination of (multiple) analog and / or digital hardware circuits and software / firmware, and (ii) Any portion of the hardware processor(s) having software (including (multiple) digital signal processors), software, and (multiple) memories, working together to enable an apparatus (such as a communication device or base station) to perform the various functions previously described; and (c) (Multiple) hardware circuits and / or (multiple) processors, such as (multiple) microprocessors or portions of (multiple) microprocessors, which require software (e.g., firmware) to operate, but the software may be absent when operation is not required.

[0181] The definition of "circuit system" applies to all uses of the term in this application, including in any claim. As another example, as used in this application, the term "circuit system" also encompasses only hardware circuitry or a processor (or multiple processors) or portions of hardware circuitry or a processor and its accompanying software and / or firmware implementation. The term "circuit system" also encompasses, for example, integrated devices.

[0182] The foregoing description provides a complete and informative description of some embodiments through exemplary and non-limiting examples. However, when read in conjunction with the accompanying drawings and appended claims, various modifications and adjustments will become apparent to those skilled in the art in light of the above description. Nevertheless, all such and similar modifications to the teachings will still fall within the scope defined in the appended claims.

Claims

1. An apparatus comprising: At least one processor; as well as At least one memory storing instructions, which, when executed by the at least one processor, cause the means to: Receive instructions from another device to perform radio beam prediction for cellular communication using a machine learning model, wherein the machine learning model is associated with a complexity value; Based on the complexity value, determine whether to use the machine learning model to perform the radio beam prediction for cellular communication; as well as Send instructions to other devices, indicating whether the devices have determined to use the machine learning model to perform the radio beam prediction for cellular communications.

2. The apparatus of claim 1, wherein the at least one memory stores instructions that, when executed by the at least one processor, cause the apparatus to: Based on the complexity value, determine at least one of the following: preference, ability, or lack of ability to use the machine learning model; and Send the indication of the preference, capability, or lack thereof using the machine learning model to the other devices.

3. The apparatus of claim 2, wherein the indication of the preference using the machine learning model specifies at least one of the following: The preference is to use different machine learning models with a complexity value that is different from the complexity value of the machine learning model. Prefer to use different machine learning models with specific complexity values; Prefer to use specific machine learning models; Prefer to perform specific radio beam prediction; The preference is to use the machine learning model with a different number of parameters; The preference is to use the machine learning model with a specific prediction period; The preference is to use the machine learning model with different prediction periods; The preference is to operate in an operating mode that is different from the current operating mode, and the different operating modes require the use of different machine learning models; or The preference is to operate in specific operational modes that require the use of different machine learning models.

4. The apparatus of claim 3, wherein the indication of the preference using the machine learning model comprises: The preference for using different machine learning models with a complexity value lower than the complexity value of the machine learning model, based on at least one of the following: The device operates in a low-complexity operation mode, a low-computation operation mode, or a low-power operation mode. The device operates in a state of low mobility; The device has its panel closed; The device has a computational cost that exceeds a threshold. The device has a power consumption higher than a threshold. The device performs another task; or The device has a temperature above a threshold.

5. The apparatus of claim 3 or claim 4, wherein the indication of capability or inability using the machine learning model indicates at least one of the following: The ability to use the machine learning model; The inability to use the machine learning model described above; or The inability to perform the aforementioned radio beam prediction.

6. The apparatus according to any one of claims 1 to 5, wherein the at least one memory stores instructions, which, when executed by the at least one processor, cause the apparatus to: Receive the machine learning model or information instructing the machine learning model from the other device.

7. The apparatus according to any one of claims 1 to 6, wherein the at least one memory stores instructions, which, when executed by the at least one processor, cause the apparatus to: Receive instructions from the other device to perform the radio beam prediction using the machine learning model or a different machine learning model based on the preference, capability, or lack thereof indicated by the device. The radio beam prediction is performed using the machine learning model, or the different machine learning models described herein. as well as The radio beam prediction is reported to the other devices.

8. The apparatus of claim 7, wherein the at least one memory stores instructions that, when executed by the at least one processor, cause the apparatus to: Receive instructions from the other devices to suspend or interrupt the use of the machine learning model, or a different machine learning model, to perform the radio beam prediction; and Pause or interrupt the use of the machine learning model, or a different machine learning model, to perform radio beam prediction.

9. The apparatus of claim 7, wherein the at least one memory stores instructions that, when executed by the at least one processor, cause the apparatus to: Receive instructions from the other device to suspend or interrupt reporting the radio beam prediction to the other device; and Suspend or interrupt the reporting of the radio beam prediction to the other devices.

10. The apparatus according to any one of claims 1 to 9, wherein the machine learning model has a complexity value based on at least one of the following: The number of parameters in the machine learning model; The number of layers in the machine learning model; The number of rounds in the machine learning model; The power consumption of the machine learning model; or The computational cost of the machine learning model.

11. The apparatus of claim 10, wherein the complexity value is an absolute complexity value or a relative complexity value.

12. The apparatus according to any one of claims 1 to 11, wherein the radio beam prediction comprises at least one of the following: Prediction of radio beams used for communication via channels; or Prediction of the quality of radio beams used for communication via channels.

13. The apparatus of any one of claims 1 to 12, wherein the instruction indicating whether the apparatus has determined to use the machine learning model to perform the radio beam prediction for cellular communication is transmitted via at least one of the following: Layer 1 signaling; Layer 2 signaling; or Layer 3 signaling.

14. The apparatus according to any one of claims 1 to 13, wherein the apparatus is a user equipment and the other apparatus is a base station; or The aforementioned device is a base station distributed unit, and the other devices are base station central units.

15. An apparatus comprising: At least one processor; as well as At least one memory storing instructions, which, when executed by the at least one processor, cause the means to: Sending instructions to another device to perform radio beam prediction for cellular communication using a machine learning model, wherein the machine learning model is associated with a complexity value; and Receive an instruction from the other device indicating whether the other device has determined, based on the complexity value, to use the machine learning model to perform radio beam prediction for cellular communication.

16. The apparatus of claim 15, wherein the at least one memory stores instructions, which, when executed by the at least one processor, cause the apparatus to: Based on the indication, determine whether the other devices use the machine learning model, or a different machine learning model, to perform the radio beam prediction; and Instructions are sent to the other devices to perform the radio beam prediction using the machine learning model, or the different machine learning model.

17. The apparatus of claim 15 or claim 16, wherein the at least one memory stores instructions that, when executed by the at least one processor, cause the apparatus to: Receive from the other device an indication of preference, capability, or inability to perform the radio beam prediction using the machine learning model.

18. The apparatus according to any one of claims 15 to 17, wherein the at least one memory stores instructions, which, when executed by the at least one processor, cause the apparatus to: Send the machine learning model, or an instruction to the machine learning model, to the other devices.

19. The apparatus according to any one of claims 15 to 18, wherein the at least one memory stores instructions that, when executed by the at least one processor, cause the apparatus to: Receive an instruction from the other device to suspend or interrupt the use of the machine learning model to perform the radio beam prediction; or Instructions to suspend or interrupt reports of radio beam prediction to the other device from the other device.

20. The apparatus according to any one of claims 15 to 19, wherein the apparatus is a base station and the other apparatus is a user equipment; or The device mentioned above is a base station central unit, and the other devices mentioned above are base station distributed units.

21. A method comprising: The device receives instructions from another device to perform radio beam prediction for cellular communication using a machine learning model, wherein the machine learning model is associated with a complexity value; The device determines, based on the complexity value, whether to use the machine learning model to perform the radio beam prediction for cellular communication; as well as The device sends an instruction to the other devices, indicating whether the device has determined to use the machine learning model to perform the radio beam prediction for cellular communication.

22. A method comprising: The device sends an instruction to another device to perform radio beam prediction for cellular communication using a machine learning model, wherein the machine learning model is associated with a complexity value; as well as Receive an instruction from the other device indicating whether the other device has determined, based on the complexity value, to use the machine learning model to perform radio beam prediction for cellular communication.

23. A computer program comprising computer-executable instructions that, when executed on one or more processors, perform the steps of the method according to claim 21 or claim 22.

Images