Method for functionality framework to ai / ML for beam management in wireless communication networks

Abstract

The invention provides a method and system for AI / ML-based beam management in wireless communication networks. It enhances beam prediction accuracy by separating beam sets for training (Set A) and inference (Set B), linked through an Associated ID to ensure contextual consistency. The approach enables intelligent beam selection across spatial and temporal domains while reducing signalling overhead and computational complexity. A Model Performance Monitoring (MPM) mechanism periodically evaluates model accuracy using metrics such as L1-RSRP and prediction reliability, allowing dynamic updates. The framework supports data collection, training, inference, and monitoring across both UE and network entities. By optimizing beam alignment and resource allocation, the invention improves communication reliability, reduces latency, and supports scalable AI / ML integration in 5G Advanced and 6G systems.

Description

[0001]Method for functionality framework to AI / ML for BM includes data collection for model training, AI / ML model inference and Model performance monitoring Field of the Invention The present invention relates to wireless communication technology, specifically to methods and systems for enhancing beam management and prediction high-frequency communication networks such as 6G / 5G-NR networks. It involves the use of artificial intelligence and machine learning (AI / ML) models for optimizing for intelligent beam prediction, configuration, and performance monitoring in beamforming systems operating in millimetre wave (mmWave) and other high-band frequency spectrums. Background of the Invention Efficient and adaptive beam management (BM) has emerged as a cornerstone requirement in the evolution of wireless communication systems, particularly with the deployment of 5G and the anticipated advancements in 6G technologies. Beam management is essential to ensure reliable connectivity, reduced latency, and enhanced throughput in networks that operate in high- frequency bands, including millimetre wave (mmWave) and beyond. These frequency bands, while enabling high bandwidth and spectral efficiency, also present significant challenges in terms of signal propagation and channel dynamics. In 5G networks, especially in mmWave deployments, signals are subjected to much higher path loss than those at lower frequency ranges. To compensate for this loss, both the base station (gNB) and user equipment (UE) are equipped with large antenna arrays that generate narrow directional beams, thereby achieving high beamforming gain. However, this high directionality introduces a fundamental challenge: maintaining proper alignment between the transmit (Tx) and receive (Rx) beams becomes increasingly difficult, particularly in mobile scenarios or under rapidly changing channel conditions. Even minor deviations in device position or orientation can degrade the beam alignment, leading to significant performance drops. Furthermore, as beamwidth narrows, a larger number of beams is required to cover a given spatial area. This makes beam sweeping, selection, and tracking computationally intensive and control-signalling heavy, thereby increasing the power consumption on the UE side and impacting user experience. Conventional beam management techniques typically rely on predefined codebooks, iterative beam refinement, and periodic feedback, which suffer from scalability issues and are insufficient in addressing the dynamic demands of real-time communication environments. To overcome these limitations, the application of artificial intelligence (AI) and machine learning (ML) has been explored to enhance beam prediction and optimization. AI / ML-based beam management approaches can learn complex mappings between signal observations and optimal beam directions across spatial and temporal domains. These data-driven methods offer the potential to reduce the overhead associated with legacy approaches by enabling proactive and adaptive beam selection. However, existing AI / ML solutions in the art are often incomplete, lacking practical integration with radio signalling, resource configuration, and lifecycle management of models. In particular, they fail to distinguish between training-phase beam data and inference-phase operational beams, do not track model evolution or context using identifiers, and lack real- time performance monitoring capabilities. Moreover, current AI / ML-based approaches do not adequately define mechanisms to ensure that model training assumptions remain consistent during inference. Without such consistency, predictions become unreliable, especially in the presence of variable or unanticipated network conditions. Furthermore, these solutions do not consider the efficient signalling and resource allocation needed to operate in compliance with evolving 3GPP standards, particularly in Release 19 and beyond, where AI / ML support for beam management is gaining emphasis. Accordingly, there exists a need for a robust and functional AI / ML framework for beam management in advanced wireless communication systems. The framework must include explicit mechanisms for separating and configuring training and inference beam sets, maintaining context and implementing model monitoring routines to ensure sustained accuracy under real-world conditions and should enable both UE-side and gNB-side model deployment, accommodate spatial and temporal prediction use cases and operate with minimal signalling overhead while ensuring consistency, traceability, and adaptability. Objective of the Invention The principal objective of the present invention is to enhance efficient beam management in wireless communication systems, particularly in 5G Advanced and 6G networks operating in high-frequency bands such as millimeter wave (mmWave), by leveraging artificial intelligence (AI) and machine learning (ML) techniques. Another objective of the present invention is to enhance AI / ML model inference by leveraging beam measurements to predict optimal beams, enabling accurate beam selection in both spatial and temporal domains Another objective of the present invention is to improve data collection for AI / ML model training by utilizing an Associated ID to ensure efficient dataset creation and model consistency without disclosing proprietary network information. Another objective of the present invention is to introduce and utilize an Associated ID to maintain contextual consistency between the training and inference stages of the AI / ML model, thereby ensuring reliable and reproducible model behaviour. Another objective of the present invention is to support dynamic model performance monitoring, either periodically or aperiodically, to detect degradation in prediction accuracy and enable timely model adaptation, selection, or replacement. A further objective of the present invention is to enable robust model performance monitoring through periodic or aperiodic evaluations at intervals, ensuring adaptability to dynamic channel conditions and maintaining prediction accuracy Summary of the Invention This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the detailed description. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter. The present invention relates to a method and system for enhancing beam management in 5G Advanced and 6G wireless communication networks through the application of artificial intelligence and machine learning (AIML) techniques. This invention addresses the technical challenges associated with high- frequency communications, particularly the need for accurate beam alignment in dynamic and high-pathloss environments such as those operating in millimeter wave (mmWave) bands. The method comprises a functional AI / ML framework consisting of data collection, model training, management, inference, and model storage. Beam management is implemented by distinguishing two sets of beams Set A for training and validation and Set B for real-time inference. An Associated ID is introduced to maintain contextual linkage between beam sets, ensuring consistency across model lifecycle stages and allowing traceable configuration and deployment. During the training phase, both Set A and Set B beams, along with the Associated ID, are transmitted by the gNB to the UE for measurement and dataset generation. AI / ML models are trained on these datasets to predict optimal beam directions based on environmental and signal features. The model may be hosted at either the UE or gNB, depending on system architecture. In the inference phase, only Set B beams are transmitted. Based on measurements of these beams, the AI / ML model predicts the best corresponding Set A beams, which are then reported to the gNB. The gNB may use this report to activate a suitable Transmission Configuration Indicator (TCI) state. This selective beam transmission and intelligent prediction reduce overhead, latency, and UE power consumption. Additionally, the invention incorporates a performance monitoring mechanism whereby beam prediction accuracy and signal quality metrics such as L1-RSRP are continuously or periodically evaluated. Monitoring intervals may be configured at 5 ms, 10 ms, or 20 ms, adapting to real-time channel variability. Based on these metrics, the system may retain, update, or switch AI / ML models to maintain optimal performance. This method enables scalable, context-aware, and low-overhead beam management in future wireless systems. It supports both spatial and temporal beam prediction, facilitates efficient resource usage, and provides robustness in fast-varying or complex propagation environments, thereby enhancing the reliability and efficiency of next-generation wireless communication networks. Brief description of the drawings The figures described below depict various aspects of the system and methods disclosed herein. It should be understood that each figure depicts an embodiment of a particular aspect of the disclosed system and methods, and that each of the figures is intended to accord with a possible embodiment thereof. Further, wherever possible, the following description refers to the reference numerals included in the following figures, in which features depicted in multiple figures are designated with consistent reference numerals. FIG. 1 illustrates an AI / ML-based Beam Management (BM) system designed to optimize beam selection and enhance communication efficiency in wireless networks (100). The system comprises two distinct models: the User Equipment (UE)-side model and the Network (NW)-side model FIG.1(A) depicts an AI / ML-based Beam Management (BM) system with a user equipment (UE)-side model (100(a)), in accordance with one embodiment of the present invention. FIG.1(B) presents an AI / ML-based Beam Management (BM) system with a network (NW)-side model (100(b)), in accordance with one embodiment of the present invention. FIG. 2 illustrates different mapping scenarios configurations between an Associated ID and one or more Model Identifiers (Model IDs), used within an AI / ML-based beam management system in a wireless communication network, in accordance with embodiments of the present invention. FIG.2(A) illustrates a mapping scenario where one Associated ID is linked to a single Model ID (200(a)), in accordance with one embodiment of the present invention. FIG.2(B) depicts a mapping scenario where one Associated ID is linked to multiple Model IDs (200(b)), in accordance with one embodiment of the present invention. FIG. 3 represents a timing diagram for a model performance monitoring system (300), designed to track and evaluate the AI / ML model's efficiency and adaptability, in accordance with one embodiment of the present invention. FIG. 4 illustrates two exemplary configurations of AI / ML model inference processes for beam management (BM) in a wireless communication system, in accordance with embodiments of the present invention. FIG.4(A) illustrates the AI / ML model inference process in Case 1 (400(a)), in accordance with one embodiment of the present invention. FIG.4(B) illustrates the AI / ML model inference process in Case 2 (400(b)), in accordance with one embodiment of the present invention. FIG. 5 depicts a flow diagram of an AI / ML-based beam management method in a wireless communication system (500), in accordance with one embodiment of the present invention. FIG.6 is a block diagram illustrating an example of a schematic hardware configuration of the network node (600) according to one embodiment of the present invention. Persons skilled in the art will appreciate that elements in the figures are illustrated for simplicity and clarity and may have not been drawn to scale. For example, the dimensions of some of the elements in the figure may be exaggerated relative to other elements to help to improve understanding of various exemplary embodiments of the present disclosure. Throughout the drawings, it should be noted that like reference numbers are used to depict the same or similar elements, features, and structures. Detailed Description of the Invention The following description with reference to the accompanying drawings is provided to assist in a comprehensive understanding of exemplary embodiments of the invention as defined by the claims and their equivalents. It includes various specific details to assist in that understanding but these are to be regarded as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the invention. In addition, descriptions of well-known functions and constructions are omitted for clarity and conciseness. The terms and words used in the following description and claims are not limited to the bibliographical meanings but are merely used by the inventor to enable a clear and consistent understanding of the invention. Accordingly, it should be apparent to those skilled in the art that the following description of exemplary embodiments of the present invention are provided for illustration purpose only and not for the purpose of limiting the invention as defined by the appended claims and their equivalents. It is to be understood that the singular forms “a,” “an,” and “the” include plural referents unless the context clearly dictates otherwise. Thus, for example, reference to “a component surface” includes reference to one or more of such surfaces. By the term “substantially” it is meant that the recited characteristic, parameter, or value need not be achieved exactly, but that deviations or variations, including for example, tolerances, measurement error, measurement accuracy limitations and other factors known to those of skill in the art, may occur in amounts that do not preclude the effect the characteristic is intended to provide. Figures discussed below, and the various embodiments used to describe the principles of the present disclosure in this patent document are by way of illustration only and should not be construed in any way that would limit the scope of the disclosure. Those skilled in the art will understand that the principles of the present disclosure may be implemented in any suitably arranged system. The terms used to describe various embodiments are exemplary. It should be understood that these are provided to merely aid the understanding of the description, and that their use and definitions, in no way limit the scope of the invention. Terms first, second, and the like are used to differentiate between objects having the same terminology and are in no way intended to represent a chronological order, unless where explicitly stated otherwise. A set is defined as a non-empty set including at least one element. FIG. 1 illustrates an AI / ML-based Beam Management (BM) system designed to optimize beam selection and enhance communication efficiency in wireless networks (100). The solution provides a method for a functional AI / ML framework for beam management, encompassing data collection for model training, AI / ML model inference, measurement reporting, and model performance monitoring, significantly reducing overhead compared to traditional non-AI / ML beam management approaches. The system comprises two distinct models: the User Equipment (UE)-side model and the Network (NW)-side model, tailored for 5G-Advanced and 6G technologies. The system enables a transmitting entity, such as a network node, to transmit a set of beams including a first set of beams (Set A) and a second set of beams (Set B), each associated with an identifier (Associated ID) conveying time and spatial characteristics, and a receiving entity, such as a user equipment, to measure the beams and transmit responses including measurement information and the Associated ID, wherein the measurement information is used to train an AI / ML model for predicting optimal beams from Set A, with Layer 1 Reference Signal Received Power (L1-RSRP) as a model output, and to monitor model performance by calculating metrics such as prediction accuracy and L1-RSRP differences. This invention offers key benefits, including reduced signaling overhead, improved beam selection efficiency, adaptability to dynamic channel conditions, lower power consumption due to optimized resource use, and faster beam prediction through AI / ML inference. Its applications include beam management in mobile networks (e.g., vehicular communications), high-frequency mmWave systems for high-throughput scenarios, and fixed wireless access (FWA) for static deployments. The system is distinguished by its use of an Associated ID for model lifecycle management, optimized configuration of Set A and Set B beams, periodic model performance monitoring, resource allocation optimization, and seamless integration of AI / ML with beam measurement processes. FIG.1(A) depicts an AI / ML-based Beam Management (BM) system with a user equipment (UE)-side model (100(a)), in accordance with one embodiment of the present invention. T The figure depicts the interaction between a network entity, referred to as the Network (NW), and one or more User Equipment (UE) devices in a wireless communication system, specifically tailored for 5G Advanced or 6G technologies. The procedure involves signalling exchanges between a transmitting entity, such as a network node (NW) (e.g., a next- generation Node B or gNB), and a receiving entity, such as a user equipment (UE). The method illustrates the process flow of training and inference using an AI / ML model located at the UE side. Initially, the network node (NW) allocates radio resources and transmits two sets of downlink beams Set A and Set B to the user equipment (UE) along with an Associated ID. The Associated ID labels and tracks data for model training, conveying gNB-side assumptions (e.g., temporal and spatial beam properties) to the UE without disclosing proprietary network information. The UE collects input data comprising measurements on Set A and Set B beams. The known model structure, including model type (e.g., Transformer, Convolutional Neural Networks), input / output specifications, number of layers, layer types / structures, layer sizes, and connections between layers, is communicated between the gNB and UE using a Model ID to ensure interoperability. These datasets are used for training, validating, and selecting AI / ML models, with the UE identifying the best-performing models based on performance metrics. During inference, the NW transmits only Set B beams to the UE, which are used as inputs to the trained AI / ML model. The model predicts the most suitable Set A beams based on current Set B measurements, producing outputs such as predicted Top-K or Top-1 beams using Layer 1 Reference Signal Received Power (L1-RSRP). The UE reports inference results to the NW, including beam information on predicted Top-K beams for BM Case 1 (spatial-domain prediction) or multiple time instances in one report for BM Case 2 (temporal-domain prediction). For BM Case 2, the ability to predict multiple time instances in a single report, as specified in the invention’s claims, enhances efficiency in temporal- domain beam forecasting. Based on the UE’s report, the NW performs Transmission Configuration Indicator (TCI) configuration and activation, either directly using predicted beams or after refined measurements. The NW transmits control signalling to indicate the TCI state for downlink reception. This embodiment optimizes resource utilization by transmitting only Set B beams during inference, with Set A beams reserved for training, and uses the Associated ID for robust lifecycle management (LCM). Model performance monitoring involves the NW allocating resources for the UE to measure Set A and Set B beams over configured resources, with the UE reporting measurements or calculating performance metrics, including prediction accuracy and L1-RSRP differences, for the NW to assess model effectiveness, wherein monitoring is conducted periodically at intervals of 5 milliseconds, 10 milliseconds, or 20 milliseconds, or aperiodically based on channel conditions. Monitoring can be periodic (e.g., every 5ms, 10ms, or 20ms) or aperiodic, depending on channel variability. Thus, FIG. 1(A) represents a UE-centric AI / ML framework for beam management, incorporating data collection, model training, inference, reporting, and configuration signalling. Although not depicted in this figure, model performance monitoring may also be performed in the network. In such cases, the NW allocates resources for UE beam measurements, and the UE reports metrics such as prediction accuracy or L1-RSRP differences, or the NW calculates these KPIs to monitor the model’s effectiveness over time. Monitoring can be executed periodically (e.g., every 5ms, 10ms, or 20ms) or aperiodically, depending on channel variability. Although not depicted in this figure, model performance monitoring may also be performed in the network. In such cases, the NW allocates resources for UE beam measurements, and the UE reports metrics such as prediction accuracy or L1-RSRP differences, or the NW calculates these metrics to monitor the model’s effectiveness over time, determining whether to select, activate, deactivate, or switch the AI / ML model based on performance degradation. Monitoring can be executed periodically (e.g., every 5ms, 10ms, or 20ms) or aperiodically, depending on channel variability. Thus, the sequence illustrated in FIG. 1(A) represents a UE-centric functional AI / ML framework for beam management in 5G-Advanced and beyond wireless systems, incorporating beam data collection, model training, inference, reporting, and configuration signalling between a transmitting network node (NW) and a receiving user equipment (UE). FIG.1(B) presents an AI / ML-based Beam Management (BM) system with a network (NW)-side model (100(b)), in accordance with one embodiment of the present invention. The figure illustrates the signaling exchange and functional operation between a transmitting entity, such as a network node (NW) for example, a gNB and a receiving entity, such as user equipment (UE), in a wireless communication system where the AI / ML model resides on the NW side. The operations comprise beam measurement, reporting, AI / ML model training and inference, and TCI configuration. In another embodiment, the network node (NW) configures and transmits two sets of beams Set A and Set B to the user equipment (UE). The UE receives the beams, performs measurements on both Set A and Set B, and reports the measurements back to the NW along with an Associated ID. The Associated ID serves to label the data and represents the underlying assumptions of the NW regarding beam transmission characteristics such as spatial and temporal features. The use of the Associated ID facilitates consistent model behaviour and enables lifecycle management (LCM) of the AI / ML model across training and inference phases. Following this, the NW uses the measurement reports and Associated ID received from the UE to train one or more AI / ML models and subsequently selects the best-performing model based on performance criteria. The AI / ML model structure is known to both NW and UE through a Model ID, and the mapping between the Model ID and the Associated ID is determined by the data sets used during training. After model training and selection, the NW transmits only Set B beams, along with the corresponding Associated ID, to the UE. The UE measures the Set B beams and reports the Top-K beams to the NW. The number of bits needed for Channel State Information Reference Signal Resource Indicator (CRI) and Synchronization Signal Block Resource Indicator (SSBRI) depends on the total number of reported beams, and the maximum potential number of beams reporting on Uplink Control Information (UCI) on Physical Uplink Control Channel (PUCCH) is limited (e.g., up to 121 beams for full reports or 243 beams without CRI, or 256 beams with differential RSRP). These beams are identified based on Layer-1 Reference Signal Received Power (L1-RSRP) measurements. The UE may report beam indices within an X dB gap to the maximum RSRP value, where the parameters K (the number of top beams) and X (threshold margin) are configured by the NW. The reported content may include beam identifiers alone or with corresponding RSRP values. The NW then performs AI / ML model inference by using the measurement reports and the Associated ID as input to the trained AI / ML model. The model infers the likely optimal Set A beam for the given input conditions. Depending on the beam management scenario, the inference may follow either of the following modes: • BM Case 1 (spatial-domain prediction): The AI / ML model uses current- time measurements of Set B to predict the best beam from Set A for immediate use. • BM Case 2 (temporal-domain prediction): The AI / ML model uses historical Set B measurements to predict the most appropriate beam from Set A for future time instances. Based on the inference results, the NW configures and activates Transmission Configuration Indicator (TCI) states. TCI configuration may be based directly on the predicted Top-K beams or may follow refined measurements by the NW before activation. The UE receives the TCI configuration via downlink control signalling and uses the indicated beams for downlink reception. The described approach reduces the UE's computational complexity by shifting AI / ML training and inference responsibilities to the NW, while maintaining measurement and reporting capabilities at the UE. The use of Set A and Set B beams ensures separation between training and inference stages, thereby optimizing air interface resource usage. Furthermore, the Associated ID provides consistency across model phases, enabling robust and traceable AI / ML operation. In UE-assisted performance monitoring, the UE measures Set A and Set B beams over allocated resources and reports measurements or calculates performance metrics, including prediction accuracy and L1-RSRP differences, to the NW, which decides to retain, switch, or deactivate the model based on performance degradation, with monitoring conducted periodically at intervals of 5 milliseconds, 10 milliseconds, or 20 milliseconds, or aperiodically depending on channel conditions. Accordingly, FIG. 1(B) illustrates a functional AI / ML framework for beam management in which training, inference, and decision-making operations are executed at the transmitting network node (NW), while measurement and reporting functions remain at the receiving user equipment (UE), enabling efficient and adaptive beamforming in 5G-Advanced and 6G networks. FIG. 2 illustrates different mapping scenarios configurations between an Associated ID and one or more Model Identifiers (Model IDs), used within an AI / ML-based beam management system in a wireless communication network, in accordance with embodiments of the present invention. The Associated ID functions as a unique reference identifier that encapsulates environmental context, spatial and temporal characteristics, and resource configuration parameters relevant to the beam measurements collected from user equipment (UE) or network entities. These identifiers ensure consistency between training and inference phases and facilitate scalable, traceable model lifecycle operations. FIG.2(A) illustrates a mapping scenario where one Associated ID is linked to a single Model ID (200(a)), in accordance with one embodiment of the present invention. In this embodiment, a one-to-one correspondence is established between an Associated ID and a Model ID. The Associated ID, as signalled by a transmitting entity such as a network node (NW), for example, a gNB, is transmitted alongside beam configurations comprising Set A and Set B beams to a receiving entity, such as user equipment (UE). The Associated ID encapsulates assumptions of the NW related to spatial and temporal configurations of the beams, while explicitly avoiding the exposure of any proprietary internal details. Upon receiving the Associated ID and beam configurations, the UE performs beam measurements and uses the collected measurement data to build a data set. The mapping between Model ID and Associated ID depends on the datasets used for models training. This data set is used to train an AI / ML model locally at the UE. The data set corresponding to one Associated ID is used to train a single model, resulting in a unique Model ID. Thus, the mapping between Associated ID and Model ID is strictly one-to-one. The Model ID serves to reference the trained model and may be used for subsequent inference steps, signalling, model update, or performance tracking. The Model ID may represent not only the identity of the trained model but also the model architecture, including model type (e.g., convolutional neural network, transformer), input / output definitions, number of layers, type of each layer (e.g., fully connected, activation), parameter sizes, and connectivity between layers. This known structure of the model ensures interoperability and consistency across different modules of the AI / ML framework. The Associated ID plays a key role in lifecycle management (LCM) by allowing the trained model to be reused consistently during inference under similar radio conditions as experienced during training. This one-to-one mapping simplifies model selection logic and is applicable to cases where the training conditions and operational context are static or well-defined. FIG.2(B) depicts a mapping scenario where one Associated ID is linked to multiple Model IDs (200(b)), in accordance with one embodiment of the present invention. In another embodiment, one-to-many mapping between an Associated ID and multiple Model IDs, where a single data collection context is used to generate more than one trained model. The Associated ID, transmitted by a network node (NW) to a user equipment (UE), represents the NW's assumptions about the configuration of beam sets (Set A and Set B) and their expected spatial or temporal distribution. The UE performs measurements based on the received beams and collects corresponding data sets that are labelled with the Associated ID. The mapping between Model ID and Associated ID depends on the datasets used for models training. However, unlike the one-to-one case in FIG.2(A), the UE uses the same data set to train multiple AI / ML models, each having distinct objectives, characteristics, or optimizations. Each model trained using this common data set is assigned a unique Model ID (e.g., Model ID 1, Model ID 2, ..., Model ID N). These models may differ based on the model architecture, number of layers, optimization goals (e.g., accuracy, latency, prediction time), or intended prediction targets (e.g., different future time instances or different beam domains). This mapping approach enables advanced functionality, such as multi-model inference, ensemble learning, or scenario-specific model switching, while still maintaining traceability through a common Associated ID. The one-to-many mapping configuration illustrated in FIG. 2(B) is particularly advantageous in scenarios where the same beam measurement context (defined by the Associated ID) can serve multiple operational needs or where models are trained to handle diverse aspects of beam prediction (e.g., spatial vs. temporal domains). During inference, the NW or UE may reference the Associated ID to select an appropriate model from the list of Model IDs linked to it, enabling runtime adaptability. This mapping also supports versioning and incremental training where newer models can be introduced without discarding older models linked to the same context. Together, FIG. 2(A) and FIG. 2(B) demonstrate flexible and scalable model-to-data mapping schemes that underpin the invention’s ability to manage AI / ML model lifecycles for beam management with traceability, modularity, and performance assurance in 5G Advanced and 6G systems. FIG. 3 represents a timing diagram for a model performance monitoring system (300), designed to track and evaluate the AI / ML model's efficiency and adaptability, in accordance with one embodiment of the present invention. The operation of periodic model performance monitoring in the context of an AI / ML- enabled beam management (BM) framework for advanced wireless communication systems, such as 5G-Advanced and 6G. The purpose of periodic monitoring is to ensure that the AI / ML model, once deployed, continues to provide reliable and accurate beam predictions under dynamic channel conditions, mobility scenarios, or changing configuration contexts. In one embodiment, the transmitting entity, such as a network node (NW) or gNB, initiates the monitoring process at fixed time intervals. The receiving entity, such as the user equipment (UE), participates by performing beam measurements or returning reports, depending on the monitoring strategy. The periodic intervals may be configured at 5 milliseconds (ms), 10 ms, or 20 ms, based on system requirements, particularly to account for the rapid changes in wireless propagation environments. Performance monitoring should be conducted in either a periodic or aperiodic manner, with aperiodic monitoring preferred in fixed wireless access (FWA) scenarios and periodic monitoring necessary in certain situations with intervals of 5ms, 10ms, or 20ms, as channel conditions can change rapidly within these time frames. At the beginning of each interval, the network allocates specific resources for transmitting a predefined set of beams, which may include Set A and Set B beams. These beams are configured for the purpose of performance testing and are not necessarily used for data communication during this phase. Upon receiving these beam transmissions, the UE performs signal strength measurements such as Layer-1 Reference Signal Received Power (L1-RSRP) over the indicated beam resources. In UE-side performance monitoring, the UE calculates the KPI and reports it back to the gNB to make a decision, and in gNB- side performance monitoring, the gNB calculates the KPI (e.g., L1-RSRP or differential L1-RSRP or prediction accuracy) and makes a decision. Another embodiment, where the AI / ML model resides on the network side, the UE transmits the raw measurement results back to the network, and the NW computes various key performance indicators (KPIs), including but not limited to absolute L1-RSRP values, differential RSRP between measured and predicted beams, and beam prediction accuracy. In an alternative embodiment, where the model resides on the UE side, the UE may locally compute such KPIs and send the processed performance indicators to the NW. In both cases, the periodic reports form the basis for determining whether the deployed model remains valid or needs to be updated. The network uses the information obtained from each periodic monitoring cycle to make decisions regarding the AI / ML model. Based on observed performance degradation, the network may decide to retain, retrain, deactivate, or switch to an alternate model. Additionally, if performance consistently falls below a configured threshold, the system may invoke a fallback procedure, reverting to non-AI-based beam management techniques. This decision logic ensures that system performance does not deteriorate due to outdated or poorly adapted models. The use of periodic monitoring is particularly important in high-mobility scenarios, such as vehicular environments, where the radio channel evolves rapidly. In contrast, aperiodic monitoring may be better suited to static deployment scenarios, such as Fixed Wireless Access (FWA), where channel conditions remain relatively stable over time. However, even in FWA scenarios, periodic monitoring may be selectively activated in response to detected environmental changes, ensuring robustness and continuity of service. Accordingly, FIG. 3 demonstrates the temporal arrangement and operational logic of periodic AI / ML model monitoring, providing a systematic mechanism for tracking model validity, accuracy, and relevance across time in real-world deployments. This ensures the AI / ML-based beam management system remains responsive, efficient, and adaptable under varying network and user conditions. FIG. 4 illustrates two exemplary configurations of AI / ML model inference processes for beam management (BM) in a wireless communication system, in accordance with embodiments of the present invention. The AI / ML inference procedures are performed using pre-trained models that consume beam measurement data as input and produce predictions of optimal beam selections across spatial or temporal domains. The system supports flexible inference architectures based on the placement of the AI / ML model, either at the receiving entity (e.g., user equipment (UE)) or the transmitting entity (e.g., gNB), as described respectively in BM Case 1 and BM Case 2. FIG.4(A) illustrates the AI / ML model inference process in Case 1 (400(a)), in accordance with one embodiment of the present invention.In this embodiment, referred to as Beam Management (BM) Case 1, the AI / ML model inference procedure is performed at the receiving entity, such as user equipment (UE), using real-time measurements collected from a current transmission instance. The transmitting entity (e.g., a gNB) configures and transmits a predefined second set of beams, referred to as Set B, using reference signal configurations such as CSI-RS. Upon receiving the beams, the UE performs beam measurements on Set B and generates signal quality metrics, including Layer-1 Reference Signal Received Power (L1-RSRP) and other relevant physical-layer indicators. These measurements are combined with an Associated ID, which encapsulates spatial, temporal, and beam configuration parameters reflecting the environment under which the model was originally trained. The Associated ID enables consistent alignment between the training dataset and the real-time inference context. The measurement data and Associated ID are then provided as input to a pre-trained AI / ML model hosted at the UE. The AI / ML model processes the Set B measurement data and infers the most suitable downlink beams from Set A, a set of candidate beams not transmitted during inference. The inference output may include a Top-K list of beams from Set A, along with their estimated signal strengths or prediction confidence values. This enables rapid beam selection and directional transmission optimization under dynamically changing channel conditions. The UE subsequently transmits the inference result to the gNB in the form of a second response, which includes the predicted beam identifiers and associated signal strength metrics. The gNB uses this information to configure and activate a Transmission Configuration Indicator (TCI) state, enabling targeted downlink beamforming aligned with the AI / ML model’s prediction. To support this inference and reporting framework, various fields are encoded with specified bitwidths to ensure efficient signalling and compliance with uplink payload constraints. These field definitions and their respective bitwidths are presented in Table 1 below. Field Bitwidth CRI [ log2 (KSCRI-RS)] SSBRI [ log2 (KSSSB)] RSRP 7 DIFFERENTIAL RSRP 4 Capability Index 2 Table 1: Bitwidth Allocation for Reporting Fields in AI / ML-Based Beam Management In Table 1, the CRI and SSBRI fields represent indices for CSI-RS and SSB beams, respectively, and their bitwidths are determined by the number of active signal sources. The RSRP field encodes absolute signal strength, while Differential RSRP allows compact encoding of signal deltas, reducing reporting overhead. The Capability Index encodes the receiver’s AI / ML reporting capability class, enabling the network to interpret and adapt beam prediction accordingly. This bitwidth structure is designed to optimize UCI reporting under payload limits, especially when the number of reported beams (Top-K) increases. The use of compressed or differential reporting enables support for up to 121 full beam reports or 62 differential reports within a typical PUCCH format, depending on configuration. Accordingly, FIG. 4(A), illustrates an end-to-end inference-driven beam prediction and reporting process under BM Case 1. The approach supports low- latency beam selection, minimizes uplink overhead, and provides high adaptability to time-varying radio conditions in 5G Advanced and future 6G systems. FIG.4(B) illustrates the AI / ML model inference process in Case 2 (400(b)), in accordance with one embodiment of the present invention. In this embodiment, referred to as Beam Management (BM) Case 2, the AI / ML model inference is performed using historical measurements of Set B beams rather than current- time observations. This approach is particularly suited for temporal-domain beam prediction, where the objective is to forecast optimal beam configurations for future time instances based on prior performance trends. In the illustrated scenario, the network node (NW) transmits Set B beams to the user equipment (UE) at multiple earlier time instances. The UE stores the measurements corresponding to these transmissions, building a time-series dataset of past Set B beam observations. These historical measurements, along with the Associated ID reflecting the context of the original beam configuration, are used as input to the AI / ML model for prediction. The AI / ML model processes this time-series input to infer which beams from Set A will most likely offer the best performance at a future target time instance. The model output includes predictions for Top-1 / N beams from Set A based on historical data and is used to infer the optimal beams for downlink transmission. The model output may include predicted Top-K beams from Set A, as well as expected signal quality metrics (e.g., L1-RSRP estimates) for those beams. The UE reports the predicted beam information back to the NW, allowing the network to anticipate beam quality trends and perform proactive TCI configuration and activation. BM Case 2 provides a predictive mechanism for anticipatory beam switching or pre-scheduling, particularly useful in high-mobility or latency- sensitive applications, such as vehicular communications or handover scenarios. By leveraging prior beam quality observations and applying time-series modelling, the AI / ML system can improve reliability and reduce the signalling overhead associated with reactive beam management.\ In both Case 1 and Case 2, the use of the Associated ID ensures lifecycle continuity between model training and inference, preserving the contextual integrity of the AI / ML inference process. These two cases, as represented in FIG. 4(A) and FIG. 4(B), together enable a versatile AI / ML framework that supports both real-time and predictive beam management in advanced wireless networks. FIG. 5 depicts a flow diagram of an AI / ML-based beam management method in a wireless communication system (500), in accordance with one embodiment of the present invention. The method enables efficient, intelligent beam selection and transmission configuration using measurement-driven inference models. The procedure supports both model training and inference stages and facilitates continuous performance monitoring and model lifecycle control. The method begins at step 505, where a transmitting entity, such as a network node (e.g., gNB), initiates the procedure by transmitting a set of beams that are divided into two categories namely Set A and Set B. Each set of beams is transmitted with an associated identifier, herein referred to as the Associated ID. The Associated ID indicated to the UE will not disclose any proprietary information about the gNB to the UE. This Associated ID conveys contextual information including temporal and spatial characteristics of the beam configuration and serves as a key reference to ensure lifecycle consistency during AI / ML model training and inference without disclosing proprietary network- side parameters to the receiver. At step 510, the transmitting entity configures reference signal resources, such as Channel State Information Reference Signals (CSI-RS), particularly for Set B beams. These resources are allocated in a manner compliant with 3GPP standards to ensure that the receiving entity, such as the user equipment (UE), can perform standardized and accurate signal measurements. These reference signals enable precise channel quality evaluation, which forms the basis of the subsequent AI / ML model processing. At step 515, the receiving entity receives the Set B beams and conducts signal measurements using the allocated reference signal resources. These measurements may include but are not limited to Layer 1 Reference Signal Received Power (L1-RSRP), signal-to-noise ratios, and other channel quality indicators. The resulting measurements serve as the foundation for both model training (in earlier stages) and inference (during operational deployment). At step 520, the receiving entity transmits a first response to the transmitting entity, containing the measured signal metrics for the Set B beams along with the originally received Associated ID. This feedback establishes a clear contextual link between the beam quality data and the training environment, allowing the network to map measurements to the correct AI / ML model configuration. At step 525, the transmitting entity collects the reported measurements and maps the Associated ID to one or more Model IDs. This mapping facilitates the training of one or more AI / ML models, with each Model ID corresponding to a distinct model architecture or training output. The mapping strategy supports both one-to-one and one-to-many model configurations depending on the use case, and ensures that all models retain a contextual linkage to the original beam configuration conditions signified by the Associated ID. At step 530, the system performs AI / ML model training using the beam measurement data from Set A and Set B. In this embodiment, training occurs at the UE for UE-sided models or at the NW using UE-reported measurements for NW sided models, leveraging datasets labeled with the Associated ID to ensure consistency. At step 535, during the model inference phase, only Set B beams are transmitted by the transmitting entity. The beam prediction can happen in the spatial domain (BM Case 1) or time domain (BM Case 2), with the scope of spatial-domain prediction being to predict the best DL TX beam or DL TX / RX beam pairs in different spatial locations, and time domain prediction aiming to predict the best DL TX beam or DL TX / RX beam pairs for future time instances. The receiving entity performs signal measurements over the Set B beams, and these measured values are then used as input to the previously trained AI / ML model. The model processes the input to predict the best-performing beams from Set A, based on previously learned relationships between Set B measurements and Set A performance. At step 540, the receiving entity generates and transmits a second response containing the Top-K predicted beams, based on the model’s inference output. This response may include beam indices and their associated L1-RSRP values, enabling the network to make informed decisions about directional transmission configurations. At step 545, based on the predicted beam information received from the UE, the transmitting entity configures a Transmission Configuration Indicator (TCI) state. The TCI state guides the UE on which beam or beam group to use for receiving downlink data transmissions, thereby optimizing beam alignment and improving communication efficiency. At step 550, the system performs performance monitoring of the deployed AI / ML model. This monitoring may occur in a periodic or aperiodic manner and includes metrics such as RSRP difference between predicted and actual beams and overall prediction accuracy. These metrics are used to evaluate whether the AI / ML model remains valid in the face of changing channel conditions or mobility dynamics. At step 555, the system compares the monitored metrics against predefined thresholds. If performance degradation is detected or channel conditions change significantly, a decision is made on whether to switch to an alternate model, retrain the current model, or deactivate the AI / ML inference system temporarily or permanently. This adaptive mechanism ensures robust beam management and maintains service quality across dynamic operating scenarios. At step 560, to further optimize deployment, especially at the receiving end, the system ensures efficient resource use by transmitting only Set B beams during inference, reducing signalling overhead compared to traditional beam management. Accordingly, FIG. 5 outlines a comprehensive end-to-end AI / ML-driven beam management method for wireless communication systems. The method includes model lifecycle processes ranging from data collection and training to inference, adaptation, and performance assurance, all contextualized by the Associated ID to maintain consistency and traceability across operational phases. FIG.6 is a block diagram illustrating an example of a schematic hardware configuration of the network node (600) according to one embodiment of the present invention. The system represents any transmitting entity, such as a gNB, or receiving entity, such as a UE, capable of performing signal measurement processing, beam prediction, and model management functions to enable intelligent directional communication in 5G Advanced or 6G systems. In one embodiment, the system (600) comprises a network interface (610), a processor (620), a memory (630), and a storage unit (640), operatively connected to execute AI / ML-based beam prediction methods, support training and inference operations, and manage beam configuration parameters for transmission optimization. The network interface (610) is configured to transmit, by a transmitting entity, a set of beams including a first set of beams (Set A) and a second set of beams (Set B), each associated with an identifier (Associated ID) conveying time and spatial characteristics, and to receive, by a receiving entity, the set of beams and a response including measurement information for the second set of beams and the Associated ID, wherein the measurement information includes at least one of L1-RSRP, channel state information (CSI), or beam identifier data. The interface comprises transceiver hardware, analog / RF chains, and digital baseband units capable of CSI-RS configuration, beam switching, and reference signal management to enable beam-level signal collection and control signalling. The processor (620) is configured to configure resources for the first and second sets of beams using a reference signal configuration; measure, by the receiving entity, the first and second sets of beams over the configured resources; train an AI / ML model using the measurement information from the first and second sets of beams to predict one or more optimal beams from the first set of beams; perform AI / ML model inference using measurements of the second set of beams to predict a Top-K or Top-1 beam in a spatial or temporal domain, wherein K is configurable and L1-RSRP is the model output; configure a transmission configuration indicator (TCI) state based on the predicted beams; monitor performance of the AI / ML model by calculating performance metrics, including prediction accuracy and L1-RSRP differences, at periodic intervals of 5 milliseconds, 10 milliseconds, or 20 milliseconds, or aperiodically, wherein the performance metrics include a comparison of predicted Top-K beam identifiers with a ground truth Top-1 beam identifier; transmit a performance monitoring response with differential L1-RSRP values and prediction accuracy metrics to evaluate model performance; and determine, based on the performance metrics, whether to select, activate, deactivate, or switch the AI / ML model when prediction accuracy or channel conditions degrade. The processor is further configured to collect a dataset including measurement information of the first and second sets of beams and the Associated ID and map the Associated ID to one or more model identifiers to support one-to-one or one-to-many mappings for training one or multiple AI / ML models without disclosing proprietary model parameters. Additionally, the processor supports efficient inference by using only Set B beam measurements, reducing computational and signaling overhead. The memory (630) is configured to store real-time beam measurement data, model prediction outputs, signal quality indicators such as L1-RSRP, channel state information (CSI), beam identifier data, AI / ML model parameters, and performance metrics used during inference and monitoring. The memory includes volatile storage (e.g., DRAM, SRAM) to buffer datasets, intermediate inference results, and model accuracy metrics that support decision logic for beam switching or retraining. The storage unit (640) is configured to retain AI / ML training datasets tagged with Associated IDs, model identifier mappings, and executable instructions for AI / ML training, inference, and performance evaluation. The storage includes model libraries and lookup tables that define valid beam sets for inference and reporting under PUCCH constraints, limiting the response to a maximum of 121 beams with beam identifiers or 256 beams with differential L1- RSRP values, based on uplink control information payload capacity as defined in 3GPP standards. The combination of these components enables the system (600) to perform intelligent and context-aware beam management using AI / ML models, reduce control signalling overhead, and maintain beam alignment accuracy under rapidly changing radio conditions. The system architecture supports deployment flexibility with model hosting on either the UE or network side and provides built-in support for training multiple AI / ML models using common datasets via Associated ID mapping. This includes data-driven beam prediction, inference-driven beam configuration, and performance-aware model adaptation in next-generation wireless networks. A number of implementations have been described. Nevertheless, it will be understood that various modifications may be made without departing from the spirit and scope of the disclosure. Accordingly, other implementations are within the scope of the following claims.

Claims

We Claim:

1. A method for beam management in a wireless communication system using artificial intelligence or machine learning (AI / ML) for beam prediction, comprising: transmitting, by a transmitting entity, a set of beams including a first set of beams (Set A) and a second set of beams (Set B), each associated with an identifier (Associated ID) conveying time and spatial characteristics; configuring resources for the second set of beams using a reference signal configuration; receiving, by a receiving entity, the set of beams and measuring the second set of beams over the configured resources; transmitting, by the receiving entity, a first response including measurement information for the second set of beams and the Associated ID; training an AI / ML model using the measurement information from Set A and Set B beams to predict one or more optimal beams from the first set of beams; performing AI / ML model inference using measurements of the second set of beams to predict the one or more optimal beams in a spatial or temporal domain, wherein Layer 1 Reference Signal Received Power (L1-RSRP) is considered as a model output; configuring a transmission configuration indicator (TCI) state based on the predicted one or more optimal beams; and monitoring performance of the AI / ML model by calculating performance metrics, including prediction accuracy and L1-RSRP differences.

2. The method as claimed in claim 1, further comprising: collecting a dataset including measurement information of the first set of beams (Set A) and the second set of beams (Set B) with the Associated ID; and mapping the Associated ID to one or more model identifiers to support training one or multiple AI / ML models.

3. The method as claimed in claim 2, wherein the Associated ID labels datasets for training the AI / ML model without disclosing proprietary model parameters and enables training multiple AI / ML models using a single dataset.

4. The method as claimed in claim 1, wherein performing AI / ML model inference includes: transmitting, by the receiving entity, a second response including beam information for a Top K predicted beam, wherein K is configurable and L1-RSRP is the model output; and using only the second set of beams (Set B) for inference to reduce resource overhead.

5. The method as claimed in claim 1, wherein training the AI / ML model includes collecting measurements from the first set of beams (Set A) and the second set of beams (Set B) with the Associated ID to train one or more models for spatial or temporal beam prediction.

6. A method for beam management in a wireless communication system using artificial intelligence or machine learning (AI / ML) for performance monitoring, comprising: transmitting, by a transmitting entity, a set of beams including a first set of beams (Set A) and a second set of beams (Set B), each associated with an identifier (Associated ID); configuring resources for the first and second sets of beams using a reference signal configuration; receiving, by a receiving entity, the set of beams and measuring the first and second sets of beams over the configured resources; transmitting, by the receiving entity, a response including measurement information for the second set of beams; monitoring performance of an AI / ML model used for beam prediction by calculating performance metrics, including prediction accuracy and L1-RSRP differences at the transmitting entity or receiving entity; anddetermining, based on the performance metrics, whether to select, activate, deactivate, or switch the AI / ML model when prediction accuracy or channel conditions degrade.

7. The method of claim 6, further comprising: calculating performance metrics, including L1-RSRP differences and prediction accuracy, based on measurements of the first and second sets of beams; and transmitting a performance monitoring response including the performance metrics, calculated by either the receiving entity or the transmitting entity.

8. The method as claimed in claim 6, wherein monitoring performance is conducted periodically at intervals of 5 milliseconds, 10 milliseconds, or 20 milliseconds, or aperiodically based on channel conditions 9. The method as claimed in claim 6, wherein monitoring performance is performed at intervals to adapt to dynamic channel conditions in fixed wireless access or mobile scenarios.

10. The method of claim 6, wherein determining whether to select, activate, deactivate, or switch the AI / ML model is based on performance metrics indicating degradation in prediction accuracy or channel conditions.

11. The method of claim 7, wherein the performance monitoring response includes differential L1-RSRP values compared to reference values measured over the first set of beams.

12. The method as claimed in claim 11, wherein the performance metrics include a comparison of predicted Top-K beam identifiers with a ground truth Top- 1 beam identifier.

13. A network node for beam management in a wireless communication system using artificial intelligence or machine learning (AI / ML), comprising:a network interface configured to: transmit, by a transmitting entity, a set of beams including a first set of beams (Set A) and a second set of beams (Set B), each associated with an identifier (Associated ID) conveying time and spatial characteristics; receive, by a receiving entity, the set of beams and a response including measurement information for the second set of beams and the Associated ID; a processor configured to: configure resources for the first and second sets of beams using a reference signal configuration; measure, by the receiving entity, the first and second sets of beams over the configured resources; train an AI / ML model using the measurement information from the first and second sets of beams to predict one or more optimal beams from the first set of beams; perform AI / ML model inference using measurements of the second set of beams to predict a Top-K or Top-1 beam in a spatial or temporal domain, wherein K is configurable and L1-RSRP is the model output; configure a transmission configuration indicator (TCI) state based on the predicted beams; monitor performance of the AI / ML model by calculating performance metrics, including prediction accuracy and L1-RSRP differences, at periodic intervals of 5 milliseconds, 10 milliseconds, or 20 milliseconds, or aperiodically; transmit a performance monitoring response with differential L1- RSRP values and prediction accuracy metrics to evaluate model performance; and determine, based on the performance metrics, whether to select, activate, deactivate, or switch the AI / ML model when prediction accuracy or channel conditions degrade.a memory operably connected to the processor and configured to store measurement information, AI / ML model parameters, and performance metrics; and a storage unit configured to retain AI / ML training datasets and Associated ID mappings.

14. The network node as claimed in claim 13, wherein the processor is further configured to: collect a dataset including measurement information of the second set of beams and the Associated ID; train the AI / ML model using the dataset, mapping the Associated ID to one or more model identifiers to support one-to-one or one-to-many mappings for training one or multiple AI / ML models without disclosing proprietary model parameters..

15. The network node as claimed in claim 13, wherein the measurement information includes at least one of L1-RSRP, channel state information (CSI), or beam identifier data.

16. The network node as claimed in claim 13, wherein the performance metrics include a comparison of predicted Top-K beam identifiers with a ground truth Top-1 beam identifier.

17. The network node as claimed in claim 13, wherein the processor is configured to limit the response to a maximum of 121 beams with beam identifiers or 256 beams with differential L1-RSRP values, based on uplink control information payload capacity as defined in 3GPP standards.

Images