Ai-ML based positioning enhancements

Abstract

The invention introduces a novel method and system for enhancing positioning accuracy in wireless communication networks by utilizing advanced AI-ML (Artificial Intelligence-Machine Learning) techniques. The proposed system monitors and optimizes the performance of AI-ML models used for direct and assisted positioning based on measurements from reference nodes. It employs a framework for periodic Model Performance Monitoring (MPM), adjusting the AI-ML models based on real-time performance metrics such as position error, range error, and angular error. The method facilitates effective data collection from various transmission reception points (TRPs), ensuring consistent model training and inference. By integrating contextual information and measurements, the invention significantly improves positioning accuracy while reducing overhead in reporting and data management. This approach not only enhances the reliability of positioning in diverse environments but also ensures compatibility with existing infrastructure, thereby advancing the overall performance and efficiency of modern wireless communication systems.

Description

[0001]AI-ML based Positioning Enhancements Field of the Invention The present invention relates to wireless communication technology and, more particularly, to AI / ML-based methods for enhancing the positioning accuracy of User Equipment (UE) in Non-Line-of-Sight (NLoS) environments, especially within 5G and next-generation networks. Background of the Invention Accurate and reliable positioning of user equipment (UE) in wireless communication networks has become increasingly important with the advent of 5G and the anticipated rollout of 6G systems. A wide range of emerging applications including autonomous vehicles, industrial automation, augmented reality (AR), emergency response, and location- based services depend on high-precision location information to function effectively. These applications often require sub-meter or even centimeter- level accuracy, including in challenging environments such as dense urban areas, complex indoor spaces, or environments with a high density of obstacles. Conventional wireless positioning techniques, such as Time of Arrival (ToA), Angle of Arrival (AoA), Time Difference of Arrival (TDoA), and Enhanced Cell ID (E-CID), are commonly used in cellular systems. These methods typically rely on geometric relationships between signal measurements and known base station locations to estimate the UE position. While effective under ideal conditions particularly when a clear Line-of-Sight (LoS) path exists between the UE and one or more base stations the accuracy of these methods deteriorates significantly in Non- Line-of-Sight (NLoS) scenarios. In such cases, radio signals are subject to reflection, diffraction, and scattering from obstacles such as walls, buildings, or moving objects, introducing biases and errors in range and angle measurements. The problem is further exacerbated at high-frequency bands such as millimeter wave (mmWave), which are more susceptible to blockage and multipath effects. To improve positioning accuracy under such conditions, recent research and industry efforts have explored the use of artificial intelligence (AI) and machine learning (ML) techniques. These approaches include direct data-driven models that infer UE position based on input features such as signal strength, channel characteristics, and device context, as well as AI-assisted positioning, where machine learning models are used to refine or augment traditional positioning methods. Such methods offer the potential to improve accuracy, particularly in dynamic or NLoS environments, by learning complex relationships from empirical data. The Third Generation Partnership Project (3GPP), in its Release 17, has acknowledged the relevance of AI in positioning and has introduced mechanisms for indicating the presence of LoS or NLoS conditions. However, current standards do not define how such LoS / NLoS classification should be performed, leaving its implementation to vendors and operators. Existing solutions for LoS / NLoS detection typically rely on heuristic rules, threshold-based methods, or static models that lack adaptability and may perform poorly across diverse environments and use cases. Furthermore, these approaches often fail to generalize to complex or changing radio conditions, resulting in unreliable classification that ultimately limits positioning accuracy. Accordingly, there exists a need for more robust, consistent, and scalable method for identifying LoS and NLoS conditions and improving UE positioning in wireless networks. In particular, there is a need to address the limitations of traditional signal-based methods and heuristic-based AI models by providing more accurate, context-aware positioning support under a wide range of operating conditions. Objective of the Invention The principal objective of the present invention is to enhance the accuracy of User Equipment (UE) positioning in Non-Line-of-Sight (NLoS) environments using Artificial Intelligence (AI) and Machine Learning (ML) models. Another objective of the present invention is to optimize the processing of Channel Impulse Response (CIR) data by reducing its volume through up- sampling, down-sampling, or truncation while maintaining the precision required for accurate positioning. Another objective of the present invention is to improve the training of AI-ML models by utilizing site-specific datasets derived from both real- world measurements and simulated data to better recognize environmental factors and NLoS characteristics affecting positioning. Another objective of the present invention is to enhance the inference capabilities of the trained AI-ML models, allowing them to analyze processed Channel State Information (CSI) data and predict accurate positioning results, even in complex NLoS conditions. A further objective of the present invention is to enable continuous performance monitoring of the AI-ML models by comparing predicted positioning outputs against reference location data and dynamically adjusting model parameters based on defined performance metrics. 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. An aspect of the present invention introduces relates to a method for improving the positioning accuracy of User Equipment (UE) in Non-Line-of- Sight (NLoS) environments using Artificial Intelligence (AI) and Machine Learning (ML) models within 6G / 5G / NR communication networks. This invention addresses the technical challenges associated with NLoS conditions, which can severely impact the accuracy and reliability of positioning systems. The method comprises the processing of Channel State Information (CSI), specifically Channel Impulse Response (CIR), by employing techniques such as up-sampling, down-sampling, or truncation to reduce the number of significant CIR samples. This optimization facilitates efficient data transmission while maintaining the accuracy required for effective positioning. Furthermore, the invention introduces a dataset identification (ID) that uniquely characterizes site-specific attributes, environmental conditions, and network configurations. This dataset ID ensures consistency during both training and inference phases of the AI-ML models. The AI-ML models are trained using site-specific datasets obtained from a combination of real-world measurements and simulated data, enabling the models to effectively recognize and adapt to various environmental factors and propagation conditions that influence positioning accuracy. During the inference stage, the trained AI-ML models analyze the processed CSI data to predict precise positioning outcomes, even under adverse NLoS conditions. The invention also incorporates a continuous performance monitoring mechanism, wherein the predicted positioning outputs are compared against reference location data, allowing for dynamic adjustment of model parameters based on established performance metrics. This method ensures seamless integration with existing communication infrastructures and significantly enhances the overall performance and efficiency of positioning systems. The invention is designed to support advanced applications in future communication networks, including 6G, thereby providing a robust and effective strategy for addressing the complexities of UE positioning in diverse and challenging environments. 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 various positioning methodologies within a communication network (100). positioning methodologies includes: FIG. 1(A) depicts the location server-based UL positioning, according to one embodiment of the present invention. FIG. 1(B) represents the location server-based DL positioning, according to one embodiment of the present invention. FIG. 1(C) depicts the User equipment-based positioning, according to one embodiment of the present invention. FIG.2 illustrates two methods for computing the measurements for reporting CIR samples (200), according to one embodiment of the present invention. FIG.3 illustrates the monitoring of model performance for different AI-ML-based positioning methods that utilize measurements from reference nodes (300), comprising: FIG.3(A) outlines the model performance monitoring for direct AI-ML based positioning using measurements from reference nodes, according to one embodiment of the present invention. FIG. 3(B) outlines the model performance monitoring for AI-ML assisted positioning using measurements from reference nodes, according to one embodiment of the present invention. FIG. 3(C) depicts the model performance monitoring for AI-ML assisted positioning without the measurements from reference nodes, according to one embodiment of the present invention. FIG. 4 illustrates the timing framework for monitoring the performance of the AI-ML models (400), according to one embodiment of the present invention. FIG. 5 is a block diagram presents an example of a schematic hardware configuration of the network node (500) 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 various positioning methodologies within a communication network (100), including location server-based uplink (UL) positioning, location server-based downlink (DL) positioning, and user equipment-based positioning, according to one embodiment of the present invention. These methodologies enable accurate positioning of a User Equipment (110) through the integration of signal measurements, network coordination, and artificial intelligence / machine learning (AI / ML)-based inference by receiving reference signals at one or more receiver nodes from one or more Transmission Reception Points (TRPs) over orthogonal resources in at least one of the time, frequency, or spatial domains, estimating channel information at the receiver node, processing the estimated channel information, reporting the processed information to a positioning entity located at one of a receiver node, a transmitter node, or a location server, determining the receiver node's location at the positioning entity using an AI / ML model, and inferring the receiver node's location through AI / ML-based positioning or AI / ML-assisted positioning. In one embodiment, FIG. 1(A) illustrates a system and method for location server-based uplink (UL) positioning in a wireless communication network, wherein the system comprises a Training Server (105), User Equipment (UE) (110), a Location Server (115), and nodes of the Next Generation Radio Access Network (NG-RAN) (120). The positioning procedure begins when the Location Server (115) receives a Location Services (LCS) request, thereby initiating a sequence of signaling exchanges to determine the position of the User Equipment (110). Upon receipt of the LCS request, the UE (110) transmits its positioning-related capabilities to the Location Server (115). These capabilities may include, but are not limited to, supported reference signal configurations, positioning method preferences, processing limitations, and AI / ML inference capabilities. Simultaneously or subsequently, the Location Server (115) retrieves information regarding the Transmission Reception Points (TRPs) from the NG-RAN (120) by means of a TRP information transfer procedure. Using the combined capability information and TRP configuration, the Location Server selects appropriate uplink positioning methods, such as UL Time Difference of Arrival (UL-TDoA), UL Angle of Arrival (UL-AoA), and / or multi-Round Trip Time (multi-RTT). The Location Server (115) issues a positioning information request to the NG-RAN (120), which in turn determines and returns the configuration of UL reference signal (UL RS) resources, such as time-frequency allocations and beamforming parameters. Based on this configuration, the Location Server provides the UL RS configuration to the UE (110) and initiates positioning activation, prompting the UE (110) to begin transmission of the configured UL reference signals. During this transmission phase, the UE (110) transmits one or more uplink reference signals to the plurality of NG-RAN nodes (120). Each node is configured to receive and process these signals to obtain channel impulse response (CIR) data. The CIR data may be raw or pre-processed and may include, but is not limited to, the delay profile (DP), power delay profile (PDP), or other signal features such as time of arrival (ToA), angle of arrival (AoA), Doppler shift, and carrier phase information. These measurements are either locally estimated at the NG-RAN nodes or transmitted to the Location Server for centralized processing. Upon receipt of these signal metrics, the Location Server (115) initiates a request for location information from the NG-RAN (120) and subsequently receives the measurements. The Location Server then performs position estimation using the received data in conjunction with one or more AI / ML models. These models are trained at the Training Server (105) using datasets associated with site-specific and configuration-specific context parameters. Such context includes transmission filters, oversampling factors, TRP identifiers, resource configurations, and receiver capabilities, all of which are encoded in a unique "Associated ID" used to ensure consistency between the training and inference stages of the model lifecycle. The estimated position is returned in response to the original LCS request. In parallel, a Model Performance Monitoring (MPM) mechanism is initiated to validate the accuracy and reliability of the AI / ML model employed. The Location Server transmits a model performance monitoring request, prompting the computation of MPM metrics such as position error, range error, or angular error. If reference node measurements (e.g., positioning reference units or high-accuracy ground-truth nodes) are available, the MPM process uses these for validation; otherwise, approximate or historical estimates may be employed. Based on the evaluation, if the model performance deviates beyond a preconfigured threshold, the Training Server (105) is triggered to perform AI / ML model identification, training, or re-selection. Updated models are then transmitted to the Location Server, along with contextual assistance data, for deployment in future inference. This process is governed by a periodic model performance monitoring mechanism that ensures continuous reliability, adaptability to environmental changes, and lifecycle integrity of the AI / ML model. Figure 1(B) illustrates the location server-based downlink (DL) positioning methodology, a key embodiment of the invention. This figure presents a detailed sequence diagram involving four main entities: the Training Server (105), User Equipment (UE) (110), Location Server (115), and the Next Generation Radio Access Network (NG-RAN) (120). Together, these entities coordinate to enable accurate and AI / ML-enhanced UE positioning in 5G-Advanced and 6G networks, especially in Non-Line-of- Sight (NLoS) environments. The positioning process is initiated when the Location Server (115) receives a Location Services (LCS) request from an application or network function. In response, the User Equipment (110) sends its positioning capabilities, which may include the ability to process channel impulse responses (CIR), extract ToA (Time of Arrival), AoD (Angle of Departure), Doppler shifts, carrier phase, and utilize AI / ML-based positioning models. These capabilities are essential for the Location Server (115) to determine which positioning method(s) are best suited to the current environment. Common methods include DL-TDoA (Downlink Time Difference of Arrival), DL-AoD, and enhanced Cell ID (ECID). Once the positioning method is selected, the Location Server (115) triggers TRP (Transmission Reception Point) Information Transfer to the NG-RAN (120), enabling it to configure and transmit the required downlink reference signals. Simultaneously, Assistance Data is shared with both the UE (110) and NG-RAN (120), containing signal configuration parameters, reference signal scheduling, beam IDs, and resource block information that ensure all network components are synchronized for positioning. At this stage, the Training Server (105) plays a crucial role by sending Context Assistance Data to the User Equipment (110). This data enables AI / ML Model Identification, Training, or Selection, ensuring that the AI / ML model used during inference matches the environmental and network context of the training phase. This is facilitated through an Associated-ID, a unique identifier that binds site-specific transmission, reception, and environmental configurations. The UE (110) responds with an AI-ML Model Response, which includes the selected model's identifier, confirming its readiness for use. The Location Server (115) then sends a Positioning Activation request to the NG-RAN (120), prompting it to Transmit Reference Signals (RS). The User Equipment (110) receives these downlink signals and processes them to compute the Channel Impulse Response (CIR) using predefined signal processing techniques. From the CIR, the UE (110) derives several measurements such as ToA, AoD, Doppler, and potentially carrier phase. These measurements, either in raw or processed form (e.g., truncated CIR, power delay profiles), are reported back to the Location Server (115). Notably, the invention includes efficient quantization and encoding methods to reduce overhead during this reporting step. Upon receiving the measurement data, the Location Server (115) uses it to Estimate the Position of the UE (110). This step may also include applying a selected AI / ML model, trained on high-resolution datasets with site-specific signal characteristics. The Associated-ID is used to verify that the model inference context aligns with its training conditions, thereby enhancing prediction accuracy. The resulting location estimate is sent back in the Response LCS Request message. In the lower half of Figure 1(B), the Model Performance Monitoring (MPM) framework is depicted. Periodically, the Location Server (115) sends a Model Performance Monitoring Request to the Training Server (105) and UE (110). The Training Server (105) again provides Context Assistance Data and triggers performance evaluation based on recent positioning results. The AI / ML model's performance is assessed using metrics such as position error, range error, or angular error. If the model underperforms, a Model Update Information step occurs, either updating or replacing the model in use. This ensures that only the most effective models are deployed in real-time positioning tasks. Figure 1(C) presents the User Equipment (UE)-based positioning methodology, a distinct embodiment of the invention in which User Equipment (110) takes on a more autonomous role in estimating its own position. This architecture supports AI / ML-enhanced, on-device positioning, designed for improved performance in environments with complex or degraded radio conditions especially in Non-Line-of-Sight (NLoS) scenarios. The sequence begins with a Location Services (LCS) request received by the Location Server (115). The UE (110) initiates a capabilities exchange, informing the Location Server about its support for local CIR processing, positioning measurement derivation, and AI / ML model execution. These capabilities are crucial for the server to determine the feasibility of UE-side positioning and to select an appropriate assistance mechanism. Based on this information, the Location Server (115) initiates TRP Information Transfer with the NG-RAN (120) to configure reference signal transmission. Following this, Assistance Data which includes signal configuration, resource allocation, and reference signal scheduling—is provided to both the User Equipment (110) and NG-RAN (120). The Training Server (105) concurrently sends Context Assistance Data to the UE. This data enables the UE to identify or select an AI / ML model appropriate for its current environment. Using Associated-IDs that encode site-specific, transmitter, and receiver configurations, the UE selects a model trained under matching conditions and returns an AI-ML Model Response. Once the setup is complete, Positioning Activation is triggered. The NG-RAN (120) transmits Reference Signals (RS) over configured radio resources. The UE (110) receives these signals and computes the Channel Impulse Response (CIR) using internal signal processing techniques. From this CIR, the UE estimates a range of positioning-related metrics, including Time of Arrival (ToA), Angle of Arrival (AoA), Angle of Departure (AoD), delay, Doppler, and carrier phase. These measurements can also be used to infer LoS / NLoS probability, which further enhances positioning accuracy. The major innovation in this embodiment lies in the UE’s ability to execute AI / ML models locally. These models, trained on high-resolution datasets corresponding to specific environmental and radio configurations, enable the UE to autonomously estimate its position in real time. The result is a reduction in network signaling load, faster response times, and more robust performance in rapidly changing or harsh environments. Once the UE computes its position, it optionally transmits the location information to the Location Server (115) or other authorized entities, after which the server responds to the original LCS request. The bottom section of Figure 1(C) introduces the Model Performance Monitoring (MPM) framework, similar to the one in Figure 1(B). The Location Server (115) periodically initiates a Model Performance Monitoring Request. The Training Server (105) supplies updated Context Assistance Data, and the UE (110) evaluates the performance of the AI / ML model using predefined error metrics (e.g., position error, range error, angle error). Based on this evaluation, the UE may initiate model updates, select an alternative model, or report performance results via the Model Performance Monitoring Response message. This closed-loop monitoring mechanism ensures sustained performance even as channel conditions, mobility, or device configurations evolve. FIG.2 illustrates two methods for computing the measurements for reporting CIR samples (200), the process for computing and reporting channel impulse response (CIR) samples and the subsequent training data collection for AI-ML models used in positioning within a wireless communication network. The process enhances the efficiency of reporting CIR measurements in non-line-of-sight (NLOS)-dominated scenarios by reducing signalling overhead and ensures consistency between training and inference stages of AI-ML models through structured data logging, enabling accurate position estimation in 5G-Advanced and 6G networks. The process is divided into two primary methods for computing and reporting CIR measurements, followed by the logging of these measurements for AI- ML training. In one embodiment, FIG. 2(A) illustrates the first method for computing the measurements for reporting CIR samples, a receiver node measures the CIR using reference signals transmitted by one or more nodes over orthogonal resources, which are orthogonal in at least one of the time, frequency, or space domains. The receiver processes the raw CIR to compute measurements such as time of arrival (ToA), angle of arrival (AoA), angle of departure (AoD), Doppler measurements, or carrier phase measurements for at least one propagation path. Alternatively, the receiver may directly estimate the position of a target node using the CIR information. The processed measurements include at least one of the delay profiles (DP) of the channel, consisting of the sample indices of the CIR, or the power and delay profile (PDP) of the channel, representing the absolute square values of the coefficients of the CIR. In this method, the complete CIR, complete PDP, or complete DP is reported to a location server (115) for position estimation, as specified in Table 1 (below given). Specifically, for the complete CIR, the number of samples N is set to the Fast Fourier Transform size NFFT with the window of reported samples Nw' equal to the total window size Nw starting from sample index n0 = 0, with no quantization applied to amplitude or phase. For the complete PDP, the same parameters apply, with phase quantization set to qP = 0. For the complete DP, amplitude quantization is set to qa = 1 and phase quantization to qP = 0. FIG. 2(B) illustrates the second method for computing the measurements for reporting CIR samples, In the second method, to mitigate the significant overhead associated with reporting the raw CIR, raw PDP or other raw measurements, the receiver employs a processing technique involving up-sampling or down-sampling of the CIR by a factor of k , followed by truncation within a window that captures a significant amount of the power contained in the CIR. This window consists of Nw = 2k× N samples, starting from an initial sample index n0. From these processed samples, the strongest Nw' samples are selected, with their amplitude quantized using qa bits and phase quantized using qP bits. The location of these Nw’ strongest samples is reported using a bitmap [!",!"#$,…,&], where the bit value bi is set to 1 if the sample is reported and cleared to 0 otherwise. This method corresponds to the truncated CIR, truncated PDP, or truncated DP reporting schemes in Table 1, where the number of samples N is less than NFFT, and no specific values for Nw', n0, qa or qp are predefined, allowing flexibility based on receiver or location server configuration. Reporting Measurement ' '()*&+,+- Complete CIR'.. / '(0 -- -- Complete PDP '.. / '(0 -- 0 Complete DP '.. / '(0 1 0 Truncated CIR < '.. / -- -- -- --Truncated PDP < '.. / -- -- -- 0Truncated DP < '.. / -- -- 1 0Table 1: Special Cases of measurement reporting scheme The location server (115) utilizes these processed measurements to estimate the position of the target node, with the values of k and N or Nw either chosen by the receiver node or configured by the location server to control the reporting overhead. Following the measurement reporting, the process proceeds to next step where the measurements are logged for training AI-ML models. The measurements are collected under specific transmission contexts (e.g., number of symbols, COMB factor, resource blocks, muting pattern, beam ID) and reception contexts (e.g., sample rate, spatial filters, oversampling factor). The collected data includes positioning parameters, enhanced channel information, node locations, Doppler information, and timestamps, ensuring comprehensive datasets for AI-ML training. To ensure consistency between training and inference stages, the measurements are logged with an associated identifier (Associated-ID), which encodes the site, TRP information, transmission context, and reception context. The Associated- ID is generated as a function of one or more parameters, including the global cell ID (GCID), physical cell ID (PCID), public land mobile network (PLMN), TRP ID, positioning frequency layer ID (PFL), reference signal (RS) resource set ID, RS resource ID, Nw, Nw' and k. The mapping for the Associated-ID is defined as: *,0012,345 = 6&. *89:; + 6$. *?9:; + 6@. *?AB! + 6C. * / D? + 6E. *?.A Associated-ID is unique for each combination of TRP, transmission context and reception context. For parameters not required in a given mapping, the corresponding 6Kis set to 0. The logged measurements are stored in a network data such as a Network Data Analytics Function (NWDAF), or an over-the-top (OTT) server, with a structure that includes the Associated-ID, CIR samples, UE location estimate, TRP location, and timestamp. The structure for logging the positioning measurements is shown in Table 2: Parameters Structure Associated-ID integer cir samples array ('(),) UE location estimate [LM4 , OM4 , PM4(RSTUR*VW)]TRP location [L3Y-, O3Y-, P3Y-(RSTUR*VW)]timeStamp Table 2: Reference Structure for Training Data with Associated IDs Multiple AI / ML models are trained using datasets collected from various transmitter-receiver configurations under static and dynamic Doppler conditions, ensuring robustness across different scenarios. The Associated-ID ensures that datasets are site and configuration-specific, enabling AI-ML models to be trained on consistent data. During inference, the same Associated-ID is used to verify that the inference context matches the training context, ensuring reliable performance. The logged measurements are used to train AI-ML models, which are identified by unique model IDs, and the mapping between model IDs and Associated-IDs is sent to the network to facilitate lifecycle management of the AI-ML models. The process depicted in FIG.2 enables the receiver to either report the full CIR, PDP, or DP for maximum information retention or employ truncation and quantization to reduce signalling overhead, while ensuring that the reported measurements are systematically logged for AI-ML training, thereby optimizing the trade-off between reporting efficiency and positioning accuracy in AI-ML-based positioning systems. FIG. 3 illustrates the monitoring of model performance for different AI-ML-based positioning methods that utilize measurements from reference nodes (300). The methods depicted are integral to the lifecycle management (LCM) of artificial intelligence and machine learning (AI-ML) models in wireless communication networks, enabling the system to assess and maintain the accuracy of positioning estimates for user equipment (UE). These methods are tailored to different positioning approaches, including direct AI-ML positioning and AI-ML assisted positioning, and account for the availability of measurements from reference nodes, such as positioning reference units (PRUs) or other nodes with accurate location estimates. The model performance monitoring (MPM) procedures calculate performance metrics based on factors such as position error (2D or 3D), range error, or angular error (azimuth, elevation, or both), with thresholds configured by the network or the node using the AI-ML model to determine whether model updates or fine-tuning are required. FIG.3(A) outlines the model performance monitoring for direct AI-ML based positioning using measurements from reference nodes. In this approach, the AI-ML model directly estimates the position of the target UE without relying on intermediate measurements such as time of arrival (ToA) or angle of arrival (AoA). The MPM procedure utilizes reference measurements provided by reference nodes to compute the position error (2D or 3D) as the primary MPM metric, as specified in Table 3. The position error is calculated by comparing the AI-ML model’s inferred position estimate with the known location of the reference node. If the position error exceeds a predefined threshold, configured by the location server or the node, the model under test (MUT) is updated or fine-tuned to ensure continued accuracy. Otherwise, the same model is retained for inference. The availability of reference measurements ensures high reliability in assessing the model’s performance, particularly in non-line-of-sight (NLOS) scenarios. FIG. 3(B) outlines the model performance monitoring for AI-ML assisted positioning using measurements from reference nodes. In AI-ML assisted positioning, the AI-ML model enhances traditional positioning methods, such as downlink time difference of arrival (DL-TDoA), uplink time difference of arrival (UL-TDoA), multi-round trip time (m-RTT), downlink angle of departure (DL-AoD), uplink angle of arrival (UL-AoA), or hybrid measurements, by refining intermediate measurements like ToA, TDoA, RTT, AoA, or AoD. The MPM procedure leverages reference measurements to compute multiple metrics, depending on the positioning method, as outlined in Table 3. For ToA or RTT-based methods, both position error (2D or 3D) and range error are calculated. For TDoA-based methods, position error and range error are used. For AoA or AoD-based methods, position error and angular error are computed. For hybrid methods combining TDoA, ToA, AoA, or AoD, the MPM metrics include position error, range error, and angular error. For TDoA-based positioning methods, the entity computes the rangedifference using TDoA measurements for inference, where RK = ZK − Z& ,and the range difference is ∆rK= _*RK. The true range difference is computedusing the locations of the TRP and the reference node, ∆rK̀ = ‖SK − Ŝ‖@ −‖S& − Ŝ‖@, with the deviation calculated as cK = |∆rK − ∆rK̀|. The overallMPM metric is determined by selecting the measurement with the highest line-of-sight (LoS) probability or using error averaging when LoS probabilities are low, as given by: Kì c ∗ where U∗K = argmax SA1Gï K1For computes therange using ToA measurements for inference, where RK = ZK, and the rangeis rK= _*RK. The true range is computed using the location of the referencenode, rK̀ = ‖SK − Ŝ‖@, with the deviation calculated as cK = |rK − rK̀|. Theoverall MPM metric is determined by selecting the measurement with the highest LoS probability or using error averaging when LoS probabilities are low, as given by: ìc ∗ oℎc}c U∗K = argmax SKA1GK1An with a default probability of 0.5 assumed if unavailable. If the MPM metric exceeds the configured threshold, the MUT is updated or fine-tuned. FIG. 3(C) depicts the model performance monitoring for AI-ML assisted positioning without the measurements from reference nodes. In the absence of reference measurements, the MPM procedure relies solely on the measurements available at the entity, such as ToA, TDoA, RTT, AoA, or AoD, depending on the positioning method. The MPM metrics are limited to those that do not require reference node locations, as specified in Table 3. For ToA or RTT-based methods, the range error is computed. For TDoA- based methods, the range error is used. For AoA or AoD-based methods, the angular error is calculated. For hybrid methods, both range error and angular error are employed. The range error for ToA is calculated as cK = |rK − rK̀|, and for TDoAas cK = |∆rK − ∆rK̀|, using inferred measurements reference nodevalidation. The MPM metric is computed similarly to FIG. 3(B), using the “best measurement method” or error averaging based on LoS probabilities. The lack of reference measurements reduces the accuracy of position error calculations, making range or angular errors the primary metrics. If the MPM metric exceeds the threshold, the MUT is updated or fine-tuned, ensuring model reliability under varying channel conditions. Type of AI-ML Ref Position MPM metric model measurements measurement Direct AI-ML Available -- Position error positioning (2D / 3D) AI- ML Available ToA / RTT Position error assisted (2D / 3D), range positioning error AI-ML Not available ToA / RTT range error assisted positioning AI-ML Available TDoA Position error assisted (2D / 3D), range positioning error AI-ML Not available TDoA range error assisted positioning AI-ML Available AoA / AoD Position error assisted (2D / 3D), angle positioning error AI-ML Not available AoA / AoD angle error assisted positioning AI-ML Available Hybrid (TDoA or Position error assisted ToA / AoA or AoD) (2D / 3D), range positioning error, angle error AI-ML Not available Hybrid (TDoA or Position error assisted ToA / AoA or AoD) (2D / 3D), range positioning error, angle error Table 3 MPM Metric Selection for AI-ML Positioning Based on Measurement and Reference Context The methods in FIG. 3 ensure robust MPM by adapting to the availability of reference measurements and the specific positioning method, enhancing the reliability of AI-ML models for positioning in 5G-Advanced and 6G networks, particularly in NLOS-dominated scenarios. FIG. 4 illustrates the timing framework for monitoring the performance of the AI-ML models (400), used for positioning within a wireless communication network, according to one embodiment of the present invention. This framework enables periodic evaluation of the inference accuracy of the deployed AI-ML model, thereby supporting lifecycle management, model reliability assessment, and adaptive optimization in real-time systems. In one embodiment, the framework consists of a sequence of discrete time intervals, referred to as slots, which are indexed sequentially (e.g., 0 to 29). Each slot corresponds to a transmission or processing opportunity in the wireless communication system’s timeline. The system is configured to trigger model performance monitoring (MPM) at specific slots determined by two key timing parameters: *-4YK15and *1^^043. The parameter *-4YK15defines the periodicity of monitoring in terms of the number of slots between successive performance checks. The parameter *1^^043specifies the exact offset within each period at which the model evaluation is to occur. Accordingly, the slot index at which the model performance monitoring is executed is computed using the expression: Slot index = k × nperiod + noffset where k is a non-negative integer representing the index of the monitoring cycle. In an example embodiment, the first monitoring event occurs at slot index 0, and subsequent monitoring events occur at slot indices 6, 12, 18, and 24. This indicates that nperiod = 6 and noffset = 0 for this specific example. These values are configurable and may vary based on network requirements, deployment scenarios, or implementation policies. At each designated monitoring slot, the AI-ML model under test (MUT) is evaluated using one or more MPM metrics, such as position error in two or three dimensions, range error, or angular error. Additionally, if the probability of line-of-sight (LoS) or non-line-of-sight (NLoS) is available, it may be incorporated into the performance evaluation; otherwise, a default probability (e.g., 0.5) may be assumed. Upon computing the MPM metric at the scheduled slot, the system compares the result against a predefined threshold value. If the metric exceeds the threshold, indicating degradation in the model’s inference quality, the system may initiate one or more remedial actions, such as fine-tuning the model, switching to an alternative model, retraining the model, or temporarily disabling the model. If the metric is within acceptable bounds, the model is retained for continued use. This periodic monitoring mechanism provides a structured and programmable approach for ensuring the consistent and accurate operation of AI-ML models under live network conditions. The slot-based monitoring strategy allows the system to maintain real-time awareness of model effectiveness, adapt to dynamic changes in channel conditions or site configurations, and ensure that AI-ML inference remains trustworthy throughout its deployment lifecycle. Thus, FIG. 4 presents a foundational component of the invention’s lifecycle management architecture for AI-ML- based positioning in 5G-Advanced and future 6G networks. FIG.5 outlines an exemplary schematic hardware architecture (500) of a network node configured to support artificial intelligence and machine learning (AI / ML)-based positioning functionalities within a wireless communication system, according to one embodiment of the present invention. The illustrated network node may be implemented as part of a base station (e.g., gNB), edge server, location management function (LMF), or a centralized / cloud-based positioning server in 5G / 6G networks. The network node (500) includes a network interface module (510), which is operably coupled to the radio access network (RAN) components, core network functions, and / or other external systems and configured to receive reference signals from one or more Transmission Reception Points (TRPs) over orthogonal resources in at least one of the time, frequency, or spatial domains. The network interface module (510) may support various standard protocols (e.g., NRPPa, F1-C / U, N2, N3, NGAP, Xn) for transmitting and receiving data related to user equipment (UE) measurements, positioning reference signals, assistance information, and model metadata. In certain embodiments, the network interface (510) may also facilitate connectivity to a Network Data Analytics Function (NWDAF), OAM systems, or cloud-native orchestration platforms. A processing unit (520) is provided to perform computational tasks, including estimating channel information based on the received reference signals, where the channel information includes at least one of CIR, DP, or PDP, processing the estimated channel information to extract positioning parameters by computing the receiver node’s position through estimating at least one of ToA, AoA, or AoD using processed CIR data and determining enhanced channel characteristics, Non-Line-of-Sight (NLoS) or Line-of-Sight (LoS) probabilities, or positioning parameters such as ToA, AoA, AoD, Doppler shift, or carrier phase, reporting the processed information to a positioning entity located at a receiver node, transmitter node, or location server, determining the receiver node’s location using an AI / ML model and inferring the receiver node’s location through AI / ML-based positioning or AI / ML- assisted positioning. The processing unit (520) is further configured to periodically monitor AI / ML model performance using at least one performance metric, including position error, range error, or angular error, and update the AI / ML model when its performance metrics exceed a predefined threshold, and minimize reporting overhead by performing at least one of up-sampling or down- sampling channel impulse response (CIR) samples, truncating CIR samples, or quantizing the strongest CIR samples before reporting and ensure consistency between AI / ML training and inference stages by generating an associated identifier (ID) derived from transmission and reception parameters. The processing unit (520) may be implemented using one or more general-purpose processors (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), or AI / ML accelerators (e.g., tensor processing units or neural processing units). The architecture further includes a memory unit (530) operably connected to the processing unit and configured to temporarily store intermediate computation results, session-specific configuration data, buffering inputs / outputs of the AI / ML inference engines, runtime control states, channel information, processed positioning data, AI / ML model parameters, and training datasets generated from processed CIR, DP, or PDP measurements. The memory may include volatile storage such as DRAM and high-bandwidth cache modules to support low-latency operation. A persistent storage module (540) is used to store a plurality of datasets and models, including trained AI / ML models, calibration data, beamforming configurations, reference signal configurations, model performance monitoring logs, AI / ML training datasets collected from multiple transmitter-receiver configurations under static and dynamic Doppler conditions, performance metrics, and historical positioning data. The persistent storage may be implemented using non-volatile storage such as SSDs or NVMe arrays and may be further secured using encryption modules to protect model integrity and proprietary datasets. In one embodiment, the network node (500) is also configured to support model lifecycle management operations, including model updates, training data ingestion, periodic re-training triggers, and rollback to previous model versions, ensuring that the positioning models deployed within the node remain accurate and robust to changing environmental and propagation conditions. The node may further interface with multiple UEs and RAN entities simultaneously, aggregating measurement data from distributed sources and coordinating collaborative positioning sessions. In certain embodiments, the node also supports federated learning, where updated model weights are aggregated from edge devices without requiring raw data transfer, thereby preserving user privacy. The modular and scalable nature of the architecture allows it to be deployed in various configurations, such as: · Edge computing nodes, co-located with distributed gNBs or MEC platforms for low-latency inferencing. · Centralized location servers, hosted in core networks or regional data centers. · Cloud-native positioning platforms, which manage positioning services across a wide area and coordinate across multiple domains. Accordingly, the architecture of network node (500) provides the computational, storage, and communication capabilities necessary to support high-accuracy, AI / ML-enhanced positioning services 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: method for estimating the User Equipment (UE) position in a communication network, the method comprising: receiving reference signals at one or more receiver nodes from one or more Transmission Reception Points (TRPs) over orthogonal resources in at least one of the time, frequency, or spatial domains; estimating channel information at the receiver node; processing the estimated channel information; reporting either the processed information or the intermediate positioning measurement to a positioning entity, the positioning entity being located at one of a receiver node, a transmitter node, or a location server; training the AI-ML model using datasets derived from collected positioning data; determining the target node's location at the positioning entity either directly using an AI / ML model with processed or raw CIR as model input or using intermediate measurements inferred by the AI-ML model; inferring either the target node's location or the intermediate positioning measurement through AI / ML-based positioning or AI / ML- assisted positioning respectively; checking the consistency between the training and inference phase of the AI-ML model by the positioning entity; andperforming the life cycle management of the AI-ML model which include model performance monitoring, activating, deactivating, retraining, fine-tuning and redeploying the AI-ML.

2. The method as claimed in claim 1, further comprising: periodically monitoring the AI / ML model’s performance using at least one performance metric, including position error, range error, or angular error; updating, finetuning, or deactivating the AI / ML model when its performance metrics deteriorate beyond a predefined threshold; reducing reporting overhead by performing at least one of up- sampling or down-sampling CIR samples, truncating CIR samples, or quantizing the strongest CIR samples before reporting; and ensuring consistency between AI / ML training and inference by generating an associated identifier (ID) based on transmission and reception parameters.

3. The method as claimed in claim 2, wherein the channel impulse response is truncated within a window of length (Nw) samples beginning at a defined sample index (n0), the truncated channel impulse response being up-sampled or down-sampled, normalized in amplitude and phase based on power or peak values, and a number of strongest samples (N′w) is quantized and their indices reported using a bitmap corresponding to the window length, wherein at least one of the window length, the number ofstrongest samples or a scaling factor (k) used to configure the window length relative to a base number of samples is determined by the receiver or a location server.

4. The method as claimed in claim 2, wherein the CIR is truncated within a configurable window, a plurality of strongest samples are selected and quantized following normalization based on either peak or power values in amplitude, power, or phase, and the locations of the quantized samples are reported in a bitmap, and the receiver optionally reports a quantized value of either the absolute power or the absolute amplitude of the strongest sample together with its corresponding sample index.

5. The method as claimed in claim 4, wherein the bitmap is configured by setting bits corresponding to the sample indices of the strongest CIR samples selected for reporting to a logic value of one and clearing all remaining bits to a logic value of zero.

6. The method as claimed in claim 1, wherein the channel information includes at least one of a channel impulse response (CIR), delay profiles (DP), power delay profiles (PDP), and channel frequency response (CFR).

7. The method as claimed in claim 1, wherein the AI / ML model estimates the position of the target node either directly using the channel information or the processed channel information.

8. The method as claimed in claim 7, wherein the receiver node is configured to report, to the positioning entity, at least one of raw channel information, processed channel information, or processed and truncated channel information, when the positioning entity is distinct from the receiver node.

9. The method as claimed in claim 7 wherein the processed channel information comprises at least one of: a power delay profile; or a delay profile derived from a channel impulse response; wherein power values are quantized when the processed channel information is reported to the positioning entity, and wherein a time domain representation is up-sampled or down-sampled by a factor of a non- negative integer.

10. The method as claimed in claim 1, wherein the AI-ML infers at least one of Time of Arrival (ToA), Angle of Arrival (AoA), or Angle of Departure (AoD) using processed CIR data which is used to estimate the location of the target node.

11. The method as claimed in claim 6, wherein processing channel information includes one or multiple of delay profile (DP) information, power delay profile (PDP) information computed using channel impulse response (CIR).

12. The method as claimed in claim 1, wherein reporting at least one of the estimated position, refined channel information, or intermediate positioning measurements to a positioning entity at a receiver node, transmitter node, or location server.

13. The method as claimed in claim 1, wherein training an AI-ML model utilizes a dataset generated from one of the processed CIR, DP, or PDP measurements.

14. The method as claimed in claim 2, wherein processing of the channel impulse response (CIR) further comprising at least one of: up-sampling or down-sampling the CIR by a factor k prior to selecting a window of samples; or selecting a window of samples prior to up-sampling or down- sampling, the window commencing at an index n0; either peak or power normalizing the windowed samples; selecting a subset of strongest samples; quantizing at least one of amplitude, power, or phase of the selected samples; finding the sample index and obtaining either a quantized power value or a quantized amplitude value corresponding to the strongest CIR sample; converting the quantized values into bits using a bitmap representation; andreporting the measurement report to a positioning entity.

15. The method as claimed in claim 1, wherein positioning data collection includes positioning measurements, enhanced channel information, transmitter and receiver node locations, Doppler information, and timestamps.

16. The method as claimed in claim 1, wherein model performance monitoring (MPM) is performed periodically using performance metrics, wherein the performance metric is computed using multiplicity of position error, range error, or angular error.

17. The method as claimed in claim 3, wherein an associated identifier (ID) is computed using one or more of the global cell ID (GCID), physical cell ID (PCID), public land mobile network (PLMN) ID, positioning frequency layer ID, positioning reference signal (PRS) or sounding reference signal (SRS) resource ID, PRS / SRS resource set ID, transmission / reception point (TRP) ID, and parameters Nw, N′w, and k.

18. A network node for estimating the User Equipment (UE) position in a communication network, the network node comprising: a receiver configured to receive reference signals from one or more Transmission Reception Points (TRPs) over orthogonal resources in at least one of the time, frequency, or spatial domains; a processor configured to:estimate channel information based on the received reference signals and RS configurations shared by the LMF; processing the estimated channel information, the processing being configured to either extract positioning measurements or to compute processed channel information or processed and truncated channel information; and report the processed information to a positioning entity located at a receiver node, transmitter node, or location server; train the AI / ML model using datasets derived from collected positioning data; determine the receiver node’s location based on the processed channel information using an AI / ML model; infer the receiver node’s location through direct AI / ML-based positioning or AI / ML-assisted positioning; a memory operably connected to the processor and configured to store channel information, processed channel information, and AI / ML model architecture and model parameters; and a storage unit configured to retain AI / ML training datasets, performance metrics, and historical positioning data.

19. The network node as claimed in claim 18, wherein the processor sitting at the positioning entity is further configured to: periodically monitor AI / ML model performance using at least one performance metric, including position error, range error, or angular error;update the AI / ML model when its performance metrics deteriorate beyond a predefined threshold; reduce reporting overhead by performing at least one of up-sampling or down-sampling channel impulse response (CIR) samples, truncating CIR samples, or quantizing atleast one of the phase, amplitude and time information of the CIR samples from a window of samples before reporting; and ensure consistency between AI / ML training and inference stages by generating an associated identifier (ID) derived from transmission and reception configuration parameters.

20. The network node as claimed in claim 19, wherein the window of channel impulse response (CIR) samples includes a subset of consecutive samples that capture all the sample have power higher than threshold fraction of total received signal power contained in the CIR.

21. The network node as claimed in claim 18, wherein the channel information includes at least one of CIR, DP, PDP, or CFR.

22. The network node as claimed in claim 18, wherein the processor is configured to compute the receiver node’s position either directly using CIR information or by estimating at least one of ToA, AoA, or AoD using processed CIR data.

23. The network node as claimed in claim 18, wherein the processor is configured to estimate channel characteristics which includes atleast one of the Non-Line-of-Sight (NLoS) or Line-of-Sight (LoS) probabilities, or extract positioning measurements such as ToA, AoA, AoD, Doppler shift, or carrier phase.

24. The network node as claimed in claim 18, wherein the processor is configured to report at least one of the estimated position, refined channel information, processed channel information, or positioning measurements to a positioning entity at one of the receiver node, transmitter node, or location server.

25. The network node as claimed in claim 18, wherein the memory stores training datasets generated from processed CIR, DP, or PDP measurements.

26. The network node as claimed in claim 18, wherein the storage unit retains datasets collected from one or more receiver using the reference signal transmitted by one or more transmitter using one or more reference signal, transmitter configurations and receiver configurations under static and dynamic Doppler conditions.

27. The network node as claimed in claim 18, wherein the processor performs model performance monitoring (MPM) periodically using performance metrics including position error, range error, or angular error.

28. The method as claimed in claim 27, wherein MPM metric is computed either based on the error in the most reliable measurement or based on the weighted average error of all the measurements wherein the weightage is computed based on the reliability of each measurement.

29. The method as claimed in claim 28, wherein the reliability of the measurement is determined based either solely on the quality of the communication link including metrics such as signal-to-noise ratio (SNR), interference levels, and the probability of the link being line-of-sight (LoS) or based on a consensus derived from intermediate measurements, computed using a least squares-based intersection error method combined with a subset selection algorithm and the link quality metrics.

30. The method as claimed in claim 27, wherein MPM is performed periodically at fixed intervals defined in units of seconds, slots, subframes, or frames, and wherein the specific time instants, slots, subframes, or frames at which MPM is executed are determined using an offset specified in the same unit, the MPM periodicity and offset being either autonomously determined by the node where the model is deployed or configured to the said node by one of a multiplicity of the base stations, a location management function, or an AI / ML lifecycle management (LCM) entity.

31. The method as claimed in claim 27, wherein measurement pattern management (MPM) is performed using assistance information comprisingat least one of: time of arrival (ToA), round trip time (RTT), reference signal time difference (RSTD) analogous to 5G-NR RSTD, relative time of arrival (RToA) analogous to 5G-NR RToA, truncated or raw channel impulse response (CIR), truncated or raw power delay profile (PDP), truncated or raw delay profile (DP), multipath time of arrival measurements, multipath angle measurements, multipath power measurements, and multipath phase measurements; and further comprising line-of-sight (LoS) and non-line-of- sight (NLoS) probability indicators, as well as location estimates obtained from nodes other nodes in the area surrounding the node whose location is being inferred using the AI / ML model.

32. The method as claimed in claim 27, wherein, when MPM is performed by a receiver node, the MPM decision is reported by the receiver to one of a plurality of base stations, a location management function, or an AI / ML lifecycle management (LCM) entity, via one of the communication mechanisms which include an uplink grant, a positioning protocol analogous to LPPs, or a communication protocol analogous to RRC.

Images