A method and system for processing user equipment (UE) environment information (UEI) in a wireless communication network

Abstract

The invention introduces an AI / ML-based CSI compression and reconstruction system for next-generation wireless networks, enabling efficient and scalable Channel State Information (CSI) reporting. It extends traditional spatial and frequency domain compression to include the temporal domain, significantly reducing feedback overhead while maintaining high downlink throughput and beamforming accuracy. The system leverages adaptive AI / ML models such as autoencoders, LSTMs, and transformers to exploit multi-dimensional CSI correlations. A server-driven architecture selects optimal models and hyperparameters based on real-time CSI statistics, enabling deployment across base stations and user equipment (UE) with minimal signaling. Compressed CSI is transmitted using standard uplink / downlink channels and reconstructed at the base station to recover full channel or precoding matrices. Quality metrics and feedback indicators ensure model consistency and performance monitoring. This approach achieves superior CSI fidelity, reduces quantization loss, lowers processing complexity, and supports dynamic adaptation to channel conditions making it ideal for 5G-Advanced and future 6G systems with massive MIMO configurations.

Description

[0001] A METHOD AND SYSTEM FOR PROCESSING USER EQUIPMENT (UE) ENVIRONMENT INFORMATION (UEI) IN A WIRELESS COMMUNICATION NETWORK

[0002] Field of the Invention

[0003] The present invention relates to wireless communication networks, specifically to methods and systems for enhancing downlink transmission efficiency by leveraging spatial, temporal, and frequency correlations to optimize network performance.

[0004] Background of the Invention

[0005] Wireless communication systems are widely deployed to provide a variety of communication services, including voice, data, and multimedia, to multiple users by efficiently utilizing shared system resources such as bandwidth and transmission power. Such systems commonly employ multiple access techniques, including but not limited to Code Division Multiple Access (CDMA), Frequency Division Multiple Access (FDMA), Time Division Multiple Access (TDMA), Orthogonal Frequency Division Multiple Access (OFDMA), and Single Carrier Frequency Division Multiple Access (SC-FDMA). These systems facilitate robust communication between a base station (BS) or network entity (e.g., gNodeB or gNB) and user equipment (UE) by enabling precise channel estimation and feedback mechanisms. A fundamental aspect of optimizing wireless communication performance, particularly in downlink transmission, involves the accurate acquisition and reporting of Channel State Information (CSI). CSI represents the state of the communication channel between the network and the UE, enabling efficient beamforming and resource allocation. Conventionally, CSI is obtained at the UE through the reception and processing of CSI-Reference Signals (CSI-RS) transmitted by the network. The UE estimates the downlink channel based on these reference signals and provides feedback to the network, typically in a compressed or encoded form, to facilitate adaptive transmission strategies. The CSI-RS ports are mapped to physical antenna ports at the network via a precoder, which is often proprietary and opaque to the UE, adding complexity to the feedback process.

[0006] As standardized up to 3GPP Release 17, CSI feedback mechanisms predominantly relied on non-AI-based approaches, such as codebookbased reporting and compression techniques limited to the spatial and frequency domains, focused on reducing feedback overhead by exploiting spatial and frequency correlations. However, these approaches fail to leverage temporal correlations in the channel, resulting in suboptimal compression efficiency and a loss of critical channel information, which adversely impacts downlink throughput.

[0007] With the advent of 5G advanced systems and the anticipated transition to 6G, as contemplated in 3GPP Release 18 and beyond, the demands on wireless networks have intensified. These demands include supporting explosive data traffic, dramatically increased per-user transmission rates, accommodation of a growing number of connected devices, ultra-low end-to-end latency, and enhanced energy efficiency. Technologies such as massive Multiple Input Multiple Output (MIMO), full- duplex communication, non-orthogonal multiple access (NOMA), and ultra- wideband operation are being explored to meet these requirements. In this context, CSI feedback mechanisms must evolve to handle the increased complexity and data volume, particularly in massive MIMO deployments where the number of antenna elements generates substantial CSI data.

[0008] In Release 18 introduced the potential of Artificial Intelligence and Machine Learning (AI / ML) for CSI compression, with an initial emphasis on temporal domain compression. However, these efforts remain nascent and lack a holistic approach that integrates spatial, temporal, and frequency domains concurrently. The inability of existing solutions to fully exploit the multidimensional nature of CSI restricts their effectiveness in nextgeneration networks, where maintaining high throughput and minimizing feedback overhead are critical.

[0009] Therefore, there is a need in the art for an improved system and method for CSI compression and reconstruction that overcomes the deficiencies of prior art by leveraging AI / ML techniques to efficiently compress CSI across spatial, temporal, and frequency dimensions, thereby enhancing downlink throughput and meeting the stringent performance demands of advanced wireless communication systems.

[0010] Objective of the Invention

[0011] The principal objective of this invention is to improve the performance of a wireless communication network by providing a method and system for efficiently processing and compressing User Equipment Environment Information (EUI) through the application of Artificial Intelligence / Machine Learning (AI / ML) techniques, wherein said processing and compression exploit correlations across spatial, frequency, and temporal domains.

[0012] Another objective of this invention is to provide a method for compressing and reconstructing EUI utilizing specific AI / ML models, including Long Short-Term Memory (LSTM) autoencoders, regular autoencoders, or CSINeT, within an AI / ML framework.

[0013] Another objective of this invention is to facilitate the selection and tuning of AI / ML models and their hyperparameters to achieve optimal compression efficiency while preserving essential environmental information.

[0014] Another objective of this invention is to enhance wireless network performance metrics, such as throughput or Signal-to-Noise Ratio (SNR), by enabling accurate reconstruction of EUI from compressed data at a base station using an AI / ML framework. Further objective of this invention is to ensure adaptability of the wireless network by providing a mechanism for updating AI / ML models employed in EUI processing through an over-the-air update process, said updates being responsive to real-time performance feedback.

[0015] Summary of the Invention

[0016] The present invention pertains to a method, system, base station, server, and computer-readable medium for enhancing downlink transmission efficiency in a wireless communication network by processing User Equipment Environment Information (EUI) through advanced compression and reconstruction techniques utilizing an Artificial Intelligence / Machine Learning (AI / ML) framework. The invention is directed to optimizing network performance by efficiently managing EUI, which comprises data indicative of environmental conditions affecting one or more User Equipment (UE), such as interference, mobility, or location-specific factors, distinct from conventional Channel State Information (CSI).

[0017] In one embodiment the first configuration information from the UE, said information including data sets with a rank indicator, beam indexes, compression ratio, quantization levels, and feedback data and analysing the said data sets to determine an optimum AI / ML model and hyperparameters for EUI compression and configuring the network with the selected model and parameters. In another embodiment a network server operationally coupled to the base station analyses the collected EUI datasets to select and tune the AI / ML model and its hyperparameters, ensuring optimal compression efficiency while preserving critical environmental information for subsequent reconstruction.

[0018] In another embodiment the selected model identified by a unique identifier (ID) or accompanied by its hyperparameters is transmitted from the server to the base station and subsequently to the UE via downlink channels, such as Downlink Control Information (DCI) or Physical Downlink Control Channel (PDCCH), while the UE may signal the model ID back to the network via uplink channels like Physical Uplink Control Channel (PUCCH) or Physical Uplink Shared Channel (PUSCH) to ensure the synchronization of the AI / ML framework across both ends of the communication link for consistent compression and reconstruction.

[0019] In another embodiment the first configuration information determines the optimum AI / ML model, compresses the EUI using said model exploiting spatial, temporal, and frequency correlations and transmits the compressed EUI as second configuration information to the base station for reducing uplink feedback overhead while retaining essential environmental details for network optimization such as beamforming.

[0020] In further embodiment the base station receives the compressed EUI, reconstructs it using the corresponding AI / ML model to recover environmental condition data or precoding vectors, and measures performance metrics like throughput or Signal-to-Noise Ratio (SNR), thereby providing accurate EUI for effective downlink beamforming and enhanced transmission throughput.

[0021] The invention further encompasses adaptability, wherein the base station updates the AI / ML model over-the-air based on real-time performance feedback, ensuring sustained efficiency under dynamic conditions. By leveraging AI / ML models such as LSTM autoencoders, regular autoencoders, or CSINeT. The invention achieves superior compression and reconstruction of EUI, extending beyond traditional CSI processing to optimize wireless network operations.

[0022] Brief description of the drawings

[0023] 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.

[0024] Figure 1 illustrates a conventional flowchart for non-AI-based CSI compression (100), according to one embodiment of the present invention. Figure 2 illustrates the system architecture for AI / ML-based CSI compression (200), according to one embodiment of the present invention.

[0025] Figure 3 illustrates the autoencoder and decoder structure for AI / ML- based CSI compression (300), according to one embodiment of the present invention.

[0026] FIG. 4 is a flow chart of a method for processing User Equipment (UE) environment information (UEI) in a wireless communication network according to one embodiment of the present invention.

[0027] Figure 5 illustrates a timing diagram for CSI compression using AI / ML (500), according to one embodiment of the present invention.

[0028] 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.

[0029] Throughout the drawings, it should be noted that like reference numbers are used to depict the same or similar elements, features, and structures.

[0030] Detailed Description of the Invention

[0031] 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.

[0032] 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.

[0033] 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.

[0034] 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.

[0035] 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.

[0036] 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.

[0037] Figure 1 illustrates a conventional flowchart for non-AI-based CSI compression (100), The figure depicts the interaction between a network entity (110), such as a base station (BS), and one or more user equipment (UE) devices (120) in a wireless communication system.

[0038] According to example embodiment of the invention, the network entity (110) initiates the CSI acquisition process by transmitting a channel state information reference signal (CSI RS) to the UEs (120). The CSI RS is a predefined signal that enables the UEs to estimate the condition of the downlink (DL) wireless channel from the network to each UE.

[0039] Upon receiving the CSI RS, the UEs (120) perform downlink channel estimation based on the received reference signals. As shown in block (130), each UE compresses the estimated CSI in the spatial and frequency domains. This compression is performed using non-AI-based approaches, typically including predefined codebook quantization methods or low- complexity signal processing techniques that apply dimensionality reduction or sub-band-based compression. These conventional methods are designed to reduce the amount of feedback data while maintaining an acceptable level of accuracy in the reconstructed CSI at the network side.

[0040] After compressing the CSI, the UEs (120) transmit the compressed CSI data back to the network entity (110) over an uplink channel, such as the Physical Uplink Shared Channel (PUSCH) or Physical Uplink Control Channel (PUCCH), depending on system configuration and resource allocation.

[0041] Once the compressed CSI is received, the base station (BS), as part of the network entity (110), performs decompression of the CSI data, as illustrated in block (140). The reconstructed CSI is then utilized by the BS to facilitate downlink transmission strategies, including but not limited to, beamforming, beam nulling and dynamic scheduling of transmission resources. These operations rely on the accuracy of the reconstructed CSI to achieve optimal spectral efficiency and interference management. The procedure illustrated in FIG. 1 represents the conventional approach to CSI compression and feedback as employed in legacy 3GPP systems up to Release 17. While effective for basic spatial and frequency domain compression, this approach is inherently limited by its inability to exploit temporal redundancy in the channel statistics. Moreover, the reliance on static or codebook-based compression techniques can result in suboptimal reconstruction quality, particularly in highly dynamic or dense deployment scenarios, thereby constraining the overall downlink throughput and system performance.

[0042] Figure 2 illustrates the system architecture for AI / ML-based CSI compression (200). The system is logically partitioned into two primary domains: a network-side architecture (210) and a user equipment (UE) side architecture (220). The architecture enables context-aware, model-driven CSI compression and reconstruction across spatial, temporal, and frequency dimensions using artificial intelligence and machine learning (AI / ML) models.

[0043] In one embodiment, the network-side architecture (210) comprises a computational subsystem (230) configured for AI / ML model training, validation and inference. The computational subsystem (230) includes, individually or in combination, graphical processing units (GPUs), central processing units (CPUs), cloud-based servers, or edge computing nodes capable of high-performance computation. It is responsible for ingesting CSI datasets collected from UEs and executing training algorithms to derive optimal model architectures and corresponding hyperparameters for CSI compression and reconstruction. The hyperparameters comprise at least one of the following: the number of layers in the model, the number of nodes per layer, the kernel size used for convolutional operations; the activation function, the mini-batch size indicating the number of channel samples processed per training iteration, the learning rate used for gradient descent optimization, the choice of optimization algorithm and a regularization method including, but not limited to, dropout or weight regularization, wherein the hyperparameters are tuned using one or more validation datasets to optimize model performance.

[0044] The computational subsystem (230) interfaces with a base station (BS) node (250) through an intermediate protocol interface module (240). The protocol interface (240) facilitates bi-directional data communication between the BS (250) and the computational subsystem (230). In particular, the BS transmits CSI datasets to the computational subsystem via interface (240) for model training and validation. Upon completion of the training phase, the computational subsystem returns one or more trained AI / ML models and associated hyperparameters to the BS over the same interface. The selected AI / ML model is configured to exploit correlations in one or more of the spatial, temporal, and frequency domains of the CSI to improve compression efficiency and increase downlink throughput.

[0045] The base station (250) may include, but is not limited to, a Central

[0046] Unit (CU), a Distributed Unit (DU), and a Radio Unit (RU), which may be configured as discrete modules or integrated combinations such as CU+Dll or DU+RU, depending on deployment architecture. These components are configured to store, manage, and apply the received AI / ML models to perform CSI compression prior to downlink transmission and to facilitate CSI reconstruction upon receiving feedback from UEs.

[0047] The UE-side architecture (220) comprises a plurality of UE devices (260). Each UE (260) is configured to receive either the complete trained model or a unique model identifier (Model ID) and associated hyperparameters from the BS (250). The UEs utilize the received model or parameters to perform CSI generation and compression. The compressed CSI is then transmitted from the UEs to the BS over a physical uplink channel, such as the Physical Uplink Shared Channel (PUSCH) or Physical Uplink Control Channel (PUCCH). The base station receives model-related feedback or configuration information from the UE via a Physical Uplink Control Channel (PUCCH) or a Physical Uplink Shared Channel (PUSCH).

[0048] In certain embodiments, the model ID transmitted to the UE may correspond to a pre-trained model stored locally at the UE or identified via a standardized lookup table. In other embodiments, the full model and its parameters may be explicitly transmitted to the UE. The UE may perform local inference using the received model to compress the CSI and subsequently indicate the model ID used during compression to the network for correct reconstruction. The system architecture shown in FIG. 2 supports dynamic model adaptation based on operating scenarios. The model selection process may be contextually tailored to specific deployment environments, such as outdoor urban cells, indoor hotspots, or high-mobility use cases. Additionally, model and hyperparameter selection may be performed at the granularity of sectors, cells, or UE clusters to improve adaptability and performance.

[0049] Figure 3 illustrates the autoencoder and decoder structure for AI / ML- based CSI compression (300). The structure (300) provides a framework for compressing and reconstructing channel state information (CSI) using a neural network-based architecture comprising an autoencoder (310) and an auto decoder (320), with a latent representation referred to as a codeword (330).

[0050] In one embodiment, the autoencoder (310) is disposed at the transmitting entity, such as a user equipment (UE), and is configured to receive CSI input data generated from channel estimation procedures. The input CSI data may include, without limitation, time-domain or frequencydomain channel coefficients, precoding matrices, or other channel quality indicators. The autoencoder (310) includes a plurality of encoding layers, which may be implemented using fully connected neural networks, convolutional layers, recurrent layers (e.g., LSTM), or any combination thereof. The selected AI / ML model comprises an autoencoder architecture selected from the group consisting of dense neural networks, convolutional neural networks (CNNs), long short-term memory (LSTM) networks, and transformer-based networks.

[0051] The autoencoder (310) processes the input CSI and outputs a lowerdimensional latent representation denoted as codeword (330). The codeword (330) is a compact encoding of the original CSI, preserving the essential information required for accurate reconstruction. The dimensionality of the codeword (330) is significantly smaller than the input CSI, thereby achieving substantial compression and reducing the uplink feedback overhead. The UE is further configured to quantize the CSI before transmission using an AI / ML encoder. The codeword (330) may be transmitted from the UE to the base station (BS) using physical uplink resources such as the Physical Uplink Shared Channel (PUSCH) or Physical Uplink Control Channel (PUCCH).

[0052] At the receiving side, the network entity, typically the BS or a network server, comprises an auto decoder (320) that is configured to receive the codeword (330) and reconstruct the original CSI therefrom. The auto decoder (320) includes a plurality of decoding layers corresponding to, or complementary to, the layers used in the autoencoder (310). The decoder (320) generates a reconstructed version of the CSI that closely approximates the original input CSI used at the UE. This reconstructed CSI may then be used by the BS to perform radio operations such as beamforming, precoding vector selection, user scheduling, and link adaptation. The compressed CSI comprises at least one of: a compressed channel matrix; or a compressed precoding matrix configured to optimize signal-to-noise ratio (SNR).

[0053] The autoencoder (310) and auto decoder (320) may be trained jointly as an end-to-end model using supervised learning techniques. The training dataset may comprise actual measured CSI data collected across a wide range of propagation environments. The training process seeks to minimize a reconstruction loss function, such as mean squared error (MSE), between the original and reconstructed CSI.

[0054] In some embodiments, the trained parameters of the autoencoder (310) and auto decoder (320) are shared between the UE and the network. In other embodiments, a model identifier (Model ID) corresponding to the encoder-decoder pair is transmitted alongside the codeword (330) to ensure that the network applies the correct decoder for reconstruction. The encoder-decoder structure may also be dynamically updated or adapted to different scenarios, such as urban macro cells, indoor hotspots, or high- mobility vehicular environments.

[0055] The structure (300) enables significant improvements over traditional CSI compression techniques. Conventional approaches, such as codebook-based or transform-domain quantization, are limited in their ability to compress multi-dimensional CSI data effectively and cannot adapt to changing channel conditions. In contrast, the AI / ML-based structure of FIG. 3 provides a data-driven mechanism for learning optimal compression mappings that exploit spatio-temporal channel correlations, thereby enabling higher fidelity CSI reconstruction, reduced uplink overhead, and enhanced spectral efficiency.

[0056] Accordingly, the autoencoder and decoder structure 300 provides an effective and scalable solution for CSI compression in advanced wireless communication systems, including 5G-Advanced and 6G networks, where massive MIMO and dense deployments necessitate efficient and adaptive CSI feedback mechanisms.

[0057] Figure 4 is a flow chart of a method for processing User Equipment (UE) environment information (UEI) in a wireless communication network (400). The method may be implemented by a network entity, such as a base station, and is directed toward enabling adaptive model selection and configuration for AI / ML-based CSI compression using input derived from UE-reported environmental and feedback information.

[0058] At step 410, the method includes receiving, at the base station, a first configuration information from one or more user equipment (UE) devices. The first configuration information comprises or indicates one or more data sets of UE environment information (UEI) associated with a reference signal transmitted by the UE. The UEI data may include, without limitation, signal quality indicators, feedback configuration parameters, and physical layer measurements related to the UE’s local channel conditions. At step 420, the base station analyses one or more received data sets. The analysed information may include, but is not limited to, a rank indicator representing the number of dominant eigenmodes in the channel, one or more beam indexes identifying the preferred transmit beams, a compression ratio associated with the CSI encoding process, quantization levels used in signal representation, and other feedback data reflecting link performance or channel quality.

[0059] At step 430, the method proceeds to determine and select at least one optimum AI / ML model and a corresponding set of hyperparameters based on the analysed dataset. The model selection may be performed by evaluating the correlation, variance, and feature distribution of the received UEI data, and matching it against pre-trained candidate models. The optimum model may be selected from a pool of available AI / ML models, such as autoencoders, LSTM-based compressors, or CSINet variants, depending on the UE scenario and system objectives.

[0060] At step 440, the method includes configuring the selected model and associated hyperparameters for deployment within the network. This configuration process may involve provisioning the model to the base station for real-time CSI compression and reconstruction, as well as transmitting a corresponding model identifier or configuration parameters to the UEs. The model configuration may be performed on a per-cell, persector, or per-UE basis, and may be adapted dynamically in response to changing channel conditions or mobility patterns. This method facilitates intelligent, data-driven model selection for CSI compression, thereby enabling higher compression efficiency and reconstruction accuracy across a wide range of propagation environments. By leveraging real-time UE environment information and dynamically selecting appropriate AI / ML models, the system overcomes limitations of static, one-size-fits-all compression techniques and supports advanced features of next-generation wireless networks including massive MIMO and dense user deployments.

[0061] Figure 5 illustrates a timing diagram for CSI compression using AI / ML (500), according to one embodiment of the present invention. The system comprises two primary domains: the network side (NW side 510), which includes a Base Station (BS) and a server (e.g., GPU-based or cloudbased), and the user equipment side (UE side 520), which includes a plurality of User Equipment (UEs). The diagram represents the sequential operations involved in the acquisition, compression, transmission, and reconstruction of Channel State Information (CSI) utilizing AI / ML models for enhanced efficiency.

[0062] Initially, the BS transmits the Channel State Information Reference Signal (CSI-RS) to the UEs. Upon reception of the CSI-RS, each UE estimates the downlink channel and generates corresponding CSI. The UE feeds back the CSI to the BS, and the BS initiates the collection of UE Information (UEI), including CSI feedback. The collected data is stored in one or more entities within the network, including the Centralized Unit (CU), Distributed Unit (DU), or Radio Unit (RU), as shown at reference step (530). The base station receives, from the UE, first configuration information comprising an estimated representation of the CSI derived from the CSI reference signal, the representation being encoded using a default or initial encoding scheme; and transmits the first configuration information to a server.

[0063] The dataset collected at the BS is then forwarded to the server. The server, upon receiving the dataset, performs AI / ML model selection and hyperparameter tuning, as denoted in step (540). Based on the observed dataset properties, the server identifies a suitable AI / ML model or parameters for CSI compression and reconstruction. The server analyses one or more datasets derived from the first configuration information, the datasets comprising at least one of: a rank indicator, beam index, compression ratio, quantization level, or feedback metric; and determines, based on the analysis, an optimal artificial intelligence or machine learning (AI / ML) model and a corresponding set of hyperparameters for compressing the CSI across spatial, temporal, and frequency domains.

[0064] Upon completion of model selection and parameter determination, the server transmits the model information or selected parameters to the BS. The BS then conveys this model or parameter information to the UE. This transmission ensures that both the UE and the network are synchronized in terms of the model used for CSI compression and decompression. The base station transmits the model identifier or the selected AI / ML model and the associated hyperparameters to the UE; requests CSI reporting resources from the UE.

[0065] After configuration, the BS again transmits the CSI-RS to the UEs. Each UE, upon receiving the CSI-RS, estimates the downlink channel (550) and proceeds to encode the CSI using an Encoder and Quantizer based on the received AI / ML model or parameters, as shown in step (560). The compressed CSI is encoded using an AI / ML encoder and is quantized prior to transmission by the UE. The UE then transmits the encoded CSI back to the network. The compressed CSI comprises at least one of: a compressed channel matrix; or a compressed precoding matrix configured to optimize a signal-to-noise ratio (SNR).

[0066] The network receives the encoded CSI and applies the preconfigured decoder model to reconstruct the original CSI, as shown in step (570). The network reconstructs the CSI from the compressed CSI and evaluates at least one network performance metric using the reconstructed CSI. This evaluation may include, for example, metrics such as reconstruction error, signal-to-noise ratio (SNR), or throughput gain, and may further involve recovering one or more of: environmental condition data associated with the UE’s geographic location; or precoding vectors for use in beamforming operations.

[0067] In one embodiment, the dataset required for AI / ML model selection may be collected and maintained at the BS, and the processing server is situated within the NW domain. In an alternative configuration, the UE-side server collects and stores the CSI dataset, which is then transferred to a cloud or over-the-top (OTT) server for AI / ML model processing. Upon finalization, the server transmits the selected model identifier or configuration to the UEs. The UE receives the model identifier or selected AI / ML model and hyperparameters from the base station, performs CSI compression using the received model and parameters, and transmits the compressed CSI to the base station. Based on the received Model ID, the network utilizes the corresponding model for CSI reconstruction.

[0068] The described timing sequence enables effective end-to-end synchronization of AI / ML-based CSI compression and reconstruction across both network and UE domains. By leveraging AI / ML models that exploit correlations in spatial, frequency, and temporal domains, the system achieves higher compression efficiency and improved reconstruction fidelity, thereby enhancing downlink throughput and overall system performance.

[0069] 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 processing and compressing Channel State Information (CSI) in a wireless communication network, the method comprising: transmitting a CSI reference signal from a base station to at least one user equipment (UE); receiving, at the base station, first configuration information from the UE, the first configuration information comprising an estimated representation of the CSI derived from the CSI reference signal and encoded using a default or initial encoding scheme; forwarding the first configuration information to a server; analyzing, at the server, one or more datasets derived from the first configuration information, the datasets comprising at least one of: a rank indicator, beam index, compression ratio, quantization level, or feedback metric; determining, based on the analysis, an optimal artificial intelligence or machine learning (AI / ML) model and corresponding hyperparameters for compressing the CSI across spatial, temporal, and frequency domains; and transmitting a model identifier or the selected AI / ML model along with the hyperparameters to the base station for use within the network.

2. The method as claimed in claim 1 , further comprising: transmitting the model identifier or the selected AI / ML model and associated hyperparameters to the user equipment (UE);requesting CSI reporting resources from the UE; receiving, from the UE, second configuration information comprising CSI compressed using the selected AI / ML model; and reconstructing the CSI from the compressed information and evaluating at least one network performance metric using the reconstructed CSI.

3. The method as claimed in claim 2, wherein the compressed CSI is encoded using an AI / ML encoder and is quantized prior to transmission by the UE.

4. The method as claimed in claim 1 , wherein the base station transmits the model identifier or the selected AI / ML model and the associated hyperparameters to the UE via Downlink Control Information (DCI) or a Physical Downlink Control Channel (PDCCH).

5. The method as claimed in claim 1 , wherein the base station receives model-related information or feedback from the UE via a Physical Uplink Control Channel (PUCCH) or a Physical Uplink Shared Channel (PUSCH).

6. The method as claimed in claim 2, wherein the compressed CSI comprises at least one of: a compressed channel matrix; or a compressed precoding matrix configured to optimize a signal-to-noise ratio (SNR).

7. The method as claimed in claim 2, wherein reconstructing the CSI comprises recovering one or more of: environmental condition data associated with the UE’s geographic location; or precoding vectors for use in beamforming operations.

8. The method as claimed in claim 1 , wherein the selected AI / ML model comprises an autoencoder architecture selected from the group consisting of: dense neural networks, convolutional neural networks (CNNs), long short-term memory (LSTM) networks, and transformer-based networks.

9. The method as claimed in claim 1 , wherein the hyperparameters comprise at least one of the following: the number of layers in the model; the number of nodes per layer; the kernel size used for convolutional operations; the activation function; the mini-batch size indicating the number of channel samples processed per training iteration; the learning rate used for gradient descent optimization; the choice of optimization algorithm; and a regularization method including, but not limited to, dropout or weight regularization. The hyperparameters are tuned using one or more validation datasets to optimize the performance of the selected artificial intelligence or machine learning (AI / ML) model.

10. The method as claimed in claim 1 , wherein the selected AI / ML model is configured to exploit correlations in one or more of the spatial, temporal, and frequency domains of the CSI, thereby improving compression efficiency and increasing downlink throughput.

11. A system for processing and compressing Channel State Information(CSI) in a wireless communication network, the system comprising: a base station, configured to: transmit a CSI reference signal to at least one user equipment (UE); receive first configuration information from the UE, the information comprising an estimated representation of the CSI encoded using a default or initial scheme; and forward the information to a server; a server configured to: analyze datasets derived from the CSI, the datasets including at least one of: a rank indicator, beam index, compression ratio, quantization level, or feedback metric; and determine an optimal artificial intelligence or machine learning (AI / ML) model and corresponding hyperparameters for compressing the CSI across spatial, temporal, and frequency domains.

12. The system as claimed in claim 11 , wherein: the server is further configured to transmit a model identifier or the selected AI / ML model along with the associated hyperparameters to the base station; the base station is configured to:transmit said model identifier or model to the UE via a control channel; request CSI reporting resources; receive compressed CSI; and reconstruct the CSI and evaluate network performance metrics; the UE is configured to: receive the model and hyperparameters; compress and quantize the CSI using an AI / ML encoder; and transmit the compressed data to the base station.

13. The system as claimed in claim 11 , wherein the AI / ML model comprises one of: dense neural networks, CNNs, LSTM networks, or transformer-based networks, with hyperparameters tuned for performance based on validation datasets.

Images