Methods and apparatus for site-specific and implicit ai model refinement for 3GPP systems

Abstract

The disclosure relates to a 5G or 6G communication system for supporting a higher data transmission rate. Embodiments disclosed herein relate to methods (600) and systems (400) for refining Artificial Intelligence (AI) models in wireless systems. The methods (600) provide site-specific and implicit AI model refinement, where site-specific dataset collection is performed as a part of operational 3GPP procedure, and new models are trained or existing models are refined for CSI prediction and CSI compression using the collected site-specific dataset. The methods (600) include training of CSI prediction and CSI compression models for UEs with the ability to download new models from the base station (402) and also for UEs without the capability.

Description

METHODS AND APPARATUS FOR SITE-SPECIFIC AND IMPLICIT AI MODEL REFINEMENT FOR 3GPP SYSTEMS

[0001] Embodiments disclosed herein relate to wireless communication networks, and more particularly to refining Artificial Intelligence (AI) models in wireless systems.

[0002] 5G mobile communication technologies define broad frequency bands such that high transmission rates and new services are possible, and can be implemented not only in “Sub 6GHz” bands such as 3.5GHz, but also in “Above 6GHz” bands referred to as mmWave including 28GHz and 39GHz. In addition, it has been considered to implement 6G mobile communication technologies (referred to as Beyond 5G systems) in terahertz (THz) bands (for example, 95GHz to 3THz bands) in order to accomplish transmission rates fifty times faster than 5G mobile communication technologies and ultra-low latencies one-tenth of 5G mobile communication technologies.

[0003] At the beginning of the development of 5G mobile communication technologies, in order to support services and to satisfy performance requirements in connection with enhanced Mobile BroadBand (eMBB), Ultra Reliable Low Latency Communications (URLLC), and massive Machine-Type Communications (mMTC), there has been ongoing standardization regarding beamforming and massive MIMO for mitigating radio-wave path loss and increasing radio-wave transmission distances in mmWave, supporting numerologies (for example, operating multiple subcarrier spacings) for efficiently utilizing mmWave resources and dynamic operation of slot formats, initial access technologies for supporting multi-beam transmission and broadbands, definition and operation of BWP (BandWidth Part), new channel coding methods such as a LDPC (Low Density Parity Check) code for large amount of data transmission and a polar code for highly reliable transmission of control information, L2 pre-processing, and network slicing for providing a dedicated network specialized to a specific service.

[0004] Currently, there are ongoing discussions regarding improvement and performance enhancement of initial 5G mobile communication technologies in view of services to be supported by 5G mobile communication technologies, and there has been physical layer standardization regarding technologies such as V2X (Vehicle-to-everything) for aiding driving determination by autonomous vehicles based on information regarding positions and states of vehicles transmitted by the vehicles and for enhancing user convenience, NR-U (New Radio Unlicensed) aimed at system operations conforming to various regulation-related requirements in unlicensed bands, NR UE Power Saving, Non-Terrestrial Network (NTN) which is UE-satellite direct communication for providing coverage in an area in which communication with terrestrial networks is unavailable, and positioning.

[0005] Moreover, there has been ongoing standardization in air interface architecture / protocol regarding technologies such as Industrial Internet of Things (IIoT) for supporting new services through interworking and convergence with other industries, IAB (Integrated Access and Backhaul) for providing a node for network service area expansion by supporting a wireless backhaul link and an access link in an integrated manner, mobility enhancement including conditional handover and DAPS (Dual Active Protocol Stack) handover, and two-step random access for simplifying random access procedures (2-step RACH for NR). There also has been ongoing standardization in system architecture / service regarding a 5G baseline architecture (for example, service based architecture or service based interface) for combining Network Functions Virtualization (NFV) and Software-Defined Networking (SDN) technologies, and Mobile Edge Computing (MEC) for receiving services based on UE positions.

[0006] As 5G mobile communication systems are commercialized, connected devices that have been exponentially increasing will be connected to communication networks, and it is accordingly expected that enhanced functions and performances of 5G mobile communication systems and integrated operations of connected devices will be necessary. To this end, new research is scheduled in connection with eXtended Reality (XR) for efficiently supporting AR (Augmented Reality), VR (Virtual Reality), MR (Mixed Reality) and the like, 5G performance improvement and complexity reduction by utilizing Artificial Intelligence (AI) and Machine Learning (ML), AI service support, metaverse service support, and drone communication.

[0007] Furthermore, such development of 5G mobile communication systems will serve as a basis for developing not only new waveforms for providing coverage in terahertz bands of 6G mobile communication technologies, multi-antenna transmission technologies such as Full Dimensional MIMO (FD-MIMO), array antennas and large-scale antennas, metamaterial-based lenses and antennas for improving coverage of terahertz band signals, high-dimensional space multiplexing technology using OAM (Orbital Angular Momentum), and RIS (Reconfigurable Intelligent Surface), but also full-duplex technology for increasing frequency efficiency of 6G mobile communication technologies and improving system networks, AI-based communication technology for implementing system optimization by utilizing satellites and AI (Artificial Intelligence) from the design stage and internalizing end-to-end AI support functions, and next-generation distributed computing technology for implementing services at levels of complexity exceeding the limit of UE operation capability by utilizing ultra-high-performance communication and computing resources.

[0008] This disclosure relates to wireless communication networks, and more particularly to a terminal and a communication method thereof in a wireless communication system.

[0009] In accordance with an aspect of the disclosure, the embodiments herein provide a method for managing an Artificial Intelligence (AI) model in a wireless system. The method comprises receiving, by a base station, at least one User Equipment (UE) capability information from at least one UE. The method comprises downloading, by the base station, one or more AI models based on the UE capability information. The method comprises selecting, by the base station, at least one AI model from the downloaded AI models, and a Channel State Information (CSI) configuration for the UE. Thereafter, the method comprises training, by the base station, the selected AI model for performing at least one CSI operation, based on the selected CSI configuration.

[0010] Aspects of the disclosure are to address at least the above-mentioned problems and / or disadvantages and to provide at least the advantages described below. Accordingly, an aspect of the disclosure is to provide efficient communication methods in a wireless communication system.

[0011] Embodiments herein are illustrated in the accompanying drawings, throughout which like reference letters indicate corresponding parts in the various figures. The embodiments herein will be better understood from the following description with reference to the following illustratory drawings. Embodiments herein are illustrated by way of examples in the accompanying drawings, and in which:

[0012] FIGs. 1A and 1B depict an Artificial Intelligence (AI) based CSI prediction, according to existing arts;

[0013] FIGs. 2A and 2B depict an AI based CSI compression, according to existing arts;

[0014] FIG. 3A shows a typical deployment cycle for non-AI methods or with AI based generic models, according to existing arts;

[0015] FIG. 3B shows a typical deployment cycle for AI based approaches, according to existing arts;

[0016] FIG. 4 depicts a system for managing an AI model in a wireless system, according to embodiments as disclosed herein;

[0017] FIG. 5 depicts a plurality of modules of base station of the proposed system, according to embodiments as disclosed herein;

[0018] FIG. 6 depicts a method for managing an AI model in a wireless system, according to embodiments as disclosed herein;

[0019] FIG. 7 depicts a message sequence diagram of performing complexity consideration based training of AI models per CSI configuration using the site-specific dataset, according to embodiments as disclosed herein;

[0020] FIG. 8 depicts a flow process for enhancing the performance of CSI compression by training new models, according to embodiments as disclosed herein;

[0021] FIG. 9 depicts a flow process for enhancing the performance of CSI compression by training decoders of shortlisted new AI models, according to embodiments as disclosed herein;

[0022] FIG. 10 depicts a flow process for enhancing the performance of CSI prediction by training new AI models, according to embodiments as disclosed herein;

[0023] FIG. 11 depicts a message sequence diagram for performing a site-specific dataset collection, according to embodiments as disclosed herein;

[0024] FIG. 12 depicts a flow process for performing dataset collection in line with implicit model refinement, according to embodiments as disclosed herein;

[0025] FIG. 13 depicts a flow process of a deployment process for AI based approaches, according to embodiments as disclosed herein;

[0026] FIG. 14 depicts a flow process for refining CSI compression models for UEs with download AI model support, according to embodiments as disclosed herein;

[0027] FIG. 15 depicts a flow process for refining CSI compression models for UEs without download AI model support, according to embodiments as disclosed herein;

[0028] FIG. 16 depicts a flow process for refining CSI compression models for UEs with download AI model support by training new AI models, according to embodiments as disclosed herein;

[0029] FIG. 17 depicts a flow process for refining CSI compression models for UEs without download AI model support, according to embodiments as disclosed herein;

[0030] FIG. 18 depicts a flow process for improving the performance of CSI compression for a UE with model transfer capability by training generic models, according to embodiments as disclosed herein;

[0031] FIG. 19 depicts a flow process for improving the performance of CSI compression for a UE without model transfer capability by training generic models, according to embodiments as disclosed herein;

[0032] FIG. 20 depicts a flow process of improving the performance of CSI prediction by training the predictor, according to embodiments as disclosed herein;

[0033] FIG. 21 illustrates a structure of a base station according to an embodiment of the disclosure; and

[0034] FIG. 22 illustrates a structure of a UE according to an embodiment of the disclosure.

[0035] Accordingly, the embodiments herein provide a method for managing an Artificial Intelligence (AI) model in a wireless system. The method comprises receiving, by a base station, at least one User Equipment (UE) capability information from at least one UE. The method comprises downloading, by the base station, one or more AI models based on the UE capability information. The method comprises selecting, by the base station, at least one AI model from the downloaded AI models, and a Channel State Information (CSI) configuration for the UE. Thereafter, the method comprises training, by the base station, the selected AI model for performing at least one CSI operation, based on the selected CSI configuration.

[0036] Accordingly, the embodiments herein provide a base station. The base station comprises a processor and a memory module. The processor is coupled with the memory module. The processor is configured to receive at least one UE capability information from at least one UE, download one or more AI models based on the UE capability information, select at least one AI model from the downloaded AI models and a CSI configuration for the UE, and train the selected AI model for performing at least one CSI operation, based on the selected CSI configuration.

[0037] These and other aspects of the example embodiments herein will be better appreciated and understood when considered in conjunction with the following description and the accompanying drawings. It should be understood, however, that the following descriptions, while indicating example embodiments and numerous specific details thereof, are given by way of illustration and not of limitation. Many changes and modifications may be made within the scope of the example embodiments herein without departing from the spirit thereof, and the example embodiments herein include all such modifications.

[0038] The following description with reference to the accompanying drawings is provided to assist in a comprehensive understanding of various embodiments of the disclosure 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 various embodiments described herein can be made without departing from the scope and spirit of the disclosure. In addition, descriptions of well-known functions and constructions may be omitted for clarity and conciseness.

[0039] The terms and words used in the following description and claims are not limited to their bibliographical meanings, but, are merely used by the inventor to enable a clear and consistent understanding of the disclosure. Accordingly, it should be apparent to those skilled in the art that the following description of various embodiments of the disclosure is provided for illustration purpose only and not for the purpose of limiting the disclosure as defined by the appended claims and their equivalents.

[0040] 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.

[0041] Before undertaking the DETAILED DESCRIPTION below, it can be advantageous to set forth definitions of certain words and phrases used throughout this patent document. The term “couple” and its derivatives refer to any direct or indirect communication between two or more elements, whether or not those elements are in physical contact with one another. The terms “transmit,” “receive,” and “communicate,” as well as derivatives thereof, encompass both direct and indirect communication. The terms “include” and “comprise,” as well as derivatives thereof, mean inclusion without limitation. The term “or” is inclusive, meaning and / or. The phrase “associated with,” as well as derivatives thereof, means to include, be included within, connect to, interconnect with, contain, be contained within, connect to or with, couple to or with, be communicable with, cooperate with, interleave, juxtapose, be proximate to, be bound to or with, have, have a property of, have a relationship to or with, or the like. The term “controller” means any device, system or part thereof that controls at least one operation. Such a controller can be implemented in hardware or a combination of hardware and software and / or firmware. The functionality associated with any particular controller can be centralized or distributed, whether locally or remotely. The phrase “at least one of,” when used with a list of items, means that different combinations of one or more of the listed items can be used, and only one item in the list can be needed. For example, “at least one of: A, B, and C” includes any of the following combinations: A, B, C, A and B, A and C, B and C, and A and B and C. For example, “at least one of: A, B, or C” includes any of the following combinations: A, B, C, A and B, A and C, B and C, and A, B and C.

[0042] Moreover, various functions described below can be implemented or supported by one or more computer programs, each of which is formed from computer-readable program code and embodied in a computer-readable medium. The terms “application” and “program” refer to one or more computer programs, software components, sets of instructions, procedures, functions, objects, classes, instances, related data, or a portion thereof adapted for implementation in a suitable computer-readable program code. The phrase “computer-readable program code” includes any type of computer code, including source code, object code, and executable code. The phrase “computer-readable medium” includes any type of medium capable of being accessed by a computer, such as Read-Only Memory (ROM), Random Access Memory (RAM), a hard disk drive, a Compact Disc (CD), a Digital Video Disc (DVD), or any other type of memory. A “non-transitory” computer-readable medium excludes wired, wireless, optical, or other communication links that transport transitory electrical or other signals. A non-transitory computer-readable medium includes media where data can be permanently stored and media where data can be stored and later overwritten, such as a rewritable optical disc or an erasable memory device.

[0043] Terms used herein to describe the embodiments of the disclosure are not intended to limit and / or define the scope of the disclosure. For example, unless otherwise defined, the technical terms or scientific terms used in the disclosure shall have the ordinary meaning understood by those with ordinary skills in the art to which the disclosure belongs.

[0044] It should be understood that “first”, “second” and similar words used in the disclosure do not express any order, quantity or importance, but are only used to distinguish different components.

[0045] As used herein, any reference to “an example” or “example”, “an implementation” or “implementation”, “an embodiment” or “embodiment” means that particular elements, features, structures or characteristics described in connection with the embodiment is included in at least one embodiment. The phrases “in one embodiment” or “in one example” appearing in different places in the specification do not necessarily refer to the same embodiment.

[0046] As used herein, “a portion of” something means “at least some of” the thing, and as such may mean less than all of, or all of, the thing. As such, “a portion of” a thing includes the entire thing as a special case, i.e., the entire thing is an example of a portion of the thing.

[0047] As used herein, the term “set” means one or more. Accordingly, a set of items can be a single item or a collection of two or more items.

[0048] In this disclosure, to determine whether a specific condition is satisfied or fulfilled, expressions, such as “greater than” or “less than” are used by way of example and expressions, such as “greater than or equal to” or “less than or equal to” are also applicable and not excluded. For example, a condition defined with “greater than or equal to” may be replaced by “greater than” (or vice-versa), a condition defined with “less than or equal to” may be replaced by “less than” (or vice-versa), etc.

[0049] It will be further understood that similar words such as the term “include” or “comprise” mean that elements or objects appearing before the word encompass the listed elements or objects appearing after the word and their equivalents, but other elements or objects are not excluded. Similar words such as “connect” or “connected” are not limited to physical or mechanical connection, but can include electrical connection, whether direct or indirect. “Upper”, “lower”, “left” and “right” are only used to express a relative positional relationship, and when an absolute position of the described object changes, the relative positional relationship may change accordingly.

[0050] The principal object of the embodiments herein is to disclose methods and systems for site-specific and implicit AI model refinement for wireless systems.

[0051] Another object of the embodiments herein is to disclose methods and systems for site-specific AI model refinement for 3GPP systems, wherein dataset collection is performed as a part of operational 3GPP procedure, and new models are trained or existing models are refined for CSI prediction and CSI compression using the collected site-specific dataset.

[0052] Another object of the embodiments herein is to disclose separate methods to train CSI prediction and CSI compression models for UEs with the ability to download new models from the base station and also for UEs without the capability.

[0053] Another object of the embodiments herein is to disclose methods and systems for UE capability based model transfer and execution, for AI operations.

[0054] Introduction of Artificial Intelligence (AI) in wireless systems is expected to improve system capacity and efficiency in Third Generation Partnership Project (3GPP) systems. Due to the proven gains of AI methods, 3GPP has taken up a study item as part of Rel-18 on introduction of AI based physical layer procedures. Currently, for Channel State Information (CSI) in 3GPP AI study, the following use cases are in focus:

[0055] - AI based CSI prediction (as depicted in FIGs. 1A and 1B); and

[0056] - AI based CSI compression (as depicted in FIGs. 2A and 2B).

[0057] Broadly, the modes of operation for AI can be categorized as:

[0058] 1) One-sided operation: An AI model is independently deployed either at Base Station (BS) or at user equipment (UE); for example, CSI prediction.

[0059] 2) Two-sided operation: A pair of AI models, one deployed at the BS and its counterpart deployed at the UE interactively execute an operation; for example, CSI compression.

[0060] AI deployment can be executed using a downloaded model at the UE-side (as depicted in FIG. 1B). In this scenario, the AI model (full or partial) is downloaded from the BS to the UE. In addition to the AI models, the UE may already have stored AI models in its memory. Using CSI compression, in this mode of operation, after attaching, the UE indicates capability to execute AI based operation with support for new model downloads. The UE receives one or more AI models to perform CSI encoding from the BS. Further, the BS indicates to the UE, the AI model to be used for CSI reporting throughput AI model indicator (AMI). The UE deploys the selected AI model to perform CSI compression. The selected AI model based CSI reports are shared with the BS, until deactivation. In this scenario, the AI model (full or partial) is available with the UE, but downloads from the BS is not supported. Hence no downloading from the network entity is involved.

[0061] AI deployment can be executed using a downloaded model at the UE-side (as depicted in FIG. 2B). In this scenario, the AI model (full or partial) is downloaded from the BS to the UE. In addition to the AI models the UE may already have stored AI models in its memory. In this scenario, the AI model (full or partial) is available with the UE, but download from the BS is not supported. Hence no downloading from the network entity is involved. After attachment, the UE indicates capability to execute AI based operation with support stored AI model selection, along with stored AI model info. The BS selects an AI model based on UE measurements, and shares indication to select the AI model in UE memory. The UE selects the BS indicated model in its memory and deploys it to perform CSI encoding. The AI model based CSI reports are shared with the BS, until deactivation.

[0062] A generic model is pre-trained using a common dataset, hence it is transparent to the site-specific behaviour. The generic model provides a good common ground, or a starting point for AI based deployments. Either at the UE memory or at the BS-side, generic models will be deployed at the initial part of AI based network deployment. Also, training generic models using a common and large dataset helps capture the problem-specific features during AI training. FIG. 3A shows a typical deployment cycle for non-AI methods or with AI based generic models. However, models that are trained with generic data may not be able to reap the full benefits of AI due to their lack of tailoring the performance specific to the deployment scenario.

[0063] AI model customization by deploying specialized models results in significant gains compared to usage of a general model. Hence, after AI deployment in a site, the network and UE may continue to fine tune the AI models in a site-specific manner to improve their performance (as depicted in FIG. 3B). FIG. 3B shows a typical deployment cycle for AI based approaches. Thus, there is a need for modem refinement of AI models after their deployment. Hence, training and dataset collection procedures are critical to support this refinement procedure.

[0064] Dataset collection using dedicated procedures pose additional resource and computational overhead at the BS. Further, performance of the trained models is correlated to the size of the dataset. Hence, efficient collection of data, at scale, is challenging.

[0065] There are challenges in training AI models and transferring them to the UE. Although, some UEs are expected to support download of models from the BS, the capability may not be supported by all the UEs in the network. Thus, such UEs that lack the AI model download support may not benefit from site-specific customization.

[0066] There are challenges in managing the complexity of AI models for inference at the UEs. Depending on the UE capability to execute AI models, the UEs attached to a BS may not be able to execute AI models of higher complexity. Hence, site-specific optimizations may have to account for UE complexity considerations. Also, AI model complexity is typically correlated to its performance.

[0067] Hence, there is a need in the art for solutions which will overcome the above mentioned drawback(s), among others.

[0068] The embodiments herein achieve methods and systems that are essential for usage of AI based operations in a deployment scenario in wireless systems. Referring now to the drawings, and more particularly to FIGS. 4 through 20, where similar reference characters denote corresponding features consistently throughout the figures, there are shown embodiments.

[0069] FIG. 4 depicts a system 400 for managing an Artificial Intelligence (AI) model in a wireless system. The system 400 comprises a base station 402, and a User Equipment (UE) 404. However, the components of the UE 404 are not limited thereto. For example, the UE 404 may include more or fewer components than those described above. In addition, the UE 404 corresponds to the UE of the FIG. 22.The base station 402 further comprises a processor 406, a communication module 408, and a memory module 410. However, the components of the base station 402 are not limited thereto. For example, the base station 402 may include more or fewer components than those described above. In addition, the base station 402 corresponds to the base station of the FIG. 21.

[0070] In an embodiment herein, the processor 406 can perform UE capability based model transfer and execution, for AI operations. The processor 406 can perform efficient site-specific dataset collection that incurs minimal resource overhead. The processor 406 can further provide customized AI models to UEs that lack the capability to dynamically deploy newer models. The AI models can be customized by training the model blocks that are executed in the base station 402. The processor 406 further comprises a model download module 502, a model selection module 504, a training module 506, and a CSI decoder module 508, as depicted in FIG. 5.

[0071] In an embodiment herein, the model download module 502 can receive at least one UE capability information from at least one UE 404. The UE capability information can include, but not limited to a computational complexity capability for each AI based functionality, and capability to download and deploy new AI models from the base station 402 during the at least one CSI operation. The AI based functionality can include, but not limited to an AI compression, an AI prediction, an AI model download, and an AI model selection. The model download module 502 can download one or more AI models based on the received UE capability information.

[0072] In an embodiment herein, the model selection module 504 can select at least one AI model from the downloaded one or more AI models. In addition, the model selection module 504 can select a Channel State Information (CSI) configuration for the at least one UE 404. The AI model, and the CSI configuration for the UE 404 are selected based on at least one of the UE capability information, a site-specific dataset collection, one or more scheduling requirements, and one or more AI model training requirements. The scheduling requirements comprise one or more reporting periodicities to the UE 404 for an AI based CSI prediction case. The AI model training requirements comprises at least one feedback bit configuration for the UE 404 for a CSI reporting in an AI compression case. In an embodiment herein, the model selection module 504 can select at least one new CSI configuration for refinement. In an embodiment herein, the model selection module 504 can select at least one existing CSI configuration for refinement. The model selection module 504 selects the existing CSI configuration, based on higher accuracy of CSI reporting for the site-specific dataset creation that meets one or more dataset quality requirements. The dataset quality requirements can be site specific in order to ensure that the dataset quality requirements meet minimum accuracy for unhindered transmission of data.

[0073] In an embodiment herein, the model selection module 504 can indicate the UE 404 to select one or more encoder models stored in the UE memory, if the UE 404 lacks the capability to download the new AI model. The model selection module 504 can select one or more new decoder models from an existing encoder-decoder model pair, based on the selected one or more encoder models. The model selection module 504 selects a high complexity class decoder, when a higher accuracy CSI reporting is desired. The model selection module 504 selects a default complexity class decoder when a desired accuracy of CSI reporting is moderate.

[0074] In an embodiment herein, the training module 506 can train the selected AI model for performing at least one CSI operation, based on the selected CSI configuration. In an embodiment herein, the training module 506 can train the selected AI model of a high complexity using the site-specific dataset collection. The training module 506 can further transfer the trained AI model to the UE 404 for execution, subject to the capability of the UE 404 to download and deploy one or more new AI models.

[0075] In an embodiment herein, the training module 506 can train a part of the selected AI model that is executed on a network side using the site-specific dataset collection. The training module 506 can further deploy the trained part of the AI model that is executed on the network side during a AI operation, while leaving the AI model at a UE side untouched. In an embodiment herein, the training module 506 can train at least one new AI model using the site-specific dataset collection, for the new CSI configuration. In an embodiment herein, the training module 506 can train the new AI model using the site-specific dataset collection, for the existing CSI configuration.

[0076] In an embodiment herein, the training module 506 can identify a set of AI model configurations and respective complexity classes for training one or more new AI models, based on a UE distribution and throughput requirements. The set of AI model configurations per complexity class is a list of complexity classes and payload sizes of the new AI models that are to be trained. The complexity class of the new AI models defines a maximum AI model complexity supported by the UE 404. The complexity class is defined for at least one of a downloaded AI model from the base station 402, and a stored AI model present in a UE memory. The complexity class of the new AI models is mapped based on at least one of a complexity in Floating Point Operations (FLOPs), an inference latency, and a combination of the complexity and the inference latency.

[0077] In an embodiment herein, the training module 506 can verify, if the UE 404 has the capability to download a new AI model, based on the received UE capability information. The training module 506 can train an encoder and decoder pair of the selected AI model for shortlisted complexity and payload sizes, for the UE 404, if the UE 404 has the capability to download a new AI model. The training module 506 can re-train the encoder and decoder pair of an existing AI model using the site-specific dataset, if the selected CSI configuration is an existing CSI compression, and the UE 404 has the capability to download a new AI model. The training module 506 can re-train, the existing AI model and transfer the re-trained existing AI model to the UE 404, if the selected CSI configuration is an existing UE-side CSI prediction, and the UE 404 has the capability to download a new AI model. The training module 506 can train a new encoder and decoder pair of the new AI model, if the selected CSI configuration is a new CSI compression, and the UE 404 has the capability to download a new AI model. The training module 506 can train new CSI predictors, if the selected CSI configuration is a new UE-side CSI prediction, and the UE 404 has the capability to download a new AI model.

[0078] In an embodiment herein, the training module 506 can train one or more new decoder models for the respective one or more encoder models stored in the UE memory, where the new decoder models are selected by the model selection module 504.

[0079] In an embodiment herein, the training module 506 can re-train the new decoder models of the existing AI model using the site-specific dataset, if the selected CSI configuration is the existing CSI compression, and the UE lacks the capability to download a new AI model. The training module 506 can train the new AI model, if the selected CSI configuration is a Base Station (BS)-side CSI prediction, and the UE lacks the capability to download a new AI model. The training module 506 can train the new decoder models using the site-specific dataset for at least one of a given CSI configuration, and AI encoder, if the selected CSI configuration is the existing CSI compression, and the UE 404 lacks the capability to download a new AI model. The training module 506 can train a new CSI prediction model to perform prediction corresponding to a new CSI application time instance using the site-specific dataset, if the selected CSI configuration is the BS-side CSI prediction, and the UE 404 lacks the capability to download a new AI model.

[0080] In an embodiment herein, the CSI decoder module 508 can receive a CSI report from the UE 404, on performing at least one CSI operation. The CSI decoder module 508 can decode the CSI report for scheduling. The CSI decoder module 508 can pre-process the decoded CSI report to extract one or more features required for a dataset. The CSI decoder module 508 can further store the pre-processed data in the dataset for future implicit training. The CSI decoder module 508 can repeat steps from decoding of the CSI report to execute the site-specific dataset collection until one or more dataset quality requirements are met.

[0081] In an embodiment herein, the processor 406 can process and execute data of a plurality of modules of the base station 402. The processor 406 can be configured to execute instructions stored in the memory module 410. The processor 406 may comprise one or more of microprocessors, circuits, and other hardware configured for processing. The processor 406 can be at least one of a single processer, a plurality of processors, multiple homogeneous or heterogeneous cores, multiple Central Processing Units (CPUs) of different kinds, microcontrollers, special media, and other accelerators. The processor 406 may be an application processor (AP), a graphics-only processing unit (such as a graphics processing unit (GPU), a visual processing unit (VPU)), and / or an Artificial Intelligence (AI)-dedicated processor (such as a neural processing unit (NPU)).

[0082] In an embodiment herein, the plurality of modules of the processor 406 of the base station 402 can communicate via the communication module 408. The communication module 408 may be in the form of either a wired network or a wireless communication network module. The wireless communication network may comprise, but not limited to, Global Positioning System (GPS), Global System for Mobile Communications (GSM), Wi-Fi, Bluetooth low energy, Near-field communication (NFC), and so on. The wireless communication may further comprise one or more of Bluetooth, ZigBee, a short-range wireless communication (such as Ultra-Wideband (UWB)), and a medium-range wireless communication (such as Wi-Fi) or a long-range wireless communication (such as 3G / 4G / 5G / 6G and non-3GPP technologies or WiMAX), according to the usage environment.

[0083] In an embodiment herein, the memory module 410 may comprise one or more volatile and non-volatile memory components which are capable of storing data and instructions of the modules of the base station 402 to be executed. Examples of the memory module 410 can be, but not limited to, NAND, embedded Multi Media Card (eMMC), Secure Digital (SD) cards, Universal Serial Bus (USB), Serial Advanced Technology Attachment (SATA), solid-state drive (SSD), and so on. The memory module 410 may also include one or more computer-readable storage media. Examples of non-volatile storage elements may include magnetic hard discs, optical discs, floppy discs, flash memories, or forms of electrically programmable memories (EPROM) or electrically erasable and programmable (EEPROM) memories. In addition, the memory module 410 may, in some examples, be considered a non-transitory storage medium. The term “non-transitory” may indicate that the storage medium is not embodied in a carrier wave or a propagated signal. However, the term “non-transitory” should not be interpreted to mean that the memory module 410 is non-movable. In certain examples, a non-transitory storage medium may store data that can, over time, change (for example, in Random Access Memory (RAM) or cache).

[0084] FIG. 4 shows example modules of the base station 402, but it is to be understood that other embodiments are not limited thereon. In other embodiments, the base station 402 may include less or more number of modules. Further, the labels or names of the modules are used only for illustrative purpose and does not limit the scope of the invention. One or more modules can be combined together to perform same or substantially similar function in the base station 402.

[0085] FIG. 6 depicts a method 600 for managing an AI model in a wireless system. The method 600 comprises receiving, by the base station 402, at least one AI based UE capability information from at least one UE 404, as depicted in step 602. The method 600 comprises downloading, by the base station 402, one or more AI models based on the UE capability information, as depicted in step 604. The method 600 comprises selecting, by the base station 402, at least one AI model from the downloaded one or more AI models, and a CSI configuration for the UE 404, as depicted in step 606. Thereafter, the method 600 comprises training, by the base station 402, the selected AI model for performing at least one CSI operation, based on the selected CSI configuration, as depicted in step 608.

[0086] The various actions in method 600 may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some actions listed in FIG. 6 may be omitted.

[0087] FIG. 7 depicts a message sequence diagram of performing complexity consideration based training of AI models per CSI configuration using the site-specific dataset. Per CSI configuration, more than one AI models are trained using site-specific dataset. Based on the UE capability info, one or more appropriate AI models are transferred at the time of CSI operation.

[0088] For example, for a two-sided CSI compression operation, as shown in FIG. 7, once a Radio Resource Control (RRC) configuration request is sent from the base station 402 to the UE 404 as depicted in step 7-2, the UE 404 completes the RRC configuration as depicted in step 7-4. Later, the base station 402 sends a UE capability enquiry to the UE 404, as depicted in step 7-6. The UE 404 responds to the base station 402 with a UE capability response, as depicted in step 7-8, where the UE capability response includes a computational complexity capability for each AI based functionality that the UE 404 is capable of downloading and deploying new AI models from the base station 402 during at least one CSI operation.

[0089] In an embodiment herein, the training configuration is selected by the base station 402. AI model inference complexity has a strong correlation to inference performance. Hence, this implies that the base station 402 should be signalled in advance, by the UE 404, its AI related capabilities and complexity information. For CSI compression and prediction, embodiments herein define the following capability elements:

[0090] AI Compression: UE support for AI based CSI Compression

[0091] AI Prediction: UE support for AI based CSI prediction

[0092] AI Model Download: UE support for AI model download from the base station 402.

[0093] AI Model Selection: UE support for selection of at least one model, from the AI models stored in UE memory, by the base station 402.

[0094] AI Model Complexity Class: Complexity class of the UE 404 for AI operation execution. This suggests the maximum AI model complexity supported by the UE 404 of AI model that can be:

[0095] - downloaded by the UE 404 from the base station 402; or

[0096] - is already present in the UE memory

[0097] An AI model is mapped to a complexity class based on:

[0098] - Complexity in FLOPs (floating point operations);

[0099] - Inference latency (Time required for inference); and

[0100] - Combination of complexity and latency.

[0101] As depicted in FIG. 7, the AI compression is indicated as supported, the AI prediction is indicated as supported, the AI model download is indicated as supported, the AI model selection is indicated as supported, and the AI model complexity class is indicated as X. A separate AI model which is pre-trained per UE complexity class to perform the same operation is defined as a AI model complexity class. Based on the UE computational complexity capability, the base station 402 selects the AI model encoder and transfers the same to the UE 404 for CSI compression, as depicted in steps 7-10 to 7-16.

[0102] The AI model download indication is sent from the base station 402 to the UE 404, as depicted in step 7-12. Further, the CSI configuration, and the AI model selection are sent from the base station 402 to the UE 404, as depicted in step 7-14. A Channel State Information Reference Signal (CSI-RS) is also sent from the base station 402 to the UE 404, as depicted in step 7-16. Later, a CSI report is sent from the UE 404 to the base station 402, as depicted in step 7-18. The base station 402 may train high complexity AI model counterparts of the AI models present at UE 404 to improve the performance. For two-sided models, balance or trade-off of the distribution of complexity s performed between the UE 404 and the base station 402, wherever possible, to optimize performance.

[0103] Embodiments herein disclose a method for selecting a AI model configuration for new model training by considering complexity consideration. For training of new models, the base station 402, first identifies the set of AI model configurations and respective complexity classes for which the new models are to be trained. The configuration is defined based on the CSI accuracy metric that should be met at the end of CSI compression or CSI prediction operation.

[0104] Embodiments herein define AI model configuration and its identification for new AI model training. For CSI compression, the requirements are dependent on:

[0105] - Payload size p∈P: Higher payload size improves reporting accuracy, but also increases the overhead

[0106] - Complexity class c∈C: Higher complexity class improves the reporting accuracy, but at the cost of higher UE computational complexity, per CSI report.

[0107] - Based on the UE distribution and throughput requirements of the cell, the BS identifies the set of tuples (c,p), that is list of complexity classes and the payload sizes, per complexity class, for which the models are to be trained.

[0108] For CSI prediction, the requirements are dependent on Complexity class c∈C, similar to CSI compression, and future CSI application time instance m∈M. For example, m=5ms implies that the CSI should be predicted 5ms into the future. Similar to CSI compression, the base station 402 identifies the set of configurations (c,m) that are to be trained for the cell, based on the cell UE distribution and throughput requirements.

[0109] FIG. 8 depicts a flow process 800 for enhancing the performance of CSI compression by training new models. For a UE with model transfer capability, a new enc-dec pair is trained. This is applicable to UEs that support download of AI encoder modules for CSI compression. Thus, model training involves training new encoders and decoders for shortlisted complexity and payload classes (c,p). First, the base station 402 determines the class of AI models based on complexity and payload sizes to meet the CSI reporting accuracy demands. Then, a separate model is trained for each of the shortlisted configuration using the dataset collection, where the dataset is collected using implicit dataset collection procedures. During CSI compression operation, the base station 402 uses one of the newly trained compression models with improved performance. Prior to its usage, the corresponding encoder model is downloaded by respective UEs to be used for CSI encoding.

[0110] As depicted, the CSI compression configuration set (i.e., set of AI model configurations) (c, p) is computed for new model training, as depicted in step 802. The base station 402 checks if the training is performed for all configurations, as depicted in step 804. If the training is performed, then the base station 402 uses the model encoder during UE model transfer for CSI compression, as depicted in step 818. A model configuration is selected for training, based on the computed set of AI model configurations, if the training is not performed, as depicted in step 806. Later, encoder-decoder pair of the AI model is trained and both encoder and decoder parameters are updated, as depicted in step 808.

[0111] Thereafter, the base station 402 checks if the enhancement criteria is met, as depicted in step 810. If the enhancement criteria is not met, then the base station 402 continues training, as depicted in step 812. If the enhancement criteria is met, then the base station 402 confirms training is done, as depicted in step 814. The base station 402 stores new encoder-decoder model pair, as depicted in step 816. The base station 402 checks if the training is done, as depicted in step 804, after storing new encoder-decoder model pair. The base station 402 uses the model encoder during UE model transfer for CSI compression, as depicted in step 818. Thereafter, the decoder is used at the base station 402 with updated parameters for CSI compression operation, as depicted in step 820.

[0112] The various actions in flow process 800 may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some actions listed in FIG. 8 may be omitted.

[0113] FIG. 9 depicts a flow process 900 for enhancing the performance of CSI compression by training decoders of shortlisted new AI models. For a UE 404 without model transfer capability, new decoder blocks can be trained. This new model training is applicable to machines that lack the capability to download new models from the base station 402. Thus, the new model training involves:

[0114] - Identifying existing encoder-decoder model pair for which new decoder blocks are to be trained.

[0115] - Note that, respective encoder models are assumed to be present in the UE memory, hence cannot be trained

[0116] - Selecting set complexity and payload set (c,p) for which new decoders are to be trained.

[0117] In this case, the base station 402 only trains the decoders of the shortlisted AI models. During CSI compression operation, the base station 402 indicates the UE 404 to select the encoder. The base station 402 further selects one of the decoders based on the complexity class and CSI reporting accuracy requirement. For example, for a given payload, when higher accuracy CSI reporting is desired, the base station 402 may select a high complexity class decoder. When desired accuracy of CSI reporting is moderate, the base station 402 may select a default complexity class decoder.

[0118] As depicted, the CSI compression configuration set (i.e., set of AI model configurations) (c, p) is computed for new model training, as depicted in step 902. The base station 402 checks if the training is performed for all configurations, as depicted in step 904. If the training is performed, then the base station 402 selects correct decoder based on the complexity class and CSI reporting requirement during CSI compression operation, as depicted in step 922. An encoder model configuration is selected for new decoder training, based on the computed set of AI model configurations, if the training is not performed, as depicted in step 906.

[0119] The base station 402 checks if the training is done for all decoder complexity class of the encoder, as depicted in step 908. The base station 402 selects the next decoder architecture for training, as depicted in step 910. New decoder design is trained for the model, and both encoder and decoder parameters are updated, as depicted in step 912. Thereafter, the base station 402 checks if the enhancement criteria is met, as depicted in step 914. If the enhancement criteria is not met, then the base station 402 continues training, as depicted in step 916. If the enhancement criteria is met, then the base station 402 confirms training is done, as depicted in step 918. The base station 402 stores new decoder model pair, as depicted in step 920. The base station 402 checks if the training is done, as depicted in step 904, after storing new decoder model pair. The base station 402 selects the correct decoder based on the complexity class and CSI reporting requirement during CSI compression operation, as depicted in step 922, after storing the new decoder model pair.

[0120] The various actions in flow process 900 may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some actions listed in FIG. 9 may be omitted.

[0121] FIG. 10 depicts a flow process 1000 for enhancing the performance of CSI prediction by training new AI models. For a UE 404 with model transfer capability, a new predictor is trained. In this mode, initially the base station 402 chooses the set of CSI prediction model configurations to be trained (c, m), where c denotes the set of complexity class, and m denotes the set of future CSI application time instances. The base station 402 trains new CSI predictor models corresponding to the shortlisted complexity and payload sizes. After completion, the newly trained predictors are transferred to the UE 404 to perform UE-side CSI prediction.

[0122] Embodiments herein can refine the prediction without UE model download support. As CSI prediction is a single-side AI model executed at the UE 404, prediction model update without UE model update is not feasible. During CSI operation, the base station 402 receives the UE capability information to know the complexity class supported by the UE 404. The base station 402 transfers at least one of the newly trained predictor models to be used at the time of CSI prediction, considering the complexity class and future CSI application window requirements. Then, for CSI reporting, the base station 402 indicates to the UE 404, the AI using which the UE 404 should perform CSI reporting, by appropriately selecting the complexity class and CSI application time instance for prediction.

[0123] As depicted, the CSI prediction configuration set (i.e., set of AI model configurations) (c, m) is computed for new model training, as depicted in step 1002. The base station 402 checks if the training is performed for all configurations, as depicted in step 1004. If the training is performed, then the base station 402 transfers the newly trained predictors to the UE 404 to perform UE-side CSI prediction, as depicted in step 1016. If the training is not performed, a new predictor model configuration is selected corresponding to shortlisted complexity and payload size, as depicted in step 1006.

[0124] Thereafter, the base station 402 checks if the enhancement criteria is met, as depicted in step 1008. If the enhancement criteria are not met, then the base station 402 continues training, as depicted in step 1010. If the enhancement criteria are met, then the base station 402 confirms training is done, as depicted in step 1012. The base station 402 stores new predictor model, as depicted in step 1014. The base station 402 checks if the training is done, as depicted in step 1004, after storing predictor model. The base station 402 transfers the newly trained predictors to the UE 404 to perform UE-side CSI prediction, as depicted in step 1016, after storing the new predictor model.

[0125] The various actions in flow process 1000 may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some actions listed in FIG. 10 may be omitted.

[0126] FIG. 11 depicts a message sequence diagram for performing a site-specific dataset collection. The dataset can be collected using regular 3GPP procedures. For example, consider selection of CSI measurement and reporting configuration based on both scheduler requirements and AI model refinement requirements. As depicted, in an example, CSI compression and reporting is regularly done by the UE 404 as per the requested base station 402 configuration.

[0127] Once a RRC configuration request is sent from the base station 402 to the UE 404 as depicted in step 11-2, the UE 404 completes the RRC configuration as depicted in step 11-4. Later, the base station 402 sends a UE capability enquiry to the UE 404, as depicted in step 11-6. The UE 404 responds to the base station 402 with a UE capability response, as depicted in step 11-8. The base station 402 selects the CSI reporting configurations based on the scheduling requirements and dataset collection requirements, as depicted in step 11-10.

[0128] Further, the CSI reporting configuration is sent from the base station 402 to the UE 404, as depicted in step 11-12. CSI-RS signal is also sent from the base station 402 to the UE 404, as depicted in step 11-14. Later, a CSI report is sent from the UE 404 to the base station 402, as depicted in step 11-16. The CSI report is decoded, as depicted in step 11-18. The decoded CSI report is further used for scheduling and site-specific model refinement, as depicted in step 11-20.

[0129] FIG. 12 depicts a flow process 1200 for performing dataset collection in line with implicit model refinement. Embodiments herein disclose the process of selecting the set of CSI configurations valid for dataset collection using conventional CSI procedures at the base station 402 side. The base station 402 selects the set of valid of CSI configurations, Ψ, valid for implicit refinement procedure. The selection is based on the higher accuracy of CSI reporting for dataset creation, that meets the dataset quality requirements, for further refinement purposes.

[0130] For example, let γ be a squared general cosine similarity (SGCS) requirement for training. Among the conventional CSI codebooks C, the base station 402 may shortlist a set of configurations meeting the squared general cosine similarity threshold,

[0131]

[0132] Similarly, when AI based compression configurations are supported, A, are deployed, the base station 402 shortlists a subset as,

[0133]

[0134] For example, in case of Rel-16 Type-II CSI codebook consists of paramCombination {1,...,8}, out of which 6 and 8 may be chosen, hence Ψ∈{6, 8} due to their high precision reporting performance.

[0135] Embodiments herein disclose the process of performing dataset collection in line with implicit model refinement using conventional CSI procedures at the base station 402 side. While configuring CSI reporting to a UE 404 during RRC configuration, the base station 402 shortlists Ψ. Then, a CSI reporting configuration that meets the accuracy requirements for training, in addition to the mandatory scheduling requirements is chosen. During a UE CSI report reception by the base station 402, the CSI report is decoded and is used for the purpose of scheduling. Then, a pre-processing is applied on the received CSI to extract the features required for dataset; for example, applying transformation on the received pre-coders. The pre-processed data is then stored in the dataset for future implicit training. The base station 402 continues to execute the dataset collection procedure until the dataset requirements are met; for example, the number of samples in the training dataset and their distribution.

[0136] As depicted in the FIG. 12, the base station 402 shortlists Ψ: set of CSI configurations valid for implicit refining procedure, as depicted in step 1202. The base station 402 checks if the UE 404 configured for CSI reporting, as depicted in step 1204. If the UE 404 is configured for CSI reporting, then the base station 402 selects CSI report configuration based on scheduling, as depicted in step 1206, and simultaneously checks for implicit refinement, as depicted in step 1208. Later, CSI reporting is configured to UE 404, as depicted in step 1212. If the implicit refinement is performed, then the base station 402 selects CSI report configuration from Ψ: based on scheduling and AI training, as depicted in step 1210, and the CSI reporting is configured to UE 404, as depicted in step 1212. If the UE 404 is not configured for CSI reporting and / or if the implicit refinement is not performed, then the base station 402 continuously checks if the UE 404 configured for CSI reporting.

[0137] Thereafter, the base station 402 checks if the CSI report is received from the UE 404, as depicted in step 1214. If the CSI report is received from the UE 404, then the CSI report is decoded, as depicted in step 1216, and implicit refinement is checked as depicted in step 1218. If the CSI report is not received and / or the implicit refinement is not performed, then the base station 402 checks continuously if the CSI report is received from the UE 404. If the implicit refinement is performed, then the base station 402 stores the received CSI in dataset as depicted in step 1220, and simultaneously completes dataset collection as depicted in step 1222. After, the dataset is collected completely, the base station 402 checks for receiving the CSI report.

[0138] The various actions in flow process 1200 may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some actions listed in FIG. 12 may be omitted

[0139] FIG. 13 depicts a flow process 1300 of a deployment process for AI based approaches. Embodiments herein can train improved AI models using the site-specific data to refine the performance of models for existing CSI configuration, train AI models for new CSI configurations, and to improve performance of the UEs that do not support new model download, enhance the AI blocks in possession of the base station 402.

[0140] A CSI compression of two-sided AI model operation is considered as an example. In this case, the UE 404 possesses a set of AI encoder models in its memory and the base station 402 would possess corresponding set of AI decoders. As shown in FIG. 13, site-specific dataset is collected by the base station 402, as depicted in step 1302. Existing CSI compression models is selected, by the base station 402, as depicted in step 1304, for refinement.

[0141] The selected models are re-trained using site-specific dataset collection by the base station 402, as depicted in step 1306. Further, new CSI configurations are selected for new model training, as depicted in step 1308. The selected models are trained using site-specific dataset collection by the base station 402, as depicted in step 1310. Therefore, the base station 402 trains new AI models using site-specific dataset for new CSI configurations or for improving performance of existing CSI configurations.

[0142] The various actions in flow process 1300 may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some actions listed in FIG. 13 may be omitted.

[0143] FIGs. 14 and 15 depict flow processes for refining CSI compression models for UEs with and without downloaded AI model support respectively. Model refinement refers to improving AI model performance using site-specific dataset by re-training available models for existing CSI configurations, and training new AI models for existing or new CSI configurations. Embodiments herein can re-train available models for existing CSI configurations. This is dependent on the UE capability to download new models.

[0144] Hence, with UE model transfer support, for CSI compression, performance enhancement of an existing encoder and decoder pair can be achieved by fine-tuning existing AI models using site-specific dataset by re-training both encoder and decoder pair. For UE-side CSI prediction, the AI model is retrained at the base station 402, and then transferred to the UE 404.

[0145] Without UE model transfer support, for CSI compression, performance enhancement can be achieved using site-specific dataset by re-training the decoder part of existing AI models while retaining the same encoder part, already available at the UE 404. For CSI prediction, new AI models can be trained to support only base station side CSI prediction.

[0146] FIG. 14 depicts a flow process 1400 for refining CSI compression models for UEs with download AI model support. At first step 1402, the base station 402 loads CSI compression models. The base station 402 also loads site-specific dataset, as depicted in step 1404. The base station 402 checks if refinement is done for all models as depicted in step 1406, and selects a model for refinement, as depicted in step 1408. The encoder-decoder models are re-trained, as depicted in step 1410.

[0147] Thereafter, the base station 402 selects a re-trained AI model for CSI compression, as depicted in step 1412. The base station 402 transfers the selected encoder model to the UE 404, as depicted in step 1414. Later, the base station 402 performs CSI reporting, as depicted in step 1416.

[0148] The various actions in flow process 1400 may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some actions listed in FIG. 14 may be omitted.

[0149] FIG. 15 depicts a flow process 1500 for refining CSI compression models for UEs without download AI model support. At first step 1502, the base station 402 loads CSI compression models. The base station 402 also loads site-specific dataset, as depicted in step 1504. The base station 402 checks if refinement is done for all models as depicted in step 1506, and selects a model for refinement, as depicted in step 1508. The base station 402 trains only decoder models, as depicted in step 1510.

[0150] Thereafter, the base station 402 selects a re-trained AI model for CSI compression, as depicted in step 1512. The base station 402 transfers the selected encoder model to the UE 404, as depicted in step 1514. Later, the base station 402 performs CSI reporting, as depicted in step 1516.

[0151] The various actions in flow process 1500 may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some actions listed in FIG. 15 may be omitted.

[0152] Embodiments herein can train new AI models for existing or new CSI configurations. Depending on the UE model download capability, with UE model transfer support, existing and new CSI configurations can be shortlisted for new model training. For CSI compression, new encoder-decoder pair can be trained for new CSI configuration. For existing CSI configuration, for example payload size, encoder-decoder or only decoder or different complexity class are retrained. For UE-side CSI prediction, new CSI predictors can be trained for new CSI configuration; for example, new CSI application time-instance. For existing CSI configuration, predictors can be trained of different complexity class.

[0153] Without UE model transfer support, for CSI compression, only new decoders are trained for a given CSI configuration or AI encoder using site-specific data, with same or different complexity. For base station-side CSI prediction, a new CSI prediction model can be trained to perform prediction corresponding to a new CSI application time instance, using site-specific data.

[0154] FIG. 16 depicts a flow process 1600 for refining CSI compression models for UEs with download AI model support by training new AI models. At first step 1602, the base station 402 identifies CSI configurations for training. The base station 402 also loads site-specific dataset, as depicted in step 1604. The base station 402 checks if refinement is done for all models as depicted in step 1606, and selects a model for refinement, as depicted in step 1608. The encoder-decoder models are re-trained, as depicted in step 1610.

[0155] Thereafter, the base station 402 selects a re-trained AI model for CSI compression, as depicted in step 1612. The base station 402 transfers the selected encoder model to the UE 404, as depicted in step 1614. Later, the base station 402 performs CSI reporting, as depicted in step 1616.

[0156] The various actions in flow process 1600 may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some actions listed in FIG. 16 may be omitted.

[0157] FIG. 17 depicts a flow process 1700 for refining CSI compression models for UEs without download AI model support. At first step 1702, the base station 402 identifies CSI configurations for training. The base station 402 also loads site-specific dataset, as depicted in step 1704. The base station 402 checks if refinement is done for all models as depicted in step 1706, and selects a model for refinement, as depicted in step 1708. The base station 402 trains only decoder models, as depicted in step 1710.

[0158] Thereafter, the base station 402 selects a re-trained AI model for CSI compression, as depicted in step 1712. The base station 402 transfers the selected encoder model to the UE 404, as depicted in step 1714. Later, the base station 402 performs CSI reporting, as depicted in step 1716.

[0159] The various actions in flow process 1700 may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some actions listed in FIG. 17 may be omitted.

[0160] FIG. 18 depicts a flow process 1800 for improving the performance of CSI compression for a UE with model transfer capability by training generic models. For UE with model transfer capability, both the encoder and decoder (enc-dec) pair is trained. In this procedure, the base station 402 refines the performance of the pre-trained AI models using the site-specific dataset collected using implicit data collection procedures. For CSI compression, both encoder and decoder pair is trained, with an intent to transfer the refinement encoder, eventually, to the UE 404 to improve CSI compression.

[0161] First, the base station 402 shortlists the AI models for further training. For example, the set of models M are such that the uplink CSI control overhead is maintained within a threshold, i.e., payload size of each model in M is less than a predetermined size. Then, for each AI model in M, the BS loads the model and trains both the encoder and decoder it to improve its performance. At the time of training, the performance improvement is tracked using intermediate KPIs such as squared general cosine similarity (SGCS).

[0162] The model refinement can be performed in the following ways:

[0163] - At the base station, when a AI training server is unavailable; and

[0164] - At a training server, after transferring the models and dataset to the server.

[0165] During CSI compression operation, the updated encoder is transferred to the UE 404, and the updated decoder is selected at the base station 402, for the selected CSI payload size or CSI configuration.

[0166] As depicted at step 1802, the base station 402 selects the CSI compression model set for refinement. The base station 402 checks if the refinement is done for all models, as depicted in step 1804. If the refinement is done, then the base station 402 uses the encoder with updated parameters during UE model transfer for CSI compression, as depicted in step 1818. A model is selected for refinement, if refinement is not performed, as depicted in step 1806. Later, encoder-decoder pair of the AI model is trained, and both encoder and decoder parameters are updated, as depicted in step 1808.

[0167] Thereafter, the base station 402 checks if the enhancement criteria is met, as depicted in step 1810. If the enhancement criteria is not met, then the base station 402 continues training, as depicted in step 1812. If the enhancement criteria is met, then the base station 402 confirms training is done, as depicted in step 1814. The base station 402 stores updated encoder-decoder model pair, as depicted in step 1816. The base station 402 checks if the refinement is done for all models, as depicted in step 1804, after storing updated encoder-decoder model pair. The base station 402 uses the model encoder with updated parameters during UE model transfer for CSI compression, as depicted in step 1818. Thereafter, the decoder is used at the base station 402 with updated parameters for CSI compression operation, as depicted in step 1820.

[0168] The various actions in flow process 1800 may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some actions listed in FIG. 18 may be omitted.

[0169] FIG. 19 depicts a flow process 1900 for improving the performance of CSI compression for a UE without model transfer capability by training generic models. For a UE without model transfer capability, only the decoder part is trained. This variant of CSI compression refinement, is meant for UEs which lack support for new model download from the network. Hence, only the decoder of an existing CSI compression model is trained. First, the base station 402 shortlists the AI models for further training.

[0170] For example, the set of models M are such that the uplink CSI control overhead is maintained within a threshold, i.e., payload size of each model in M is less than a predetermined size. Then, for each AI model in M, the base station 402 loads the model, and updates only the decoder parameters during the training. The encoder parameters are not updated as part of the training. At the time of training, the performance improvement is tracked using intermediate KPIs such as squared general cosine similarity (SGCS). The training location can be either base station 402 or a third party parameter server. During CSI compression operation, the base station 402 uses the updated decoder model to improve CSI compression performance. But, the encoder at the UE 404 remains unchanged.

[0171] As depicted at step 1902, the base station 402 selects the CSI compression model set for refinement. The base station 402 checks if the refinement is done for all models, as depicted in step 1904. If the refinement is done, then the base station 402 uses the decoder with updated parameters at the base station 402 for CSI compression, as depicted in step 1918. A model is selected for refinement, if refinement is not performed, as depicted in step 1906. Later, encoder-decoder pair of the AI model is trained, and only decoder is updated, as depicted in step 1908.

[0172] Thereafter, the base station 402 checks if the enhancement criteria is met, as depicted in step 1910. If the enhancement criteria is not met, then the base station 402 continues training, as depicted in step 1912. If the enhancement criteria is met, then the base station 402 confirms training is done, as depicted in step 1914. The base station 402 stores updated decoder for the model, as depicted in step 1916. The base station 402 checks if the refinement is done for all models, as depicted in step 1904, after storing updated decoder. The base station 402 uses the model decoder with updated parameters at the base station 402 for CSI compression, as depicted in step 1918.

[0173] The various actions in flow process 1900 may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some actions listed in FIG. 19 may be omitted.

[0174] FIG. 20 depicts a flow process 2000 of improving the performance of CSI prediction by training the predictor. For a UE with model transfer capability, the predictor can be trained. In this case, the predictor models are trained at the base station 402 using the additional dataset. The enhanced predictors are then transferred to the UEs to support improved CSI prediction. First, the BS shortlists the CSI prediction configurations for further training. Among the set of CSI application time instances for which prediction is supported, a subset P is selected. Then, for each predictor AI model in P, the base station 402 loads the model, and trains the model by updating the parameters of the predictor during training. For tracking the performance improvement, intermediate KPIs such as squared general cosine similarity (SGCS) is used. During CSI prediction operation: for a CSI prediction configuration, the base station 402 first transfers the updated predictor to the UE 404. The UE 404 then uses the predictor to generate prediction based reports which is then shared with the base station 402.

[0175] As depicted at step 2002, the base station 402 selects the CSI prediction model set for refinement. The base station 402 checks if the refinement is done for all models, as depicted in step 2004. If the refinement is done, then the base station 402 uses the predictor with updated parameters during UE model transfer for CSI prediction, as depicted in step 2018. A model is selected for refinement, if refinement is not performed, as depicted in step 2006. Later, predictor of the AI model is trained by updating its parameters, as depicted in step 2008.

[0176] Thereafter, the base station 402 checks if the enhancement criteria is met, as depicted in step 2010. If the enhancement criteria is not met, then the base station 402 continues training, as depicted in step 2012. If the enhancement criteria is met, then the base station 402 confirms training is done, as depicted in step 2014. The base station 402 stores updated predictor for the model pair, as depicted in step 2016. The base station 402 checks if the refinement is done for all models, as depicted in step 2004, after storing updated decoder. The base station 402 uses the predictor with updated parameters during UE model transfer for CSI prediction, as depicted in step 2018.

[0177] The various actions in flow process 2000 may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some actions listed in FIG. 20 may be omitted.

[0178] As per simulations, Machine Learning (ML) CSI prediction gain with ML based CSI compression exploiting time domain correlation is depicted. Table-1 shows the performance of CSI compression model trained with site-specific dataset collected using an existing AI model. In the example shown, the dataset is collected using a CSI configuration with a payload size of 564 bits. Using this dataset, the base station 402 trains a new model with 272 bits payload size. The performance of the newly trained model is around 0.9356 SGCS, which is similar to the performance of a model with centralized dataset collection procedure.

[0179]

[0180] Therefore, AI models trained with site-specific data perform better than AI models that are trained with a generic dataset. The proposed methods provide UE capability based model transfer and execution, for AI operations. The methods provide efficient site-specific dataset collection that incur minimal resource overhead. Further, to bring the benefits of customized AI models to UEs that lack the capability to dynamically deploy newer models, the proposed methods customize the AI model by only training the model blocks that are executed in the base station 402. The methods are applicable to multiple scenarios discussed in Third Generation Partnership Project (3GPP), for example CSI prediction, CSI compression, location estimation, network handover optimization, and so on.

[0181] The embodiments disclosed herein can be implemented through at least one software program running on at least one hardware device and performing network management functions to control the network elements. The network elements shown in FIG. 4 include blocks which can be at least one of a hardware device, or a combination of hardware device and software module.

[0182] FIG. 21 illustrates a structure of a base station according to an embodiment of the disclosure.

[0183] As shown in FIG. 21, the base station according to an embodiment may include a transceiver 2110, a memory 2120, and a processor 2130. The transceiver 2110, the memory 2120, and the processor 2130 of the base station may operate according to a communication method of the base station described above. However, the components of the base station are not limited thereto. For example, the base station may include more or fewer components than those described above. In addition, the processor 2130, the transceiver 2110, and the memory 2120 may be implemented as a single chip. Also, the processor 2130 may include at least one processor. Furthermore, the base station of FIG. 21 corresponds to the base station 402 of the FIG. 4.

[0184] The transceiver 2110 collectively refers to a base station receiver and a base station transmitter, and may transmit / receive a signal to / from a terminal(UE) or a network entity. The signal transmitted or received to or from the terminal or a network entity may include control information and data. The transceiver 2110 may include a RF transmitter for up-converting and amplifying a frequency of a transmitted signal, and a RF receiver for amplifying low-noise and down-converting a frequency of a received signal. However, this is only an example of the transceiver 2110 and components of the transceiver 2110 are not limited to the RF transmitter and the RF receiver.

[0185] Also, the transceiver 2110 may receive and output, to the processor 2130, a signal through a wireless channel, and transmit a signal output from the processor 2130 through the wireless channel.

[0186] The memory 2120 may store a program and data required for operations of the base station. Also, the memory 2120 may store control information or data included in a signal obtained by the base station. The memory 2120 may be a storage medium, such as read-only memory (ROM), random access memory (RAM), a hard disk, a CD-ROM, and a DVD, or a combination of storage media.

[0187] The processor 2130 may control a series of processes such that the base station operates as described above. For example, the transceiver 2110 may receive a data signal including a control signal transmitted by the terminal, and the processor 2130 may determine a result of receiving the control signal and the data signal transmitted by the terminal.

[0188] FIG. 22 illustrates a structure of a UE according to an embodiment of the disclosure.

[0189] As shown in FIG. 22, the UE according to an embodiment may include a transceiver 2210, a memory 2220, and a processor 2230. The transceiver 2210, the memory 2220, and the processor 2230 of the UE may operate according to a communication method of the UE described above. However, the components of the UE are not limited thereto. For example, the UE may include more or fewer components than those described above. In addition, the processor 2230, the transceiver 2210, and the memory 2220 may be implemented as a single chip. Also, the processor 2230 may include at least one processor. Furthermore, the UE of FIG. 22 corresponds to the UE 404 of the FIG. 4.

[0190] The transceiver 2210 collectively refers to a UE receiver and a UE transmitter, and may transmit / receive a signal to / from a base station or a network entity. The signal transmitted or received to or from the base station or a network entity may include control information and data. The transceiver 2210 may include a RF transmitter for up-converting and amplifying a frequency of a transmitted signal, and a RF receiver for amplifying low-noise and down-converting a frequency of a received signal. However, this is only an example of the transceiver 2210 and components of the transceiver 2210 are not limited to the RF transmitter and the RF receiver.

[0191] Also, the transceiver 2210 may receive and output, to the processor 2230, a signal through a wireless channel, and transmit a signal output from the processor 2230 through the wireless channel.

[0192] The memory 2220 may store a program and data required for operations of the UE. Also, the memory 2220 may store control information or data included in a signal obtained by the UE. The memory 2220 may be a storage medium, such as read-only memory (ROM), random access memory (RAM), a hard disk, a CD-ROM, and a DVD, or a combination of storage media.

[0193] The processor 2230 may control a series of processes such that the UE operates as described above. For example, the transceiver 2210 may receive a data signal including a control signal transmitted by the base station or the network entity, and the processor 2230 may determine a result of receiving the control signal and the data signal transmitted by the base station or the network entity.

[0194] Those skilled in the art will understand that the various illustrative logical blocks, modules, circuits, and steps described in this application may be implemented as hardware, software, or a combination of both. To clearly illustrate this interchangeability between hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above in the form of their functional sets. Whether such function sets are implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. Technicians may implement the described functional sets in different ways for each specific application, but such design decisions should not be interpreted as causing a departure from the scope of this application.

[0195] In the above-described embodiments of the disclosure, all operations and messages may be selectively performed or may be omitted. In addition, the operations in each embodiment do not need to be performed sequentially, and the order of operations may vary. Messages do not need to be transmitted in order, and the transmission order of messages may change. Each operation and transfer of each message can be performed independently.

[0196] Although the FIG.s illustrate different examples of user equipment, various changes may be made to the FIG.s. For example, the user equipment can include any number of each component in any suitable arrangement. In general, the FIG.s do not limit the scope of this disclosure to any particular configuration(s). Moreover, while FIG.s illustrate operational environments in which various user equipment features disclosed in this patent document can be used, these features can be used in any other suitable system.

[0197] The various illustrative logic blocks, modules, and circuits described in this application may be implemented or performed by a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic devices, discrete gates or transistor logics, discrete hardware components, or any combination thereof designed to perform the functions described herein. The general purpose processor may be a microprocessor, but in an alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors cooperating with a DSP core, or any other such configuration.

[0198] The steps of the method or algorithm described in this application may be embodied directly in hardware, in a software module executed by a processor, or in a combination thereof. The software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, register, hard disk, removable disk, or any other form of storage medium known in the art. A storage medium is coupled to a processor to enable the processor to read and write information from / to the storage media. In an alternative, the storage medium may be integrated into the processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and the storage medium may reside in the user terminal as discrete components.

[0199] In one or more designs, the functions may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, each function may be stored as one or more pieces of instructions or codes on a computer-readable medium or delivered through it. The computer-readable medium includes both a computer storage medium and a communication medium, the latter including any medium that facilitates the transfer of computer programs from one place to another. The storage medium may be any available medium that can be accessed by a general purpose or special purpose computer.

[0200] While the disclosure has been shown and described with reference to various embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the disclosure as defined by the appended claims and their equivalents.

Claims

1.A method performed by a base station (402) in a wireless communication system, the method comprising:receiving, from at least one UE (404), at least one user equipment (UE) capability information;obtaining AI models based on the at least one UE capability information;selecting at least one AI model among the AI models, and a channel state information (CSI) configuration for the at least one UE (404); andtraining the at least one AI model for performing at least one CSI operation, based on the CSI configuration.2.The method (600) of claim 1, wherein the at least one UE capability information comprises at least one of a computational complexity capability for each AI based functionality, or capability to download and deploy new AI models from the base station (402) during the at least one CSI operation, andwherein the AI based functionality comprises at least one of an AI compression, an AI prediction, an AI model download, and an AI model selection.3.The method (600) of claim 1, wherein the at least one AI model, and the CSI configuration for the at least one UE (404) are selected based on at least one of the at least one UE capability information, a site-specific dataset collection, at least one scheduling requirements, or at least one AI model training requirements,wherein the at least one scheduling requirements comprises at least one reporting periodicities to the at least one UE (404) for an AI based CSI prediction case, andwherein the at least one AI model training requirements comprises at least one feedback bit configuration for the at least one UE (404) for a CSI reporting in an AI compression case.4.The method (600) of claim 1, further comprising:training the at least one AI model of a high complexity using the site-specific dataset collection; andtransferring the trained at least one AI model to the at least one UE (404) for execution, subject to capability of the at least one UE (404) to download and deploy at least one new AI models.5.The method (600) of claim 1, wherein the method (600) comprises:training a part of the at least one AI model that is executed on a network side using the site-specific dataset collection; anddeploying the trained part of the at least one AI model that is executed on the network side during an AI operation, while leaving the at least one AI model at a UE side untouched.6.The method (600) as claimed in claim 1, further comprising:selecting at least one new CSI configuration for refinement; andtraining at least one new AI model using the site-specific dataset collection, for the at least one new CSI configuration.7.The method (600) of claim 1, further comprising:selecting at least one existing CSI configuration for refinement, wherein the base station (402) selects the at least one existing CSI configuration, based on higher accuracy of CSI reporting for the site-specific dataset creation that meets at least one dataset quality requirements; andtraining the at least one new AI model using the site-specific dataset collection, for the at least one existing CSI configuration.8.The method (600) of claim 6, further comprising:identifying a set of AI model configurations and respective complexity classes for training at least one new AI models based on a UE distribution and throughput requirements, wherein the set of AI model configurations per complexity class is a list of complexity classes and payload sizes of the at least one new AI models that are to be trained.9.A base station (402) in a wireless communication system, the base station (402) comprising:a transceiver; andat least one processor (406) coupled with the transceiver and configured to:receive, from at least one UE (404), at least one user equipment (UE) capability information,obtain AI models based on the at least one UE capability information,select at least one AI model from the downloaded among the AI models, and a channel state information (CSI) configuration for the at least one UE (404), andtrain the at least one AI model for performing at least one CSI operation, based on the CSI configuration.10.The base station (402) of claim 9, wherein the at least one UE capability information comprises at least one of a computational complexity capability for each AI based functionality, or capability to download and deploy new AI models from the base station (402) during the at least one CSI operation, andwherein the AI based functionality comprises at least one of an AI compression, an AI Prediction, an AI model download, and an AI model selection.11.The base station (402) of claim 9, wherein the at least one AI model, and the CSI configuration for the at least one UE (404) are selected based on at least one of the at least one UE capability information, a site-specific dataset collection, at least one scheduling requirements, and at least one AI model training requirements,wherein the at least one scheduling requirements comprises at least one reporting periodicities to the at least one UE (404) for an AI based CSI prediction case, andwherein the at least one AI model training requirements comprises at least one feedback bit configuration for the at least one UE (404) for a CSI reporting in an AI compression case.12.The base station (402) of claim 9, wherein the at least one processor (406) is further configured to:train the at least one AI model of a high complexity using the site-specific dataset collection, andtransfer the trained at least one AI model to the at least one UE (404) for execution, subject to capability of the at least one UE (404) to download and deploy at least one new AI models.13.The base station (402) of claim 9, wherein the at least one processor (406) is further configured to:train a part of the at least one AI model that is executed on a network side using the site-specific dataset collection, anddeploy the trained part of the at least one AI model that is executed on the network side during an AI operation, while leaving the at least one AI model at a UE side untouched.14.The base station (402) of claim 9, wherein the at least one processor (406) is further configured to:select at least one new CSI configuration for refinement, andtrain at least one new AI model using the site-specific dataset collection, for the at least one new CSI configuration.15.The base station (402) of claim 13, wherein the at least one processor (406) is further configured to:select at least one existing CSI configuration for refinement, wherein the base station (402) selects the at least one existing CSI configuration, based on higher accuracy of CSI reporting for the site-specific dataset creation that meets at least one dataset quality requirements, andtrain the at least one new AI model using the site-specific dataset collection, for the at least one existing CSI configuration.

Images