Systems and methods for CSI feedback using ai / ml

Abstract

Embodiments disclosed herein relate to methods (400) and systems (300) for providing Channel State Information (CSI) feedback in wireless communication networks using Artificial Intelligence (AI) / Machine Learning (ML), and designing CSI-Reference Signal (RS) for 5G, and beyond 5G networks. The methods (400) include providing feedback estimates of steering vectors (spatial domain), delays (frequency domain), and Doppler frequency (time domain), wherein a channel can be predicted / constructed over the stationary time. The methods (400) include determining at least one region of Normalized digital frequencies (NDFs) for each cluster of a channel, and reporting details of NDFs to a base station (304) for reconstructing the channel.

Description

SYSTEMS AND METHODS FOR CSI FEEDBACK USING AI / ML

[0001] Embodiments disclosed herein relate to wireless communication networks, and more particularly to providing Channel State Information (CSI) feedback in wireless communication networks using Artificial Intelligence (AI) / Machine Learning (ML), and designing CSI-Reference Signal (RS) for 5th generation (5G), and beyond 5G networks. It also pertains to design of CSI-RS for a plurality number of ports.

[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 3.5 GHz, but also in "Above 6 GHz" bands referred to as mmWave including 28 GHz and 39 GHz. In addition, it has been considered to implement 6th generation (6G) mobile communication technologies (referred to as Beyond 5G systems) in terahertz bands (for example, 95 GHz to 3 THz 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 multi input multi output (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 bandwidth part (BWP), new channel coding methods such as a low density parity check (LDPC) 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 vehicle-to-everything (V2X) 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, new radio unlicensed (NR-U) 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, integrated access and backhaul (IAB) 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 dual active protocol stack (DAPS) 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 Augmented Reality (AR), Virtual Reality (VR), Mixed Reality (MR) 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 Orbital Angular Momentum (OAM), and Reconfigurable Intelligent Surface (RIS), 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 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] Existing methods for Channel State Information (CSI) feedback take Fast Fourier Transform (FFT), and feedback the FFT vector (indices) and FFT coefficients.

[0009] FIG. 1A depicts FFT of a signal whose discrete frequency k=f is an integer. FIG. 1B depicts an example scenario where the signal power is spreading, when the frequency f=k is a non-integer. As depicted in an example in FIGS. 1A and 1B, consider a signal of N samples:　Y(n)= , where n: sample index, f: Normalized Digital Frequency (NDF). Sometimes f / N can be considered as NDF too. It can be understood from context. Let, Y=[y(0) y(1) ...y(N-1)]. Consider a case, where f is integer say "3". Then FFT of Y has only one bin (as shown in FIG. 1A). If f is a fraction say 3.3, FFT of Y has many bins, as FFT is a sinc shaped, it is called as leakage (as shown in FIG. 1B). So, Y can be represented by N FFT points or approximated by L << N FFT points around 3. If the frequency is an integer, then the FFT works perfectly. In case if the frequency is a non-integer, then signal power is spreading.

[0010] FIG. 1C depicts an example scenario wherein there are two frequencies. In scenarios, where there are two frequencies in a bin (as depicted in FIG. 1C), estimating two frequencies in a bin is more challenging. Estimation of multiple frequencies is another challenging task. Error associated in estimating the multiple frequencies is another challenging task.

[0011] Current solution (R. M. Dreifuerst and R. W. Heath, "SignalNet: A Low Resolution Sinusoid Decomposition and Estimation Network," in　IEEE Transactions on Signal Processing, vol. 70, pp. 4454-4467, 2022) resolves signals where NDF is such that there is only one NDF per bin.

[0012] Another Quasi NN solution (Y. Jiang, T. Zhang and W. Zhang, "Line Spectral Estimation Inspired by Quasi-Neural Network," in　IEEE Transactions on Signal Processing, vol. 70, pp. 5822-5832, 2022) can resolve up to two NDFs per FFT bin.

[0013] At N=32, the signal corresponds to three NDFs 0.1, 0.115 and 0.37. NDFs 0.1 and 0.115 lie between bins 3 and 4. FIG. 2 depicts a plurality of NDFs that is estimated by a neural network. From FIG. 2, NDFs 0.1 and 0.115 between FFT bins 3 and 4 can be resolved by ML but not by conventional signal processing algorithms like MUSIC. In 3GPP, there may be many NDFs per FFT bin, and the Artificial Intelligence (AI) / Machine Learning (ML) may fail to resolve all NDFs.

[0014] 5G networks use channel estimation from CSI-Reference Signal (RS) signals for downlink pre-coder design. Table 7.4.1.5.3-1 in 3GPP document 38.211 provides CSI-RS locations within a slot. Primarily, code-division in frequency, and time domains are used to provide unique CSI-RS patterns for supporting up to 32 ports. 6G systems envision support for much higher number of ports, sometimes going all the way up to 128 ports. Newer unique CSI-RS patterns, and sound design principles are required to systematically support such a large number of CSI-RS ports, and perhaps extend to even higher numbers in the future.

[0015] Hence, there is a need in the art for solutions which will overcome the above mentioned drawback(s), among others.

[0016] The principal object of embodiments herein is to disclose methods and systems for providing Channel State Information (CSI) feedback in wireless communication networks using Artificial Intelligence (AI) / Machine Learning (ML).

[0017] Another object of embodiments herein is to disclose methods and systems for providing feedback estimates of steering vectors (spatial domain), delays (frequency domain), and Doppler frequency (time domain), wherein a channel can be predicted / constructed over the stationary time.

[0018] Another object of embodiments herein is to disclose methods and systems for determining at least one region of Normalized digital frequencies (NDFs) for each cluster of a channel, and reporting details of NDFs to a base station for reconstructing the channel.

[0019] Another object of embodiments herein is to disclose methods and systems for designing CSI-Reference Signal (RS) for 5G, and beyond 5G networks, wherein a Kronecker product based CSI-RS design along frequency and time dimensions is used for supporting a large number of ports.

[0020] Accordingly, the embodiments herein provide a method for providing Channel State Information (CSI) feedback in a wireless communication network by a User Equipment (UE). The method comprises separating at least one channel into one or more clusters, determining at least one region of one or more Normalized digital frequencies (NDFs) for each cluster, an associated complex scalar, and a number of approximated NDFs in each cluster, approximating the channel in at least one of a time domain, a frequency domain, and a spatial domain using the determined region of the NDFs, and reporting the determined region of the NDFs, the associated complex scalar, the number of approximated NDFs in each cluster, and the approximated channel to a base station.

[0021] Accordingly, the embodiments herein provide a UE which comprises a processor, and a memory module (310). The processor is coupled with the memory module (310). The processor is configured to separate at least one channel into one or more clusters, determine at least one region of one or more Normalized digital frequencies (NDFs) for each cluster, an associated complex scalar, and a number of approximated NDFs in each cluster, approximate the channel in at least one of a time domain, a frequency domain, and a spatial domain using the determined region of the one or more NDFs, and report the determined region of the NDFs, the associated complex scalar, the number of approximated NDFs in each cluster, and the approximated channel to a base station.

[0022] 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.

[0023] 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 illustrator drawings. Embodiments herein are illustrated by way of examples in the accompanying drawings, and in which:

[0024] FIG. 1A depicts FFT of a signal whose discrete frequency k=f is an integer, according to existing arts;

[0025] FIG. 1B depicts an example scenario where the signal power is spreading, when the frequency f=k is a non-integer, according to existing arts;

[0026] FIG. 1C depicts an example scenario wherein there are two frequencies, according to existing arts;

[0027] FIG. 2 depicts a plurality of NDFs that is estimated by a neural network, according to existing arts;

[0028] FIG. 3 depicts a block diagram of a system for providing a CSI feedback in a wireless communication network, according to embodiments as disclosed herein;

[0029] FIG. 4 depicts a method for providing CSI feedback in a wireless communication network by the UE, according to embodiments as disclosed herein;

[0030] FIG. 5 depicts a representation of Y(n) with just one frequency that is fractional, according to embodiments as disclosed herein;

[0031] FIG. 6 depicts an example scenario, which is a 2D antenna layout at a transmitter, according to embodiments as disclosed herein;

[0032] FIG. 7 is a representation of the signal on a 2D FFT grid, according to embodiments as disclosed herein;

[0033] FIGS. 8A, 8B, 8C and 8D depict NDFs in different domains for the 3GPP Clustered Delay Line (CDL) model, according to embodiments as disclosed herein;

[0034] FIG. 9A depicts the actual channel being provided as feedback, according to embodiments as disclosed herein;

[0035] FIG. 9B depicts the approximate channel being provided as feedback, according to embodiments as disclosed herein;

[0036] FIG. 10 depicts a process of providing CSI feedback, according to embodiments as disclosed herein;

[0037] FIG. 11 depicts an example scenario, wherein the frequencies are estimated in a 2D domain, according to embodiments as disclosed herein;

[0038] FIG. 12A depicts a block form of neural network using LSTM where a=2, b=2, both real / imaginary values are fed to neural network, according to embodiments as disclosed herein;

[0039] FIG. 12B depicts a block form of neural network using LSTM where a=8, b=8, both real / imaginary values are fed to neural network, according to embodiments as disclosed herein;

[0040] FIG. 13 depicts a block form of neural network using CNN for a=2, b=2, according to embodiments as disclosed herein;

[0041] FIG. 14 depicts a block form of neural network using CNN for a=8, b=8, according to embodiments as disclosed herein;

[0042] FIG. 15 depicts a block form of neural network using CNN for a=4, b=4, according to embodiments as disclosed herein;

[0043] FIG. 16 depicts a block form of neural network using CNN for a=3, b=3, according to embodiments as disclosed herein;

[0044] FIG. 17 depicts an example 5G OCC for CDM4-FD2-TD2, according to embodiments as disclosed herein;

[0045] FIG. 18 depicts an example RE mapping for a CDM8-FD2-TD4 OCC, according to embodiments as disclosed herein;

[0046] FIG. 19A depicts an example RE mapping in a contiguous time-frequency grid, according to embodiments as disclosed herein;

[0047] FIG. 19B depicts an example RE mapping in a non-contiguous uniform time-frequency grid, according to embodiments as disclosed herein;

[0048] FIG. 19C depicts an example RE mapping in a non-contiguous non-uniform time-frequency grid, according to embodiments as disclosed herein;

[0049] FIG. 20A depicts an example figure for CDM group separation in a uniform time-frequency grid, according to embodiments as disclosed herein;

[0050] FIG. 20B depicts an example figure for CDM group separation in a non-uniform time-frequency grid, according to embodiments as disclosed herein;

[0051] FIG. 21A depicts an example 2-port OCC structure, according to embodiments as disclosed herein;

[0052] FIG. 21B depicts an example 3-port OCC structure, according to embodiments as disclosed herein;

[0053] FIGS. 22A, 22B, and 22C depict example three 4-order OCC structures, according to embodiments as disclosed herein;

[0054] FIGS. 23A, 23B, 23C, and 23D depict examples OCC structures, according to embodiments as disclosed herein;

[0055] FIG. 24 depicts an example of the Kronecker operation, according to embodiments as disclosed herein;

[0056] FIG. 25 depicts a 2D layout of antenna ports and / or corresponding 2D spatial channel, according to embodiments as disclosed herein;

[0057] FIG. 26 depicts training of the neural network using the FFT SRAPs, and the ground truth, according to embodiments as disclosed herein;

[0058] FIG. 27 depicts predicting FFT-ERAP from the FFT_SRAPs, according to embodiments as disclosed herein;

[0059] FIG. 28 depicts an entire frequency band of SRS (EFB_SRS), along with sub-band SRSs, according to embodiments as disclosed herein;

[0060] FIG. 29 depicts training of the neural network using the FFTs of all SB_SRS, and ground truth, according to embodiments as disclosed herein; and

[0061] FIG. 30 depicts predicting the FFT-EFB_SRS from the FFT_SB_SRSs, according to embodiments as disclosed herein.

[0062] The embodiments herein and the various features and advantageous details thereof are explained more fully with reference to the non-limiting embodiments that are illustrated in the accompanying drawings and detailed in the following description. Descriptions of well-known components and processing techniques are omitted so as to not unnecessarily obscure the embodiments herein. The examples used herein are intended merely to facilitate an understanding of ways in which the embodiments herein may be practiced and to further enable those of skill in the art to practice the embodiments herein. Accordingly, the examples should not be construed as limiting the scope of the embodiments herein.

[0063] For the purposes of interpreting this specification, the definitions (as defined herein) will apply and whenever appropriate the terms used in singular will also include the plural and vice versa. It is to be understood that the terminology used herein is for the purposes of describing particular embodiments only and is not intended to be limiting. The terms "comprising", "having" and "including" are to be construed as open-ended terms unless otherwise noted.

[0064] The words / phrases "exemplary", "example", "illustration", "in an instance", "and the like", "and so on", "etc.", "etcetera", "e.g.," , "i.e.," are merely used herein to mean "serving as an example, instance, or illustration." Any embodiment or implementation of the present subject matter described herein using the words / phrases "exemplary", "example", "illustration", "in an instance", "and the like", "and so on", "etc.", "etcetera", "e.g.," , "i.e.," is not necessarily to be construed as preferred or advantageous over other embodiments.

[0065] Embodiments herein may be described and illustrated in terms of blocks which carry out a described function or functions.　These blocks, which may be referred to herein as managers, units, modules, hardware components or the like, are physically implemented by analog and / or digital circuits such as logic gates, integrated circuits, microprocessors, microcontrollers, memory circuits, passive electronic components, active electronic components, optical components, hardwired circuits and the like, and may optionally be driven by a firmware. The circuits may, for example, be embodied in one or more semiconductor chips, or on substrate supports such as printed circuit boards and the like.　The circuits constituting a block may be implemented by dedicated hardware, or by a processor (e.g., one or more programmed microprocessors and associated circuitry), or by a combination of dedicated hardware to perform some functions of the block and a processor to perform other functions of the block.　Each block of the embodiments may be physically separated into two or more interacting and discrete blocks without departing from the scope of the disclosure.　Likewise, the blocks of the embodiments may be physically combined into more complex blocks without departing from the scope of the disclosure.

[0066] It should be noted that elements in the drawings are illustrated for the purposes of this description and ease of understanding and may not have necessarily been drawn to scale. For example, the flowcharts / sequence diagrams illustrate the method in terms of the steps required for understanding of aspects of the embodiments as disclosed herein. Furthermore, in terms of the construction of the device, one or more components of the device may have been represented in the drawings by conventional symbols, and the drawings may show only those specific details that are pertinent to understanding the present embodiments so as not to obscure the drawings with details that will be readily apparent to those of ordinary skill in the art having the benefit of the description herein. Furthermore, in terms of the system, one or more components / modules which comprise the system may have been represented in the drawings by conventional symbols, and the drawings may show only those specific details that are pertinent to understanding the present embodiments so as not to obscure the drawings with details that will be readily apparent to those of ordinary skill in the art having the benefit of the description herein.

[0067] The accompanying drawings are used to help easily understand various technical features and it should be understood that the embodiments presented herein are not limited by the accompanying drawings. As such, the present disclosure should be construed to extend to any modifications, equivalents, and substitutes in addition to those which are particularly set out in the accompanying drawings and the corresponding description. Usage of words such as first, second, third etc., to describe components / elements / steps is for the purposes of this description and should not be construed as sequential ordering / placement / occurrence unless specified otherwise.

[0068] The embodiments herein disclose methods and systems for providing Channel State Information (CSI) feedback in wireless communication networks using Artificial Intelligence (AI) / Machine Learning (ML), and for designing CSI-Reference Signal (RS) for 5G, and beyond 5G networks. Referring now to the drawings, and more particularly to FIGS. 3 through 30, where similar reference characters denote corresponding features consistently throughout the figures, there are shown embodiments.

[0069] Embodiments herein disclose providing feedback estimates of steering vectors (spatial domain), delays (frequency domain) and Doppler frequency (time domain), wherein the channel can be predicted / constructed over the stationary time. The stationary time is roughly 100x coherence time. Therefore, uplink feedback can be reduced by a great amount.　The channel over the stationary time is the scaled sum of Doppler frequencies. If Doppler frequencies can be estimated, then channel can be constructed over stationary time.

[0070] Embodiments herein create unique CSI-RS patterns to support larger number of CSI-RS ports in 5G, and beyond 5G networks. 5G systems have been using product codes along time-frequency dimensions for supporting up to 32 ports. Here, a Kronecker product based CSI-RS design is proposed along frequency and time dimensions to handle this scenario.

[0071] Although the terminology "time" and "frequency" is used herein, it is to be understood that embodiments herein can be applicable to any two independent dimensions, and not just time and frequency, for example space-time, space-frequency, and so on.

[0072] FIG. 3 depicts a block diagram of a system 300 for providing a CSI feedback in a wireless communication network. The system 300 comprises a User Equipment (UE) 302, and a base station 304. The UE 302 further comprises a processor 306, a communication module 308, and a memory module 310.

[0073] In an embodiment herein, the processor 306 can approximate at least one channel for CSI feedback, and the approximated channel can be reported by the processor 306 to the base station 304. The processor 306 can further comprise a channel approximation module 312, a feedback module 314, and CSI-RS module 316.

[0074] In an embodiment herein, the channel approximation module 312 can separate at least one channel into one or more clusters. The channel approximation module 312 can transform the channel into a transform domain using at least one transform method for obtaining a matrix of values. The transform method can include, but not limited to a two-dimensional (2D) Discrete Fourier Transform (DFT) (2D DFT), a 2D oversampled DFT, a 2D Slepian, a 2D Discrete Cosine Transform (DCT), a 1D DFT, a 1D oversampled DFT, a 1D Slepian, and a 1D DCT.

[0075] In an embodiment herein, the channel approximation module 312 can determine at least one region of one or more Normalized digital frequencies (NDFs) for each cluster, an associated complex scalar, and a number of approximated NDFs in each cluster. In an embodiment herein, the channel approximation module 312 can identify location of peaks of each cluster of the channel in at least one of the 1D oversampled DFT, and the 2D oversampled DFT. The channel approximation module 312 can obtain a cluster submatrix around each peak of each cluster in at least one of the 1D DFT, and the 2D DFT. The channel approximation module 312 can feed the cluster submatrix, and the location of peaks of each cluster to a neural network for determining co-ordinates of the region of the NDFs for each cluster.

[0076] In an embodiment herein, the channel approximation module 312 can upconvert at least one of a subset of 1D Slepian basis, and a subset of 2D Slepian basis of each cluster using at least one of a 1D frequency, and a 2D frequency for obtaining at least one of a 1D upconverted frequency, and a 2D upconverted frequency. The 1D frequency, and the 2D frequency corresponds to the location of peaks of each cluster. The channel approximation module 312 can collect the at least one of the 1D upconverted frequency, and the 2D upconverted frequency of the one or more clusters into a matrix. The channel approximation module 312 can project the channel onto the matrix for obtaining one or more Slepian coefficients of each cluster. The channel approximation module 312 can feed a submatrix of the Slepian coefficients, the at least one of the 1D upconverted frequency, and the 2D upconverted frequency of each cluster to the neural network for obtaining co-ordinates of the region of the NDFs for each cluster. Examples of neural network include, but not limited to Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and U-Net.

[0077] In an embodiment herein, the channel approximation module 312 can approximate the channel in at least one of a time domain, a frequency domain, and a spatial domain using the determined region of the NDFs. The channel approximation module 312 can approximate the channel using at least one Artificial Intelligence (AI) model. In an embodiment herein, the channel approximation module 312 can equi-space the NDFs in the region. The channel approximation module 312 can quantize a scaling factor associated with at least one of each cluster, and or the at least one region of the one or more NDFs. The channel approximation module 312 can approximate the channel with the equi-spaced NDFs in each cluster, and the quantized scaling factor.

[0078] In an embodiment herein, the feedback module 314 can report the determined region of presence of the NDFs, the associated complex scalar, the number of approximated NDFs in each cluster, and the approximated channel to the base station 304 for reconstructing the channel.

[0079] In an embodiment herein, the CSI-RS module 316 can include a 5G Orthogonal Cover Code (OCC) design. The CSI-RS module 316 can group one or more CSI-RS ports into Code-Division Multiplexing (CDM) groups comprised of contiguous or closely-located groups of Resource Elements (REs). OCC can be applied over each CDM group individually.

[0080] In an embodiment herein, the CSI-RS module 316 can apply OCC within each CDM group, where the applied OCC is a product of individual frequency-domain and time-domain OCCs. For example, the CSI-RS module 316 can create a 4-element CDM group, henceforth called CDM4, by multiplying two orthogonal 2-element sequences, one over frequency-domain and the other over time-domain. Such an OCC sequence is denoted as a CDM4-FD2-TD2 sequence.

[0081] In an embodiment herein, the processor 306 can process and execute data of a plurality of modules of the UE 302. The processor 306 can be configured to execute instructions stored in the memory module 310. The processor 306 may comprise one or more of microprocessors, circuits, and other hardware configured for processing. The processor 306 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 306 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 306 of the UE 302 can communicate via the communication module 308. The communication module 308 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 310 may comprise one or more volatile and non-volatile memory components which are capable of storing data and instructions of the modules of the UE 302 to be executed. Examples of the memory module 310 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 310 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 310 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 310 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. 3 shows example modules of the system 300, but it is to be understood that other embodiments are not limited thereon. In other embodiments, the system 300 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 system 300.

[0085] FIG. 4 depicts a method 400 for providing CSI feedback in a wireless communication network by the UE 302. The method 400 includes separating at least one channel into one or more clusters, as depicted in step 402. The method 400 includes determining at least one region of one or more NDFs for each cluster, an associated complex scalar, and a number of approximated NDFs each cluster, as depicted in step 404. Thereafter, the method 400 includes approximating the channel in at least one of a time domain, a frequency domain, and a spatial domain using the determined region of the NDFs, as depicted in step 406. Later, the method 400 includes reporting the determined region of presence of the NDFs, the associated complex scalar, the number of approximated NDFs in each cluster, and the approximated channel to the base station 304, as depicted in step 408, for reconstructing the channel.

[0086] The various actions in method 400 may be performed in the order presented, in a different order or simultaneously. Further, in some embodiments, some actions listed in FIG. 4 may be omitted.

[0087] Consider a signal of N samples:　Y(n)= , where n: sample index, f: Normalized digital frequency (NDF). Sometimes f / N can be considered as NDF too. It can be understood from context. Let, Y= [ y(0) y(1) ...y(N-1)]. Consider a case, where f is integer say "3", then Fast Fourier Transform (FFT) of Y has only one bin. If f is a fraction say 3.3, FFT of Y has many bins, as FFT is a sinc shaped, it is called as leakage. So, Y can be represented by N FFT points or approximated by L << N FFT points. However, from Y using the estimate that f=3.3, then all of N samples of Y can be reconstructed. In that case, only one value of the "f" can be provided as feedback.

[0088] In the above example, Y comprises of only one exponential frequency.　In practice, Y can be a scaled sum of sinusoids like:

[0089] Y(n)=

[0090] where there are M NDFs fm: m=0,1,2,..., M-1.

[0091] In practice, if Y is a channel across antennas, for a given subcarrier and time, each NDF is due to a ray in a

[0092] cluster and due to the corresponding steering vector with associated AoD (Angle of Departure). Embodiments herein use 1D frequency for a 1D antenna layout as an example. For a 2D antenna layout, it will be the sum of 2D frequencies.

[0093] Now, Y can be the channel in spatial, frequency, time domain.

[0094] Spatial: The exponential sinusoids are steering vectors. Y is the channel at any given instant in any sub-carrier but across antennas (for simplicity, a ID antennas layout is considered).

[0095] Frequency: The exponential sinusoids are due to delay elements (fm). Y is channel across subcarriers for a given antenna and given time instant.

[0096] Time: The exponential sinusoids are due to Doppler frequencies (fm). Y is channel across time for a given subcarrier and antenna.

[0097] If fm=3.3, consider that there is one NDF between third and fourth bin.　Consider that Y(n)= , where fmis a NDF. Consider a case where fm=3.3. FIG. 5 depicts a representation of the Y(n).

[0098] For, Y(n)= , where f1, f2are NDFs. Consider a case where f1=3.3, f2=5.5, as depicted in FIG. 1C.

[0099] FIG. 6 depicts an example scenario, which is a 2D antenna layout at a transmitter, according to an embodiment of the disclosure. Consider a case of a 2D antenna layout. For simplicity, in what follows, assume Nr=1 receive antenna at UE and one polarization only (as depicted in FIG. 6). Consider one ray at the receiver. The channel is , where G is the gain of the path.

[0100] Nr= Number of antenna rows

[0101] Nc= Number of antenna columns

[0102] dr= Distance between two elements in a row

[0103] dc= Distance between two elements in a columns

[0104]

[0105]

[0106] The above expressions are steering vectors. The ray is leaving at the angle (θ, φ). , and are complex exponential sequences of form . For , f = (-drcos(θ)Nr / λ). For , g = (-dccos(φ)Nc / λ). As an example, f=3.3 and g=4.4. In 2DD FFT grid, Nr=Nc=8. The signal channel is as follows:

[0107]

[0108] FIG. 7 is a representation of a signal on a 2D FFT grid. Since f and g are fractional in 2D FFT. Actual (f, g) = (3.3, 4.4). There may be energy spread in FFT bins around (3.3, 4.4) like bins (3, 3),

[0109] (3, 4), (4, 3), (4, 4), and so on. In conventional 3GPP, these bin indices (3, 3), (3, 4), (4, 3), and (4, 4) can be provided as feedback along with

[0110] quantized values of 2D FFT at these bins. This is design of W1in 3GPP. In the proposed method 400, embodiments herein send the location of (f, g)= (3.3, 4.4) along with the quantized value of .

[0111] Consider how the channel varies in frequency domain. In the spatial channel, it can be seen that NDF of channel comes from angle θ, φ is the form of:

[0112] dr*cos(θ) Nr / λ and dc*cos(φ) Nc / λ

[0113] The NDF of complex exponentials in a channel across frequency arises. Consider a channel across N sub-carriers h(1), h(2),..., h(N). Embodiments herein have h(k) = , where is gain of lth multipath, is timing offset of lth multipath, and k is subcarrier index.

[0114] It can be seen that h(k) is sum of L Complex Exponentials (CEs) of NDF - .

[0115] Embodiments herein show how the channel is modelled across Orthogonal Frequency Division Multiplexing (OFDM) symbols for a given antenna and sub-carrier. The channel is kth OFDM per antenna per symbol be:

[0116]

[0117] where dlis NDF due to Doppler / Frequency offset.

[0118] Table 1 depicts NDFs in CE of channel in domain.

[0119]

[0120] It can be seen that the channel in any dimension is sum of CE. Generalizing for N samples,

[0121]

[0122] where are not multiples of (1 / N). So FFT can not be used for estimation. Embodiments herein estimate .Conventionally, MUSIC, ESPIRIT are used to estimate fl. However if many flare closely spaced then it may fail. ML or Deep Learning (DL) or atomic norm based algorithms can perform better in such scenarios, and better than conventional signal processing algorithms like MUSIC and ESPIRIT. Even for ML / DL algorithms, more than two NDFs per FFT bin cannot be resolved. The 3GPP Clustered Delay Line (CDL)　model, has many rays in each cluster. All these rays may result in many NDFs per FFT bin, where even ML / DL algorithms fail.

[0123] FIGS. 8A-8D depict NDFs for the 3GPP CDL model. The NDFs associated with the various CEs are such that all CEs have roughly the same amplitude (in spatial, frequency, time domain). Embodiments herein are discussed in the spatial domain only. Similar approaches can be used to estimate channel in other domains.

[0124] Embodiments include approximation of the channel by estimating the region of presence of NDFs (which is reported by the UE 302 to the base station 304), and also reporting the number of equi-spaced NDFs in this region, and quantized scaling factor associated with at least one of each cluster, and the region of the NDFs, by the UE 302 to the base station 304.

[0125] FIGS. 9A and 9B depict the actual channel and the approximate channel being provided as feedback, respectively. FIG. 9A depicts a region of NDFs in an actual 2D spatial channel, where NDFs are randomly located in this region. FIG. 9B depicts a region of NDFs in an approximate 2D spatial channel, where NDFs are equi-spaced located in this region. A Domain Region Estimator and Channel Approximator (DRECA) is in ML. The same approach can be the channel in frequency and time domains.

[0126] As ML works better at a cluster level, embodiments herein separate the clusters. Two clusters can have same delay spread, and Doppler spread in which case the clusters cannot be separated in frequency / time domains. But they are distinct and can be separated in Spatial domain as Angle of Departure (AoD) are distinct.

[0127] FIG. 10 depicts a process of providing CSI feedback, according to embodiments of the disclosure. As depicted in FIG. 10, the channels are separated into dominant clusters. Later, the region of NDFs is determined for each cluster in various domains. The UE 302 can feedback region details, number of NDFs, and associated scaling factors to the base station 304, which are used to reconstruct the channel.

[0128] For example, the channel approximation module 312 comprises a cluster separator 1002. The cluster separator 1002 receives the channel in spatial, frequency, time domains, and separates the channel into cluster 1, cluster 2,..., cluster n in spatial, frequency, time domains. The channel approximation module 312 comprises DRECA 1004 that contains NDF region estimators in various domains, various clusters, and number of associated NDFs, scaling factors. The feedback module 314 reports feedback from the DRECA 1004 for channel approximation to the base station 304.

[0129] The methods 400 can use a U-NET architecture for finding the NDFs in various domains or the region of NDFs. The methods 400 can also use a modified form of the U-Net architecture for determining region of NDFs in various domains. The methods 400 can replace more than six NDFs / CEs but have to scale appropriately.

[0130] Furthermore, if the CEs are between say 2.2 and 2. 6, all the six equi-spaced CEs may lie between 2.1 and 2.6 as 2.1, 2.2, 2.3,..., 2.6. So the problem becomes one of the estimating the bandwidth, i.e., [2.1 - 2.6]. This may have less feedback as feedback of actual / NDF needs a lots of bits.

[0131] FIG. 11 depicts an example scenario, wherein the frequencies are estimated in a 2D domain. For example, since f and g are fractional in 2D FFT, Actual (f, g) = (3.3, 4.4). There may be energy spread in FFT bins around (3.3, 4.4) like bins (3, 3), (3, 4), (4, 3), and (4, 4) etc. In the proposed method, location of (f, g) = (3.3, 4.4) along with quantized value of is sent to the base station 304.

[0132] In an embodiment herein, for example N1×N2spatial channel (or eigen vector of covariance matrix) at any sub-band, and time instant is transformed into a transform domain by considering either a 2D-DFT, 2D-oversampled-DFT, 2D Slepian, 2D DCT or any other transform to get N1×N2matrix of values.

[0133] For example, from the 2D oversampled DFT peak (m, n) locations are identified for each cluster. ab values in the form of a×b submatrix around each peak of a cluster in the 2D DFT domain are obtained, and this is called a cluster-submatrix (of the spatial channel). This is fed to a neural network, and the output of the neural network is the co-ordinates of the region of NDFs for each cluster. The locations of the peaks of each cluster are also fed to the neural network. The neural network architectures assume location of a cluster's peak at (0, 0), and only one cluster is considered.

[0134] For example, for Slepian based transform, a subset of 2D Slepian basis is up-converted using 2D frequency, where the 2D frequency corresponds to the location of the peak of the cluster. All the up-converted 2D Slepian basis of all clusters is collected into a matrix, and the spatial channel is projected onto this matrix (using pseudo inverse of the matrix which is multiplied to spatial channel). This provides us the Slepian coefficients of each cluster. ab values in the form of a×b submatrix of these coefficients for each cluster along with the 2D up-converted frequency is given as input to the neural network, and its output is the co-ordinates of the region of NDFs of each cluster. The same procedure can be followed to compress a 2D channel across sub-bands and time as well. In the results, that follow N1=8, N2=8.

[0135] In an embodiment herein, a neural network using a Long Short-Term Memory (LSTM) can be considered. FIGS. 12A and 12B depict a block form of neural network using LSTM. a=2, b=2, both real / imaginary values are fed to neural network, corresponds to FIG. 12A. a=N1, b=N2, both real / imaginary values are fed to neural network, corresponds to FIG. 12B.

[0136] In an embodiment herein, a neural network using CNN for a=2, b=2 can be considered. FIG. 13 depicts a block form of neural network using CNN for a=2, b=2.

[0137] In an embodiment herein, a neural network using CNN for a=8, b=8 can be considered. FIG. 14 depicts a block form of neural network using CNN for a=8, b=8.

[0138] In an embodiment herein, a neural network using CNN for a=4, b=4 can be considered. FIG. 15 depicts a block form of neural network using CNN for a=4, b=4.

[0139] In an embodiment herein, a neural network using CNN for a=3, b=3 can be considered. FIG. 16 depicts a block form of neural network using CNN for a=3, b=3.

[0140] In an embodiment herein, a series of dense networks with any activation function and varying width for each layer can also be considered as another embodiment of the neural network.

[0141] In an embodiment herein, for approximating region of NDFs with equi-spaced frequency:

[0142] Consider a region R1 of NDFs, each NDF associated with a roughly similar channel gain. The sum of these NDFs is the channel that needs to feedback from the base station 304 to the UE 302. The NDFs may be randomly distributed in the region. The channel can be approximated by having uniformly spaced NDFs in this region, and one associated scalar for this region. Let the region R1 be centred at (a1, b1), and have length and height wc1 and wr1, respectively. This is in the 2D NDF plane (after taking DTFT). Similarly, let us assume another Region R2 be centred around (a2, b2). The channel is the sum of two clusters or regions.

[0143] Define

[0144] The NR×NCchannel is given as where the first summation is over thei1th 2D NDF corresponding to (ai1, bi1), and gain Ai1in Region R1and the second summation is over thei2th 2D NDF corresponding to (ai2, bi2), and gain Ai2in Region R2. Let there be k1such NDFs in R1and k2such NDFs in R2, the channel. To estimate H, R1is approximated by L1equi-spaced indices indexed by , and R2by L2equi-spaced indices indexed by . Let V(H) denote the vector of Matrix H, obtained by stacking columns of H one after another. In Matlab notation, V(H)=H(:).

[0145] The estimate of H is given as . The UE 302 feedback to the base station 304, , , R1, R2, k1, k2. R1, R2are estimated by AIML techniques, and can estimate , .

[0146] In an embodiment herein, for estimating complex saclars for each cluster / region of NDFs:

[0147] Consider two matrices B1, B2of dimensions NRNC×k1, and NRNC×k1. Let themth NDF in R1correspond to andnth NDF in R2correspond to . Here 1≤m≤k1, 1≤n≤k2. B1, B2as follows.

[0148] Denote B (:,m) as themth column of Matrix B (as in Matlab). Define , .

[0149] , are calculated from A. The equation to calculate A is V(H)=BA. From this equation A is calculated as

[0150] A=(BHB)-1BHV(H).

[0151] This can extend the same to other dimensions like time and frequency, where channel is 1D, and can have 1D versions accordingly. This can extend to many regions clusters accordingly.

[0152] In an embodiment herein, for a proposed OCC design, from a frequency domain code, , and a time domain code , the CSI-RS module 316 creates , as depicted in FIG. 17.

[0153] For consistency, the CSI-RS module 316 adopts an RE mapping convention, as shown in FIG. 18. The OCC sequence w(4)is mapped first along the frequency domain, and then along the time domain. Embodiments herein reuse this convention subsequently.

[0154] Embodiments herein provide a brief overview of the proposed 6G OCC design. The CSI-RS module 316 groups one or ports CSI-RS ports into CDM groups comprised of contiguous or closely-located groups of REs. OCC is applied over each CDM group individually. An m-order OCC can be thought of as an m×m orthogonal matrix or unitary matrix, each row comprising of individual m-length orthogonal sequences from the OCC. Within each CDM group, the applied OCC is a Kronecker product of two same or different OCCs. By taking the Kronecker product of an m-order OCC (say A) and an n-order OCC (say B), embodiments herein create a new mn-order OCC . For simplicity of reference, embodiments herein may refer to A as the outer code, and B as the inner code subsequently.

[0155] For example, the CSI-RS module 316 creates a CDM4 OCC by taking a Kronecker product of two 2-order OCCs, say A and B, respectively. The OCC sequence is then denoted as a CDM4-FD2-TD2 sequence.

[0156] The above procedure can be applied recursively to generate OCCs of arbitrarily large orders.

[0157] For example, embodiments herein disclose a 6G CSI-RS OCC generation procedure. Embodiments herein cover all possible tweaks of the generated OCCs, particularly those covered by the properties enumerated below, which can be obvious to a person with ordinary skill in the art. FIGS. 19A, 19B, and 19C depict an example RE mapping in a contiguous time-frequency grid, a non-contiguous uniform time-frequency grid; and a non-contiguous non-uniform time-frequency grid respectively.

[0158] Property 0:

[0159] The procedure can start with any orthogonal matrix. In particular, embodiments herein use (scaled / unscaled) identity matrix, DFT matrix or Hadamard matrix or DCT matrix or Slepain-based matrix. The rows or columns of a unitary matric can be the OCC for each port / user. For example, using the identity matrix would give rise to a simple FDM- / TDM-based OCC.

[0160] Property 1 (RE distribution):

[0161] The CDM group comprises of REs that might be contiguous or non-contiguous. For non-contiguous REs, they might be uniformly or non-uniformly placed. They are considered as different embodiments. The OCC is deployed over REs of a CDM group.

[0162] Property 2 (Factor permutation):

[0163] Since the Kronecker product is non-commutative, both and will represent valid mn-order OCCs for an m-order OCC A and an n-order OCC B.

[0164] Property 3 (Column permutation):

[0165] An m-order OCC can be column-permuted in any order and the resulting matrix will still consist of orthogonal rows. Hence, it would represent another valid OCC. An m-order OCC with orthogonal rows will also have m orthogonal columns. Hence, transposed and conjugate-transposed variants of an m-order OCC are also considered. Given an m-order OCC A and an n-order OCC B, the mn-order OCC will change if either of A or B are column-permuted (generating A' and B', respectively). Such OCCs as are also covered. Anm-order OCC, where an m-length OCC can be deployed to m REs in m! ways and all these are considered as different embodiments. The m REs can be distributed in a rectangular region in time-frequency with different spacings between the REs. All are considered equivalent embodiments.

[0166] Property 4 (Density):

[0167] The CDM group separation can be determined by density parameter. However, the separation might be uniform or non-uniform. FIGS. 20A and 20B depict example figures for CDM group separation in a uniform time-frequency grid, and a non-uniform time-frequency grid respectively. Many similar or different CDM groups are repeated across time and frequency for a port / user.

[0168] Property 5 (Periodicity):

[0169] The CSI-RS signals can be transmitted on a periodic, semi-periodic or aperiodic basis, like in current 5G specification.

[0170] Property 6 (Recursive construction):

[0171] Embodiments herein discuss recursive constructions of OCCs for some example number of ports. For simplicity, embodiments herein consider a single CDM group comprising of contiguous REs, and a single CSI-RS occasion. A person skilled in the art will be able to see that properties 1, 3, 4 and 5 can help in defining many more unique OCCs.

[0172] Additionally, embodiments herein denote a code-sequence from an m-order OCC W as [W(0), W(1), ...,W(m-1)], i.e., with no explicit assumption on the particular sequence used in the constituent factor OCC.

[0173] 2-port and 3-port cases:

[0174] For 2-port and 3-port cases, it is easy to see that the OCC structures, as depicted in FIGS. 21A and 21B are possible. Embodiments herein have denoted the two 2-port OCCs as A and B, and the two 3-port OCCs as C and D.

[0175] 4-port cases:

[0176] For the 4-port case, the three 4-order OCC structures as depicted in FIGS. 22A, 22B, and 22C are possible: Note that the OCC in FIG. 22C above can be generated by taking a Kronecker product of the 2-port OCCs. Embodiments herein will use this technique for recursively generating OCC structures for higher number of ports.

[0177] 5-port cases:

[0178] For the 5-port case, it is easy to see that only two OCC structures are possible, one along frequency axis and the other along time axis, just like in the 2-port and 3-port cases.

[0179] 6-port cases:

[0180] For the 6-port case, FIGS. 23A-23D depict examples OCC structures. It is easy to see that the OCCs in FIGS. 23A and 23B are similar to the 5-port case, comprising of orthogonal sequences as per Property 0. OCC in FIG. 23C is the Kronecker product of the 2-port OCC, A, and the 3-port OCC, D. Thus, it can be denoted as . Similarly, the OCC in FIG. 23D can be denoted as , for the 2-port OCC, B, and the 3-port OCC, C. Next, we consider the two other OCCs, and . Both of them would resemble the OCCs in FIGS. 23A and 23B, the only difference being that they are generated via Kronecker product of the 2-port and 3-port OCCs.

[0181] Now using Property 2 (Kronecker product is not commutative), embodiments herein can create four more OCCs, , , and . Thus, embodiments herein generate a total of 10 OCCs, out of which 8 were generated through a recursive Kronecker product construction method.

[0182] Embodiments herein count the possible OCC combinations in the 6-port case. Embodiments herein use P(m) to denote an arbitrary m-order OCC. The tuple {P(m), P(n)} would denote all possible OCCs generated via and , for any two integers m and n. This will denote an instance of mn-order OCC, denoted by P(mn). If P(m) occupies m1×m2in time-frequency domain (can be any two domains like spatial-frequency etc) and P(n) occupies n1×n2in time-frequency domain, then P(mn) occupies m1n1×m2n2in time-frequency domain. Embodiments herein use N(m) to denote the total number of OCC structures possible for an m-order OCC. Thus, the number of 6-order OCCs can be calculated as per table 2. Note that the factor of 2 coming in the third column of table 2 is due to Property 2.

[0183]

[0184] FIG. 24 depicts an example of the Kronecker operation.

[0185] There could be duplicates in N(m). Nevertheless, it shows the huge possibilities with increase in no. of antennas. The rectangular region of REs can be greater than coherence time and coherence bandwidth (region can be time- and frequency selective), in which case OCC can be used as complex exponential sequences, skip some exponential sequences (like we skip cyclic shifts in SRS). In this case, the region of REs will be greater than the number of ports which will have to be orthogonal over that region.

[0186] Table 3 depicts the construction of the higher-order OCCs.

[0187]

[0188]

[0189] Property 7 (Decodability):

[0190] Embodiments herein disclose the decodability of the OCCs generated by the proposed procedure. In particular, embodiments herein consider OCCs A1, A2, B1and B2such that A1and A2have compatible dimensions, and B1and B2have compatible dimensions (they are matrices, generalized for rectangular region in time-frequency grid). Also, it is assumed that , and , where denotes the sum of all elements of X and denotes the Hadamard product (element-wise multiplication of two matrices). Consider the well-known identity on mixed Kronecker-Hadamard products:

[0191]

[0192] For any matrices A, B, C and D of compatible dimensions.

[0193] Also have .

[0194] For any two OCCs B1and B2with , and any two appropriately dimensioned matrices A and C,

[0195]

[0196] In a very similar manner, for any two OCCs A1and A2with , and any two appropriately dimensioned matrices B and D, it can be shown that:

[0197]

[0198] Using the above two facts, it can be clearly seen that whenever or .

[0199] Thus, it can be said that whenever constituent factor OCCs re used, the resulting Kronecker product is also a valid OCC, and hence, decodable due to the constituent OCCs being orthogonal.

[0200] In an embodiment herein, if the base station 304 transmits CSI-RS (pilots) in a subset of reduced number of antenna ports, out of a superset of antenna ports, then the UE 302 feedback the channel information to the base station 304 corresponding to the superset of antenna ports using neural networks. Alternately, the UE 302 or feedback module 314 feedback the channel for the subset of reduced number of antenna ports to the base station 304, and the base station 304 computes the channel for the entire superset of antenna ports using neural networks. Example of neural networks can include, but not limited to U-Net, Generative Adversarial Network (GAN), and LSTM.

[0201] FIG. 25 depicts a 2D layout of antenna ports and / or corresponding 2D spatial channel. As depicted, FIG. 25 indicates an Entire Region of Antenna Ports (ERAP) at the base station 304. It can also depict the 2D channel associated with the ERAP. There are four (an example, can be many more / less) Sub-Regions of Antenna Ports (SRAP). The SRAP transmit pilots such as CSI-RS. The UE 302 or the channel approximation module 312 estimates the 2D channel corresponding to these SRAPs, and from that using AI / ML or neural networks constructs the 2D channel associated with ERAP.

[0202] In an embodiment herein, FFT_SRAP is the FFT of the channel of SRAP. The FFTs of all SRAPs are fed to a neural network like UNet, LSTM, CNN, GAN etc. FFT_ERAP denotes the FFT of the channel of ERAP, and is the ground truth fed to the neural network. FFTs here could also mean padding with zeros in all dimensions and taking FFT for more number of points.

[0203] Using the above inputs and ground truth, the neural network is trained. FIG. 26 depicts training of the neural network using the FFT SRAPs, and the ground truth. Alternately, the feedback module 314 feedback to the base station 304, only the channels associated with the SRAPs, and the base station 304 constructs the entire channel associated with ERAP. This reduces pilot overhead and UE feedback.

[0204] The trained neural network can now predict the FFT-ERAP from the many FFT_SRAPs, from which the estimate of the channel corresponding to ERAP can be arrived at using 2D Inverse Fast Fourier Transform (IFFT). FIG. 27 depicts predicting FFT-ERAP from the FFT_SRAPs,

[0205] In an embodiment herein, if the UE 302 transmits Sounding Reference Signals (SRS) (pilots) in a subset of subcarriers (sub-bands), out of a superset of subcarriers (entire frequency band), then the base station 304 computes the channel information corresponding to the superset of subcarriers (entire frequency band) using neural networks.

[0206] FIG. 28 depicts an entire frequency band of SRS (EFB_SRS) along with sub-band SRSs. FIG. 28 can also depict the 1D channel associated with the EFB_SRS. There are four (can be more or less) sub-bands of SRS (SB_SRS). The sub-bands of SRS transmit pilots SRS. The base station 304 estimates the 1D channel corresponding to the SB_SRS, and from that using AI / ML or neural networks constructs the 1D channel associated with EFB_SRS. This reduces SRS overhead as SRS is transmitted only in sub-bands, and not entire frequency band.

[0207] In an embodiment herein, FFT_SB_SRS is the FFT of the channel of sub-bands SB_SRS. The FFTs of all SB_SRS are fed to a neural network like UNet, LSTM, CNN, GAN etc. FFT-EFB_SRS denotes the FFT of the channel of EFB_SRS and is the ground truth fed to the neural network. Using the FFTs of all SB_SRS, and ground truth, the neural network is trained, as depicted in FIG. 29. In FFT one could pad zeros and get higher number of FFt bins as well.

[0208] The trained neural network can now predict the FFT-EFB_SRS from many FFT_SB_SRSs, as depicted in FIG. 30, from which the estimate of the channel corresponding to EFB_SRS can be arrived at using 1D IFFT.

[0209] 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 may include blocks which can be at least one of a hardware device, or a combination of hardware device and software module.

[0210] The embodiment disclosed herein describes methods 400 and systems 300 for providing CSI feedback in wireless communication networks using AIML, and for designing CSI-RS for 5G, and beyond 5G systems. Therefore, it is understood that the scope of the protection is extended to such a program and in addition to a computer readable means having a message therein, such computer readable storage means contain program code means for implementation of one or more steps of the method, when the program runs on a server or mobile deviceor any suitable programmable device. The method is implemented in at least one embodiment through or together with a software program written in e.g., Very high speed integrated circuit Hardware Description Language (VHDL) another programming language, or implemented by one or more VHDL or several software modules being executed on at least one hardware device. The hardware device can be any kind of portable device that can be programmed. The device may also include means which could be e.g., hardware means like e.g., an ASIC, or a combination of hardware and software means, e.g., an ASIC and an FPGA, or at least one microprocessor and at least one memory with software modules located therein. The method embodiments described herein could be implemented partly in hardware and partly in software. Alternatively, the invention may be implemented on different hardware devices, e.g., using a plurality of CPUs.

[0211] The foregoing description of the specific embodiments will so fully reveal the general nature of the embodiments herein that others can, by applying current knowledge, readily modify and / or adapt for various applications such specific embodiments without departing from the generic concept, and, therefore, such adaptations and modifications should and are intended to be comprehended within the meaning and range of equivalents of the disclosed embodiments. It is to be understood that the phraseology or terminology employed herein is for the purpose of description and not of limitation. Therefore, while the embodiments herein have been described in terms of embodiments and examples, those skilled in the art will recognize that the embodiments and examples disclosed herein can be practiced with modification within the scope of the embodiments as described herein.

Claims

1.A method performed by a user equipment (UE) for providing Channel State Information (CSI) feedback in a wireless communication network, the method comprising:separating at least one channel into one or more clusters;determining at least one region of one or more Normalized digital frequencies (NDFs) for each cluster, an associated complex scalar, and a number of approximated NDFs in each cluster;approximating the at least one channel in at least one of a time domain, a frequency domain, and a spatial domain using the at least one determined region of the one or more NDFs; andreporting, to a base station, the at least one determined region of the one or more NDFs, the associated complex scalar, the number of approximated NDFs in each cluster, and the at least one approximated channel.2.The method of claim 1, further comprising:transforming the at least one channel into a transform domain using at least one transform method for obtaining a matrix of values,wherein the at least one transform method comprises at least one of a two-dimensional (2D) Discrete Fourier Transform (DFT) (2D DFT), a 2D oversampled DFT, a 2D Slepian, a 2D Discrete Cosine Transform (DCT), a 1D DFT, a 1D oversampled DFT, a 1D Slepian, and a 1D DCT.3.The method of claim 1, wherein determining the at least one region of one or more NDFs for each cluster comprises:identifying location of peaks of each cluster of the at least one channel in at least one of the 1D oversampled DFT, and the 2D oversampled DFT;obtaining a cluster submatrix around each peak of each cluster in at least one of the 1D DFT, and the 2D DFT; andfeeding the cluster submatrix, and the location of peaks of each cluster to a neural network for determining co-ordinates of the at least one region of the one or more NDFs for each cluster.4.The method of claim 1, wherein determining the at least one region of one or more NDFs for each cluster comprises:upconverting at least one of a subset of 1D Slepian basis, and a subset of 2D Slepian basis of each cluster using at least one of a 1D frequency, and a 2D frequency for obtaining at least one of a 1D upconverted frequency, and a 2D upconverted frequency, wherein the at least one of the 1D frequency, and the 2D frequency corresponds to the location of peaks of each cluster;collecting the at least one of the 1D upconverted frequency, and the 2D upconverted frequency of the one or more clusters into a matrix;projecting the at least one channel onto the matrix for obtaining one or more Slepian coefficients of each cluster; andfeeding a submatrix of the one or more Slepian coefficients, the at least one of the 1D upconverted frequency, and the 2D upconverted frequency of each cluster to the neural network for obtaining co-ordinates of the at least one region of the one or more NDFs for each cluster.5.The method of claim 1, wherein the UE (302) approximates the at least one channel using at least one Artificial Intelligence (AI) model, wherein approximating the at least one channel comprises:equi-spacing the one or more NDFs in the at least one region;quantizing a scaling factor associated with at least one of each cluster, and the at least one region of the one or more NDF; andapproximating the channel with the one or more equi-spaced NDFs in each cluster, and the quantized scaling factor.6.The method of claim 1, further comprising:receiving, from the base station, a Channel State Information-Reference Signal (CSI-RS) in a subset of reduced number of antenna ports, out of a superset of antenna ports; andfeedback, to the base station, at least one channel information corresponding to at least one of the subset of reduced number of antenna ports, and the superset of antenna ports using a neural networks, on receiving the CSI-RS.7.The method of claim 1, further comprising: transmitting, to the base station, at least one Sounding Reference Signal (SRS) indicating the at least one channel information, in a subset of subcarriers out of a superset of subcarriers.8.A User Equipment (UE) comprising:a processor (306); anda memory module (310),wherein the processor (306) is coupled with the memory module (310), and is configured to:separate at least one channel into one or more clusters;determine at least one region of one or more Normalized digital frequencies (NDFs) for each cluster, an associated complex scalar, and a number of approximated NDFs in each cluster;approximate the at least one channel in at least one of a time domain, a frequency domain, and a spatial domain using the at least one determined region of the one or more NDFs; andreport, to the base station, the at least one determined region of the one or more NDFs, the associated complex scalar, the number of approximated NDFs in each cluster, and the at least one approximated channel.9.The UE of claim 8, wherein the processor (306) is further configured to transform the at least one channel into a transform domain using at least one transform method for obtaining a matrix of values, wherein the at least one transform method comprises at least one of a two-dimensional (2D) Discrete Fourier Transform (DFT) (2D DFT), a 2D oversampled DFT, a 2D Slepian, a 2D Discrete Cosine Transform (DCT), a 1D DFT, a 1D oversampled DFT, a 1D Slepian, and a 1D DCT.10.The UE of claim 8, wherein the processor (306) is configured to:identify location of peaks of each cluster of the at least one channel in at least one of the 1D oversampled DFT, and the 2D oversampled DFT;obtain a cluster submatrix around each peak of each cluster in at least one of the 1D DFT, and the 2D DFT; andfeed the cluster submatrix, and the location of peaks of each cluster to a neural network for determining co-ordinates of the at least one region of the one or more NDFs for each cluster.11.The UE of claim 8, wherein the processor (306) is configured to:upconvert at least one of a subset of 1D Slepian basis, and a subset of 2D Slepian basis of each cluster using at least one of a 1D frequency, and a 2D frequency for obtaining at least one of a 1D upconverted frequency, and a 2D upconverted frequency, wherein the at least one of the 1D frequency, and the 2D frequency corresponds to the location of peaks of each cluster;collect the at least one of the 1D upconverted frequency, and the 2D upconverted frequency of the one or more clusters into a matrix;project the at least one channel onto the matrix for obtaining one or more Slepian coefficients of each cluster; andfeed a submatrix of the one or more Slepian coefficients, the at least one of the 1D upconverted frequency, and the 2D upconverted frequency of each cluster to the neural network for obtaining co-ordinates of the at least one region of the one or more NDFs for each cluster.12.The UE of claim 8, wherein the UE (302) approximates the at least one channel using at least one Artificial Intelligence (AI) model, wherein the processor (306) is configured to:equi-space the one or more NDFs in the at least one region; andquantize a scaling factor associated with at least one of each cluster, and the at least one region of the one or more NDFs; andapproximate the channel with the one or more equi-spaced NDFs in each cluster, and the quantized scaling factor.13.The UE of claim 8, wherein the processor (306) is further configured to:receive, from the base station, a Channel State Information-Reference Signal (CSI-RS) in a subset of reduced number of antenna ports, out of a superset of antenna ports, andfeedback, to the base station, at least one channel information corresponding to at least one of the subset of reduced number of antenna ports, and the superset of antenna ports using a neural networks, on receiving the CSI-RS.14.The UE of claim 8, wherein the processor (306) is further configured to transmit, to the base station, at least one Sounding Reference Signal (SRS) indicating the at least one channel information, in a subset of subcarriers out of a superset of subcarriers.

Images