Autoencoder-based CSI feedback for o-ran networks

Abstract

A method of optimizing an autoencoder-based channel state information (CSI) feedback in an Open Radio Access Network (O-RAN) system includes: providing a CSI generation module of the autoencoder in a user equipment (UE), the CSI generation module being configured to generate a CSI payload including a codeword representing characteristics of the estimated channel as output based on estimated channel information; providing a CSI controller module in an O-RAN distributed unit (O-DU), the CSI controller module being configured to i) receive the CSI payload, ii) forward the CSI payload as CSI feedback over a fronthaul interface to an O-RAN radio unit (O-RU), and iii) forward control information to the O-RU; and providing a CSI reconstruction module in the O-RU, the CSI reconstruction module being configured to receive the CSI payload and the control information, and generate reconstructed channel information.

Description

AUTOENCODER-BASED CSI FEEDBACK FOR O-RAN NETWORKSBACKGROUND OF THE DISCLOSURE1. Field of the Disclosure

[0001] The present disclosure is related to Open Radio Access Network (O-RAN) systems, and relates more particularly to autoencoder-based Channel State Information (CSI) feedback for O-RAN systems.2. Description of Related Art

[0002] In Uplink Performance Improvement (ULP1) functional split, which has been approved by O-RAN alliance Workgroup 4, the uplink DMRS-based beamforming function and, in one of the two variants, equalization function are relocated from O-DU to the O-RU (the beamforming function and the equalization function are collectively referred to as DMRS-BF method). This DMRS-BF method leverages the received DMRS signal at the antenna level to generate the beamforming weights for the uplink PUSCH channel. Compared to the traditional Weight-based Dynamic Beamforming (WDBF) method, in which the beamforming weights are generated only based on SRS signal at the antenna level in the DU, DMRS-BF can adapt to the channel variation more timely and potentially with higher number of spatial streams than the fronthaul bandwidth allows, thus enabling the DMRS-BF to outperform the traditional WDBF in high mobility and high interference scenarios.

[0003] Most recently, there have been proposals by the O-RAN WG4 participants to additionally relocate the SRS channel estimation to the O-RU. Since the SRS can be configured to cover the entire carrier bandwidth, with SRS CE moving to O-RU, the O-RU shall have access to the full bandwidth channel estimation at the antenna level. It is conceivable that downlink beamforming weights can be generated in the O-RU as well as in a TDD system by leveraging the SRS and DMRS channel estimation. As a result, the entire beamforming generation operation can take place in the O-RU. We believe that this is the most likely technology evolution path for O-RAN WG4 going forward.

[0004] Under this context, it can be beneficial to introduce the channel state information feedback from the UE to the 0-RU. In this situation, the UE measures the channel state information according to a reference signal (e.g., CSI-RS) transmitted by the network, and reports it to the network via the uplink PUCCH or PUSCH channels. Currently, 3GPP supports CSI feedback based on various types of CSI-RS codebooks such as type I, type II, and enhanced type II CSI-RS with different triggering conditions (periodic, semi-persistent, and aperiodic). However, CSI-RS-based CSI feedback has a lower spatial resolution due to the over-sampled DFT beam and limited bandwidth for the PMI report. To improve the accuracy of the CSI feedback, starting from Release 18, AI / ML-based CSI feedback has been identified as one of the key study items for machine learning (ML) for the next-generation wireless networks by 3 GPP and has attracted significant interest from the industry. The intention is to send the UE-observed channel state information directly back to the network without the drawback of high granularity and low resolution of CSI-RS-based codebook.

[0005] To perform AI / ML-based CSI feedback, the consensus of the 3GPP participants is to adopt the so-called two-sided AI / ML model. An example of the two-sided AI / ML model is the auto-encoder architecture shown in FIG. 1, where the CSI compression part is implemented by the “encoder” 10 functionality module (which is on the “generation” side shown on the left portion of FIG. 1), and the CSI decompression part is implemented by the “decoder” 11 functionality module (which is on the “reconstruction” side shown on the right portion of FIG.1). The “encoder” 10 compresses the channel state information (CSI) (represented as X in FIG.1) into the low-dimensional latent space (denoted as low-dimensional representation “Z” 12), and the “decoder” 11 decompresses and recovers the CSI at the destination (where the recovered CSI is represented as X’ in FIG. 1).

[0006] Autoencoder-based channel feedback outperforms the traditional CSI-RS-based channel feedback as it attempts to directly send the downlink channel or some representation of the downlink channel (e.g., channel covariance or precoding vectors), which can overcome the shortcoming of the low spatial resolution of the traditional codebook based beamforming. The existing Open RAN architecture already supports legacy CSI feedback using the CSI-RS codebook, either in Predefined-beam beamforming (PDBF) or Weight-based dynamic beamforming(WDBF). But the system architecture of using an autoencoder for CSI feedback hasnot been addressed in the O-RAN architecture, especially for DMRS-BF split.

[0007] One clear drawback of DMRS-BF is the coverage imbalance between the downlink and the uplink. More specifically, the uplink is power-limited due to UE transmit power limitation, and as a result, the coverage of the uplink reference signal (e.g., DMRS) is much smaller than that of the downlink reference signal (e.g., CSI-RS). At the cell edge, the uplink DMRS-based beamforming may no longer be accurate and may underperform the codebook-based beamforming. Another drawback of the DMRS-BF is that it is not applicable to FDD system, in which the channel reciprocity can not be established. There has been interest in doing mMIMO in FDD, in this case, in order to do downlink beamforming in the 0-RU, it is apparent that the downlink CSI would be needed in the O-RU.

[0008] Accordingly, there is a need for a system and a method of optimizing the autoencoderbased CSI feedback in the O-RAN system to improve O-RAN network performance.SUMMARY

[0009] Accordingly, what is desired is a system and a method of optimizing the autoencoderbased CSI feedback in the O-RAN system to improve O-RAN network performance.

[0010] According to an example system and / or method of the present disclosure, in a scenario where the uplink DMRS-based beamforming may no longer be accurate and may underperform the codebook-based beamforming, e.g., in a scenario at the cell edge, O-RU is provided access to the autoencoder-based CSI feedback from the UE, thereby enabling higher cell edge performance and better spectrum efficiency.

[0011] According to an example system and / or method of the present disclosure, for use cases where channel reciprocity cannot be leveraged (e.g., FDD for implementing mMIMO), CSI feedback is provided to the O-RU in order to implement downlink beamforming in the O-RU.

[0012] According to an example system and / or method of the present disclosure, a new fronthaul functional split is provided to enable autoencoder-based CSI feedback in an O-RAN network, wherein the CSI reconstruction part can reside in the O-RU for the channel state information recovery.

[0013] According to an example embodiment of the method, the following steps are implemented: receiving, by the 0-DU from the user equipment (UE), the compressed channel state information in the low dimensional latent space; sending, by the O-DU, the latent space data to the O-RU over the open fronthaul interface; and recovering, by the CSI reconstruction part which can reside in the O-RU, the channel state information.

[0014] According to an example embodiment of the method, the following steps are further provided: sending, by the O-DU over the open fronthaul interface to the CSI reconstruction part, the control parameters related to the model configuration, performance target, monitoring information, life cycle management (LCM) information, and / or inter-vendor training and collaboration.

[0015] For this application, the following terms and definitions shall apply:

[0016] The term “computer” as used herein refers to a device having one or more general or special purpose processors, memory, storage, and networking components (either wired or wireless), which device can execute an operating system, e.g., a Microsoft® Windows®-compatible operating system (OS), Apple® OS X or iOS, a Linux® distribution, or Google® Android OS.

[0017] The term “network” as used herein includes both networks and internetworks of all kinds, including the Internet, and is not limited to any particular type of network or inter-network.

[0018] The terms “first” and “second” are used to distinguish one element, set, data, object or thing from another, and are not used to designate relative position or arrangement in time.

[0019] The terms “coupled”, “coupled to”, “coupled with”, “connected”, “connected to”, and “connected with” as used herein each mean a relationship between or among two or more devices, apparatus, files, programs, applications, media, components, networks, systems, subsystems, and / or means, constituting any one or more of (a) a connection, whether direct or through one or more other devices, apparatus, files, programs, applications, media, components, networks, systems, subsystems, or means, (b) a communications relationship, whether direct or through one or more other devices, apparatus, files, programs, applications, media, components,networks, systems, subsystems, or means, and / or (c) a functional relationship in which the operation of any one or more devices, apparatus, files, programs, applications, media, components, networks, systems, subsystems, or means depends, in whole or in part, on the operation of any one or more others thereof.

[0020] The above-described and other features and advantages of the present disclosure will be appreciated and understood by those skilled in the art from the following detailed description, drawings, and appended claims.DESCRIPTION OF THE DRAWINGS

[0021] FIG. l is a block diagram illustrating an example of a conventional system including the UE and the wireless network utilizing autoencoder for CSI compression feedback.

[0022] FIG. 2 is a block diagram illustrating an example embodiment of a system according to the present disclosure, in which system the encoder in the UE works with the decoder in the O-RU over the open fronthaul interface.

[0023] FIG. 3 is a block diagram illustrating an example embodiment of a system according to the present disclosure, in which system the 0-DU is configured to send model configuration message over the open fronthaul to support the decoder in the 0-RU.

[0024] FIG. 4 is a block diagram illustrating an example embodiment of a system according to the present disclosure, in which system the 0-DU is configured to send an operation control message over the open fronthaul to support the decoder in the 0-RU.

[0025] FIG. 5 is a block diagram illustrating an example embodiment of a system according to the present disclosure, in which system the O-DU is configured to send performance monitoring message over the open fronthaul to support the decoder in the 0-RU.

[0026] FIG. 6 is a block diagram illustrating an example embodiment of a system according to the present disclosure, in which system the 0-DU is configured to send training control messages over the fronthaul to support inter-vendor collaboration of the decoder in the 0-RU over the fronthaul.0017506WGU / 4688DETAILED DESCRIPTION

[0027] In massive multi-input multi -output (MIMO) systems, it is highly critical for transceivers to obtain accurate channel state information (CSI) to improve spectrum efficiency and system capacity. In the time division duplex (TDD) system, the base station can obtain the CSI through a reference signal transmitted by the user equipment (UE) by leveraging the channel reciprocity, such reference signals can be SRS or DMRS signals. However, in the frequency division duplex (FDD) system, the downlink and uplink channels exhibit different channel fading characteristics due to their operation in different carrier frequencies. Consequently, the UE must provide the base station with feedback of the downlink CSI through the uplink channel. To address this issue, 3GPP has approved codebook-based beamforming using downlink CSLRS signals. For example, the eType II codebook compresses the downlink CSI based on sparsity in the spatial and frequency domains.

[0028] To further enhance CSI recovery accuracy and reduce feedback overhead, a machine learning (ML) based autoencoder has been adopted by 3 GPP as a study item, and it has been shown that the autoencoder demonstrates superior performance over traditional codebook-based (e.g., eType II codebook) schemes. A two-sided AI / ML model for CSI feedback has been adopted by 3GPP, in which model the UE side performs the compression of the CSI information, and the Network side performs the decompression to recover the CSI information. These two sides are also referred to as the CSI “generation” part and CSI “reconstruction” part, respectively. In the autoencoder model, channel state information (CSI) can be represented in various formats and is sent to the network directly. For example, the autoencoder can take the downlink channel estimation or precoding matrix in the spatial-frequency domain in a subband granularity as the input, and provide the reconstructed channel estimation or precoding matrix in the spatial -frequency as the output.

[0029] Further studies have been conducted on means to improve the trade-off between performance, proprietary protection, and manageable overhead, such as extending from the spatial / frequency domain CSI compression to spatial / temporal / frequency CSI compression; to extend from generic Al models to cell / site-specific models and to extend from the CSI compression only to CSI compression plus prediction models. The target of complexityreduction may be measured by the model size and / or FLOPs required, overhead bits needed for signaling from the spatial / temporal / frequency basis, and the like, when compared with baseline model through, for example, squared generalized cosine similarity (SGCS) performance comparison. An inference model that has low latency is always beneficial for reacting quickly to ever-changing channel conditions.

[0030] In addition, to facilitate inter-vendor training for the two-sided model, solutions that support parameter / dataset / model exchange between the UE side and the network (NW) side are required, and the standardized over-the-air signaling, such as RRC signaling, can be considered.

[0031] In the 0-RAN networks, the network side comprises disaggregated network function (NF) components with standardized interfaces such that such components from different vendors can work seamlessly together. One such interface is the fronthaul interface, which connects the O-DU and O-RU together. The present disclosure provides that in the autoencoder-based CSI feedback architecture in the 0-RAN network, the CSI reconstruction part can reside in the O-RU, whereas the O-DU can be responsible for forwarding the output of the CSI generation part to the O-RU and managing the configuration, Life Cycle Management (LCM), training and performance monitoring of the CSI reconstruction part in the O-RU over the fronthaul interface. This method is beneficial in situations where channel reciprocity is not possible (e.g., a FDD system) or where the quality of CSI derived from the UE-transmitted reference signal (e.g., SRS and / or DMRS) degrades due to the power-limited nature of the uplink.

[0032] An example embodiment of a system architecture according to the present disclosure is shown in FIG. 2. As shown in the block diagram of FIG. 2, which illustrates a system comprising UE 20, O-DU 21 and O-RU 22 in a network, the CSI generation module 102 (also referred to as an “encoder”, since the module 102 encompasses an encoding portion) is located in the UE 20, which CSI generation module 102 utilizes the estimated channel 101 as input to generate the structured payload 103. The estimated channel 101 can be represented as a pre-processed channel matrix, a covariance channel matrix, or a precoding matrix. An example CSI generation module 102 (encoder) can comprise several layers, e.g., the first layer of the encoder can be a convolutional layer with real and imaginary parts of the channel, after which the feature maps are reshaped into a vector and use a fully connected layer to generate the codewords, which is a real-valued vector of size M. The codeword captures the essential feature of the channel at a much lower dimension (i.e., at the latent space), with each of the dimensions representing specific characteristics of the input data. The codeword is then packaged into a structured payload 103 (e.g., a transport block, or a protocol data unit) that is compliant with 3GPP, e.g., as an LI message or an L3 message, and is subsequently transmitted to the base station (e.g., to 0-DU 21) over the air interface.

[0033] The base station (e.g., 0-DU 21) receives and decodes the structured payload (also referred to as CSI payload) to retrieve the codeword as part of the decoded payload 104. The CSI Controller module 105 of the 0-DU 21 forwards the codeword (e.g., as part of the CSI feedback and / or control 23) over the fronthaul 106 to the CSI reconstruction module 107 (also referred to as a “decoder”, since the module 107 encompasses a decoding portion) in the O-RU 22. At the CSI reconstruction module 107, the codeword of size M is given to the decoding portion, wherein the first layer of the decoder may be a dense fully-connected layer which gives the initial estimate of the channel, followed by several refinement network (“refmenef ’) units to recover the reconstructed CSI (as represented by recoonstructed channel 108). It is advantageous to send the codeword over the fronthaul 106 as it helps conserve fronthaul bandwidth, compared to the conventional technique of performing the CSI reconstruction part in the 0-DU and then sending the reconstructed CSI over the fronthaul to the O-RU.

[0034] According to the example embodiment shown in FIG. 2, the CSI Controller 105 in the O-DU 21 can be responsible for the configuration, life-cycle management, and / or training of the CSI reconstruction module 107 in the O-RU 22, by communicating over the open FH interface 106.

[0035] According to an example system and / or method of the present disclosure shown in FIG.3, the CSI Controller 105 in the 0-DU 21 can send a model configuration message 301 to the CSI Reconstruction module 107 in the O-RU 22. In a two-sided autoencoder deployment, it is crucial to align the model configuration between the CSI generation module 102 (encoder) in the UE 20 and the CSI reconstruction module 107 (decoder) in order to achieve the desired performance. FIG. 3 shows that the CSI Controller module 105 in the 0-DU 21 can act as a proxy for the CSI reconstruction module 107 in the O-RU 22. The CSI Controller 105 can i)0017506WGU / 4688interwork with the CSI generation module 102 in the UE 20 through LI or L3 messaging, ii) determine the proper configuration parameters, and iii) send the model configuration message 301 to the CSI reconstruction module 107 over the open fronthaul interface. Such configuration parameters can include, but are not limited to, the following:1. Model backbone type, e.g., whether it is CNN-based or Transformer based.2. Hyperparameters of the model: For CNN, it can include kernel size, stride, and padding in each convolution operation, dimension of feedforward latent space, and the like. For Transformer, it can include the number of blocks, embedding length, hyper-parameters in selfattention, feedforward blocks.3. The quantization and / or dequantization approaches of CSI feedback, e.g., scaler and vector quantizers for the generation of the CSI-feedback bits.4. The associated model ID: Assistant information that is associated with the exchanged dataset, model and / or parameter (e.g., association ID, model ID and / or pairing ID).5. The data format and / or data type of input / output of the autoencoder.6. Performance target or indicator, e.g., squared generalized cosine similarity (SGCS) target.

[0036] According to an example system and / or method of the present disclosure shown in FIG.4, the CSI controller module 105 in the O-DU 21 can send operation control parameters 401 of the CSI reconstruction module 107 in the O-RU 22 over the fronthaul interface. In the example configuration of FIG. 4, the CSI controller module 105 in the O-DU 21 acts as a proxy and interworks with the CSI generation module 102 in the UE 20 and the L2 / L3 network functions in the O-DU 21 and / or the O-CU to determine the proper operation parameters and send an operational control message to the CSI reconstruction module 107 over the open fronthaul interface. Such operation parameters can include, but are not limited to, the following:1. Model selection function.2. Model activation / deactivation function.0017506WGU / 46883. CSI buffer length potentially adapting to the UE speed.4. CSI buffer reset function to address misalignment of historical CSI used at the UE side and the network (NW) side.5. Details of CSI reporting mechanism (periodic, semi-persistent, or aperiodic) and trigger conditions.6. Configuration of rank / layer, number of sub-bands.7. Latency requirement.

[0037] According to an example system and / or method of the present disclosure shown in FIG.5, the CSI controller module 105 in the 0-DU 21 can implement performance monitoring of the autoencoder model. In an example embodiment, the CSI controller module 105 can implement performance monitoring of the autoencoder model, e.g., to check for performance degradation against ground-truth CSI or target CSI, by sending a performance monitoring message 501 to the CSI reconstruction module 107 in the 0-RU 22 over the fronthaul interface. In an example scenario, the network may need to detect whether the current model is no longer suitable for the deployed network environment due to a mismatch of the data distribution of the UE side and the network side or due to data drift. The ground truth can be the channel state information obtained from a non-AI method, e.g., a codebook-based feedback and / or SRS-based channel estimation.

[0038] In the example configuration of FIG. 5, the CSI controller module 105 can work with the UE 20 and manage the performance monitoring of the CSI reconstruction module 107 in the O-RU 22. Performance degradation can be characterized by significant changes in statistics and / or semantics of measured CSI. The CSI controller module 105 in the 0-DU 21 can send control information related to performance monitoring to the CSI reconstruction module 107 in the O-RU 22 over the fronthaul interface, which control information can include, but are not limited to, the following:1. The activation and deactivation function of performance monitoring.2. The source of the ground truth (e.g., from a high-resolution codebook such as eTypell, or from SRS, UL DMRS, and the like).3. Instructions to isolate whether the performance degradation is due to UE-side CSI generation part, NW-side CSI reconstruction part, or data drift, and the like. This step can involve activating a local reference model of the CSI generation part in the O-DU or O-RU so that a known compression output can be generated for problem identification.4. Information related to a mitigation measure, such as changing to a different model ID or falling back to the legacy CSI feedback between the O-RU and the UE.

[0039] It is advantageous for the performance monitoring that one kind of ground truth based on channel reciprocity (e.g., SRS and / or DMRS sent by the UE) is easily accessible in the O-RU at the antenna or virtual antenna level.

[0040] According to an example system and / or method of the present disclosure shown in FIG.6, the CSI controller module 105 in the O-DU 21 provides the training data to the CSI reconstruction module 107 in the O-RU 22 via the fronthaul interface, e.g., to facilitate intervendor training collaborations. In a multi-vendor ecosystem such as a cellular wireless communications network, the CSI generation module 102 can be provided by a large number of UE manufacturers, and the CSI reconstruction module 107 can also be provided by several network component manufacturers. Accordingly, there is significant overhead for the network side to simultaneously support a large number of UEs with different model structures and parameters. Methods have been proposed to reduce inter-vendor collaboration complexity and improve feasibility for UE to use received parameters directly for inference, e.g., a proposal that at least the model parameter quantization and input data pre-processing should be standardized together with the CSI generation model structure. From the network side, using one or more standardized reference models for the UE CSI-generation module in the network side to train and / or validate the CSI reconstruction module becomes necessary to improve inter-vendor collaboration.

[0041] As shown in FIG. 6, the CSI controller module 105 in the O-DU 21 can send the training, testing, and / or validation data in a message (collectively referenced as “training control” message 601) over the fronthaul to the CSI reconstruction module 107 in the O-RU 22 to facilitate NW-side decoder training, validation and testing. In an example embodiment, the reference CSI-generation model resides in the O-DU or in the O-RU. The information related tothe inter-vendor training can include, but not limited to, the following:1. The number of samples in a dataset, the number of datasets, and how often the dataset needs to be transferred.2. Training dataset or information related to collecting training dataset.3. Testing dataset or information related to collecting testing dataset.4. Additional dataset for performance testing.5. Enable / disable ground-truth reporting, type of ground-truth CSI.6. Information related to the loss functions.7. Performance requirements, end-to-end ones (e.g., SGCS thresholds between target CSI and reconstructed CSI) or NMSE threshold between CSI feedbacks.8. Pairing information based on the conditions / additional conditions assigned to the samples of the datasets used for training of the model.

[0042] The example embodiments of the present disclosure provide a new functional mapping of the autoencoder-based CSI feedback in the context of the 0-RAN network, whereby the UE’s CSI generation module interworks with the CSI reconstruction module in the 0-RU over the fronthaul interface, and the O-DU manages the autodecoder’s parameters, the configurations, and the LCM aspects. In addition, the example embodiments of the present disclosure provide a number of new messages configured to be transmitted over the open fronthaul for the O-DU to support the CSI reconstruction module in the O-RU. Some of the advantages provided by the example embodiments of the present disclosure include:1) With autoencoder-based CSI reconstruction module in the O-RU, the cell edge performance of a TDD system can be improved. It is also useful for an FDD system where channel reciprocity cannot be established.2) The proposed architecture of the present disclosure aligns with the overall trend in the 0-RAN WG4 architecture migration of moving more functions into the O-RU and enabling full0017506WGU / 4688uplink (UL) and downlink (DL) beamforming inside the O-RU. For example, the SRS CE function is being proposed by WG4 participants to be moved to the O-RU.

[0043] While the present disclosure has been described with reference to one or more exemplary embodiments, it will be understood by those skilled in the art that various changes can be made and equivalents can be substituted for elements thereof without departing from the scope of the present disclosure. Although the example methods have been described in the context of O-RAN systems, the example methods are equally applicable to 6G and other wireless cellular networks. In addition, many modifications can be made to adapt a particular situation or material to the teachings of the disclosure without departing from the scope thereof. Therefore, it is intended that the present disclosure not be limited to the particular embodiment(s) disclosed as the best mode contemplated, but that the disclosure will include all embodiments falling within the scope of the appended claims.

[0044] For the sake of completeness, the following list of acronyms is provided:CE - Channel EstimationCNN - Convolutional Neural NetworkCSI-RS - Chanel State Information Reference SignalDMRS - Demodulation Reference SignalDMRS-BF - DMRS Based BeamformingFDD - Frequency Division DuplexFH - Fronthaul SplitFLOP - Floating Point OperationLCM - Life Cycle ManagementNMSE - Nonnalized Mean Square ErrorO-CU - 0-RAN Central UnitO-DU - 0-RAN Distributed Unit0-RAN - Open Radio Access NetworkO-RU - 0-RAN Radio UnitPDBF - Predefined BeamformingPUCCH - Physical Uplink Control ChannelPUSCH- Physical Uplink Shared ChannelSGCS - Squared Generalized Cosine SimilaritySRS - Sounding Reference SignalTDD - Time Division DuplexULPI - Uplink Performance Improvement WDBF - Weight-based Dynamic Beamforming WG4 - Work Group 4

Claims

CLAIMS:

1. A method of optimizing an autoencoder-based channel state information (CSI) feedback in an Open Radio Access Network (O-RAN) system, the method comprising:providing a CSI generation module of the autoencoder in a user equipment (UE), wherein the CSI generation module is configured to generate a CSI payload as output based on estimated channel information;providing a CSI controller module in an O-RAN distributed unit (O-DU), wherein the CSI controller module is configured to i) receive the CSI payload, ii) forward the CSI payload as CSI feedback over a fronthaul interface to an O-RAN radio unit (O-RU), and iii) forward control information to the O-RU; andproviding a CSI reconstruction module in the O-RU, wherein the CSI reconstruction module is configured to receive the CSI payload and the control information, and generate reconstructed channel information.

2. The method of claim 1, wherein:the CSI payload generated by the CSI generation module is a structured payload comprising a codeword representing characteristics of the estimated channel.

3. The method of claim 2, wherein:the structured payload is one of a transport block or a protocol data unit.

4. The method of claim 2, wherein:the O-DU receives the structured payload and decodes the codeword;the codeword is forwarded by the CSI controller module of the O-DU to the CSI reconstruction module of the O-RU; andthe CSI reconstruction module generates the reconstructed channel information based on at least the codeword.

5. The method of claim 4, wherein:the CSI controller module of the O-DU forwards control information comprising model configuration parameters for the CSI reconstruction module over the fronthaul interface to the CSI reconstruction module.

6. The method of claim 4, wherein:the CSI controller module of the O-DU forwards control information comprising operation control parameters for the CSI reconstruction module over the fronthaul interface to the CSI reconstruction module.

7. The method of claim 4, wherein:the CSI controller module of the O-DU forwards control information comprising control information related to performance monitoring of the autoencoder over the fronthaul interface to the CSI reconstruction module.

8. The method of claim 4, wherein:the CSI controller module of the O-DU forwards control information comprising at least one of training, testing, and validation data over the fronthaul interface to the CSI reconstruction module.

9. A system for optimizing an autoencoder-based channel state information (CSI) feedback in an Open Radio Access Network (O-RAN) system, comprising:a CSI generation module of the autoencoder provided in a user equipment (UE), wherein the CSI generation module is configured to generate a CSI payload as output based on estimated channel information;a CSI controller module provided in an O-RAN distributed unit (O-DU), wherein the CSI controller module is configured to i) receive the CSI payload, ii) forward the CSI payload as CSI feedback over a fronthaul interface to an O-RAN radio unit (O-RU), and iii) forward control information to the O-RU; anda CSI reconstruction module provided in the O-RU, wherein the CSI reconstruction module is configured to receive the CSI payload and the control information, and generatereconstructed channel information.

10. The system of claim 9, wherein:the CSI payload generated by the CSI generation module is a structured payload comprising a codeword representing characteristics of the estimated channel.

11. The system of claim 10, wherein:the 0-DU is configured to receive the structured payload and decodes the codeword; the codeword is forwarded by the CSI controller module of the O-DU to the CSI reconstruction module of the 0-RU; andthe CSI reconstruction module is configured to generate the reconstructed channel information based on at least the codeword.

12. The system of claim 11, wherein:the CSI controller module of the O-DU is configured to forward control information comprising model configuration parameters for the CSI reconstruction module over the fronthaul interface to the CSI reconstruction module.

13. The system of claim 11, wherein:the CSI controller module of the O-DU is configured to forward control information comprising operation control parameters for the CSI reconstruction module over the fronthaul interface to the CSI reconstruction module.

14. The system of claim 11, wherein:the CSI controller module of the O-DU is configured to forward control information comprising control information related to performance monitoring of the autoencoder over the fronthaul interface to the CSI reconstruction module.

15. The system of claim 11, wherein:the CSI controller module of the O-DU is configured to forward control information comprising at least one of training, testing, and validation data over the fronthaul interface to the CSI reconstruction module.

Images