Ai / ml based channel state information compression

Abstract

Example embodiments of this disclosure relate to channel state information (CSI) compression based on artificial intelligence (AI) / machine learning (ML). In one method, a first device determines a CSI report based on the CSI and according to a first model at the first device. The CSI report includes compressed CSI. The first device transmits the CSI report to a second device. If at least one condition is met, the first device retransmits the CSI report to the second device; or updates the CSI and the CSI report, and transmits the updated CSI report. The at least one condition includes: a first condition and / or a second condition, wherein the first condition is that a message from the second device indicates that the performance metric for channel decoding and / or decompression of the CSI report is less than a threshold, and the second condition is that the evaluated performance metric for channel decoding and / or decompression of the CSI report is less than the threshold.

Description

Cross-references to related applications

[0001] This application claims priority and interest in U.S. Provisional Application No. 63 / 751318, filed January 30, 2025, the entire contents of which are incorporated herein by reference. Technical Field

[0002] Various example embodiments of this disclosure generally relate to the telecommunications field, and more specifically to methods, apparatuses, devices, and computer-readable storage media for channel state information (CSI) compression based on artificial intelligence (AI) / machine learning (ML). Background Technology

[0003] A communication network can be used as a facility to enable communication between two or more communication devices, or to provide communication devices with access to a data network. A mobile or wireless communication network is an example of a communication network. Services can be provided to communication devices by an application server.

[0004] As communication networks and services increase in size, complexity, and the number of users, operations within these networks become increasingly complex. To improve communication performance, the use of AI / ML technologies in wireless communication networks has been proposed. AI / ML models can be applied to various communication functions in different scenarios, including but not limited to beam management (BM), CSI compression, and positioning. Summary of the Invention

[0005] In a first aspect of this disclosure, a first apparatus is provided. The first apparatus includes: at least one processor; and at least one memory, the at least one memory storing instructions that, when executed by the at least one processor, cause the first apparatus to: determine a CSI report based on a CSI and according to a first model at the first apparatus, the CSI report including compressed CSI; transmit the CSI report to a second apparatus; and, based on at least one condition being met, perform at least one of the following: retransmit the CSI report to the second apparatus; or update the CSI and the CSI report and transmit the updated CSI report, wherein the at least one condition includes at least one of the following: a first condition, the first condition being that a message from the second apparatus indicates that a performance metric for channel decoding and / or decompression of the CSI report is less than a threshold, or a second condition, the second condition being that an evaluated performance metric for channel decoding and / or decompression of the CSI report at the second apparatus is less than a threshold, the evaluated performance metric being evaluated by at least one of the following: the first model or evaluator at the first apparatus.

[0006] In a second aspect of this disclosure, a second apparatus is provided. The second apparatus includes: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the second apparatus to: receive a channel state information (CSI) report from a first apparatus, the CSI report including compressed channel state information; perform channel decoding and decompression on the CSI report, at least the decompression being based on a second model at the second apparatus; and transmit a first message to the first apparatus based on a performance metric for channel decoding or decompression of the CSI report being less than a threshold, the first message indicating that the performance metric for channel decoding and / or decompression of the CSI report is less than the threshold.

[0007] In a third aspect of this disclosure, a method is provided. The method includes: determining a CSI report based on a CSI and according to a first model at a first device, the CSI report including compressed CSI; transmitting the CSI report to a second device; and, based on at least one condition being met, performing at least one of the following: retransmitting the CSI report to the second device; or updating the CSI and the CSI report and transmitting the updated CSI report, wherein the at least one condition includes at least one of the following: a first condition, the first condition being that a message from the second device indicates that a performance metric for channel decoding and / or decompression of the CSI report is less than a threshold, or a second condition, the second condition being that an evaluated performance metric for channel decoding and / or decompression of the CSI report at the second device is less than a threshold, the evaluated performance metric being evaluated by a first model or evaluator at the first device.

[0008] In a fourth aspect of this disclosure, a method is provided. The method includes: receiving a channel state information (CSI) report from a first device, the CSI report including compressed channel state information; performing channel decoding and decompression on the CSI report, at least the decompression being based on a second model at a second device; and transmitting a first message to the first device based on a performance metric for channel decoding or decompression of the CSI report being less than a threshold, the first message indicating that the performance metric for channel decoding and / or decompression of the CSI report is less than the threshold.

[0009] In a fifth aspect of this disclosure, a first apparatus is provided. The first apparatus includes: components for determining a CSI report based on a CSI and according to a first model at the first apparatus, the CSI report including compressed CSI; components for transmitting the CSI report to a second apparatus; and components for performing at least one of the following based on at least one condition being met: retransmitting the CSI report to the second apparatus; or updating the CSI and the CSI report and transmitting the updated CSI report, wherein at least one condition includes at least one of the following: a first condition, the first condition being that a message from the second apparatus indicates that a performance metric for channel decoding and / or decompression of the CSI report is less than a threshold, or a second condition, the second condition being that an evaluated performance metric for channel decoding and / or decompression of the CSI report at the second apparatus is less than a threshold, the evaluated performance metric being evaluated by a first model or evaluator at the first apparatus.

[0010] In a sixth aspect of this disclosure, a second apparatus is provided. The second apparatus includes: means for receiving a channel state information (CSI) report from a first apparatus, the CSI report including compressed channel state information; means for performing channel decoding and decompression on the CSI report, at least the decompression being based on a second model at the second apparatus; and means for transmitting a first message to the first apparatus based on a performance metric for channel decoding or decompression of the CSI report being lower than a threshold, the first message indicating that the performance metric for channel decoding and / or decompression of the CSI report is lower than the threshold.

[0011] In a seventh aspect of this disclosure, a computer-readable medium is provided. The computer-readable medium includes instructions stored thereon for causing a device to perform at least the method according to a third or fourth aspect.

[0012] It should be understood that the summary portion is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0013] Some exemplary embodiments will now be described with reference to the accompanying drawings, in which: Figure 1 An example communication environment in which example embodiments of this disclosure may be implemented is shown; Figure 2A An example diagram of an eigenvalue compression autoencoder is shown; Figure 2B An example architecture against autoencoders is shown; Figure 2C An example diagram of an autoencoder with a self-supervised classifier is shown; Figure 2DAn example architecture of a classifier autoencoder is shown; Figure 2E An example diagram of a simplified architecture for CSI compression based on an autoencoder is shown; Figures 2F to 2G Example plots are shown for the correlation and normalized mean square error (NMSE) degradation of a 64-length codeword for a user equipment (UE) with two receive antennas. Figure 3 Example signaling streams for AI / ML-based CSI compression are shown according to some example embodiments of this disclosure; Figure 4 Example signaling flows for AI / ML model configuration and / or selection from options are shown according to some example embodiments of this disclosure; Figure 5 Example signaling flows for channel-aware CSI feedback are shown according to some example embodiments of this disclosure; Figure 6 Another example signaling flow for channel-aware CSI feedback is shown according to some example embodiments of this disclosure; Figure 7 Example signaling streams for AI / ML-based feedback information compression evaluation are shown according to some example embodiments of this disclosure; Figure 8 Example diagrams of AI / MLCSI feedback with conventional channel coding and modulation are shown according to some example embodiments of the present disclosure; Figure 9 Example diagrams of AI / ML CSI feedback with integrated channel coding according to some example embodiments of this disclosure are shown; Figure 10 Example diagrams are shown of an AI / ML decoder design with an evaluation of the decoding probability with compressed CSI feedback according to some example embodiments of the present disclosure; Figure 11 Example diagrams are shown of an AI / ML encoder design with an evaluation of the decoding probability with compressed CSI feedback according to some example embodiments of the present disclosure; Figure 12 Another flowchart of a method implemented at a first device according to some example embodiments of the present disclosure is shown; Figure 13 Another flowchart of a method implemented at a second device according to some example embodiments of the present disclosure is shown; Figure 14 Another flowchart of a method implemented at a first device according to some example embodiments of the present disclosure is shown; Figure 15Another flowchart of a method implemented at a second device according to some example embodiments of the present disclosure is shown; Figure 16 A simplified block diagram of a device suitable for implementing example embodiments of the present disclosure is shown; and Figure 17 A block diagram of an example computer-readable medium according to some example embodiments of the present disclosure is shown.

[0014] Throughout the accompanying drawings, the same or similar reference numerals denote the same or similar elements. Detailed Implementation

[0015] The principles of this disclosure will now be described with reference to some exemplary embodiments. It should be understood that these embodiments are described for illustrative purposes only and to help those skilled in the art to understand and implement this disclosure, without implying any limitation on the scope of this disclosure. The embodiments described herein may be implemented in various ways other than those described below.

[0016] In the following description and claims, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0017] References to "an embodiment," "an embodiment," "an example embodiment," etc., in this disclosure indicate that the described embodiments may include specific features, structures, or characteristics, but not every embodiment includes that specific feature, structure, or characteristic. Furthermore, such phrases do not necessarily refer to the same embodiment. In addition, when a specific feature, structure, or characteristic is described in connection with an embodiment, it is to be noted that those skilled in the art will recognize, whether explicitly described or not, that such features, structures, or characteristics apply in conjunction with other embodiments.

[0018] It should be understood that although various elements may be described herein using the preceding terms such as "first," "second," etc., such as noun(s), these elements should not be limited to these terms. These terms are used only to distinguish one element from another, and they do not restrict the order of the noun(s). For example, without departing from the scope of the exemplary embodiments, a first element may refer to a second element, and similarly, a second element may refer to a first element. As used herein, the term "and / or" includes any and all combinations of one or more of the listed terms.

[0019] As used herein, “at least one of the following: ” and “at least one of ” and similar expressions, wherein the list of two or more elements is connected by “and” or “or”, means at least one of these elements, or at least any two or more of these elements, or at least all of these elements.

[0020] As used herein, unless explicitly stated otherwise, the execution step “in response to A” does not indicate that the step is performed immediately after “A” occurs, but may include one or more intermediate steps.

[0021] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments. As used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It will also be understood that the terms “comprising,” “including,” “having,” “possessing,” “containing,” and / or “covering,” as used herein, specify the presence of the stated features, elements, and / or components, but do not exclude the presence or addition of one or more other features, elements, components, and / or combinations thereof.

[0022] As used in this application, the term "circuit" may refer to one or more or all of the following: • (a) Implemented solely with hardware circuitry (e.g., implemented with purely analog and / or digital circuitry), and (b) Combinations of hardware circuitry and software, such as (if applicable): • (i) A combination of (multiple) analog and / or digital hardware circuits and software / firmware, and (ii) Any part of a hardware processor having software (including (multiple) digital signal processors, software, and (multiple) memories, which work together to enable a device (such as a mobile phone or server) to perform various functions), and (c) The operation requires software (e.g., firmware) for the operation of (multiple) hardware circuits and / or (multiple) processors, such as (multiple) microprocessors or parts thereof, but the software may be absent when the operation does not require the software.

[0023] This definition of "circuit" applies to all use of the term in this application (including in any claim). As another example, as used in this application, the term "circuit" also covers only hardware circuitry or processors (or processors), or portions of hardware circuitry or processors and their accompanying software and / or firmware implementations. For example, where applicable to a particular claim element, the term "circuit" also covers baseband integrated circuits or processor integrated circuits for mobile devices or similar integrated circuits in servers, cellular network devices, or other computing or networking devices.

[0024] As used herein, the term "communication network" refers to a network that conforms to any suitable communication standard, such as New Radio (NR), Long Term Evolution (LTE), LTE-A Advanced (LTE-A), Wideband Code Division Multiple Access (WCDMA), High-Speed ​​Packet Access (HSPA), Narrowband Internet of Things (NB-IoT), etc. Furthermore, communication between terminal devices and network devices in a communication network can be performed according to any suitable generation of communication protocol, including but not limited to first-generation (1G), second-generation (2G), 2.5G, 2.75G, third-generation (3G), fourth-generation (4G), 4.5G, fifth-generation (5G), 5.5G, sixth-generation (6G) communication protocols and / or any other currently known or future-developed protocols. Embodiments of this disclosure can be applied to various communication systems. Given the rapid development of communications, there will inevitably be future types of communication technologies and systems that can implement this disclosure. The scope of this disclosure should not be considered limited to the systems described above.

[0025] As used herein, the term "network device" refers to a node in a communications network through which terminal devices access the network and receive services. Depending on the terminology and technology applied, a network device can refer to a base station (BS) or access point (AP), such as a Node B (NodeB or NB), an evolved Node B (eNodeB or eNB), an NR NB (also known as a gNB), a Remote Radio Unit (RRU), a Radio Head (RH), a Remote Radio Head (RRH), a relay, an Integrated Access and Backhaul (IAB) node, a low-power node (such as a femtosecond or picosecond), a non-terrestrial network (NTN) or non-terrestrial network device (such as satellite network devices, low Earth orbit (LEO) satellites, and geostationary Earth orbit (GEO) satellites), an aircraft network device, etc. In some example embodiments, the Radio Access Network (RAN) split architecture includes a centralized unit (CU) and a distributed unit (DU) at the IAB master node. The IAB node includes a mobile terminal (IAB-MT) portion facing the parent node like a UE, while the DU portion of the IAB node faces the next-hop IAB node like a base station.

[0026] The term "terminal device" refers to any terminal device capable of wireless communication. As an example and not a limitation, a terminal device may also be referred to as a communication device, user equipment (UE), subscriber station (SS), portable subscriber station, mobile station (MS), or access terminal (AT). Terminal devices can include, but are not limited to, mobile phones, cellular phones, smartphones, Voice over IP (VoIP) phones, wireless local loop phones, tablets, wearable terminal devices, personal digital assistants (PDAs), portable computers, desktop computers, image acquisition terminal devices (such as digital cameras), gaming terminal devices, music storage and playback devices, in-vehicle wireless terminal devices, wireless endpoints, mobile stations, laptop embedded devices (LEE), laptop mounted devices (LME), USB dongles, smart devices, wireless customer premises equipment (CPE), Internet of Things (IoT) devices, watches or other wearable devices, head-mounted displays (HMDs), vehicles, drones, medical devices and applications (e.g., remote surgery), industrial devices and applications (e.g., robots and / or other wireless devices operating in the context of industrial and / or automated processing chains), consumer electronic devices, devices operating on commercial and / or industrial wireless networks, etc. The terminal device may also correspond to the mobile terminal (MT) portion of an IAB node (e.g., a relay node). In the following description, the terms "terminal device," "communication device," "terminal," "user equipment," and "UE" are used interchangeably.

[0027] As used herein, the terms “resource,” “transmission resource,” “resource block,” “physical resource block” (PRB), “uplink resource,” or “downlink resource” can refer to any resource used to perform communication (e.g., communication between a terminal device and a network device), such as resources in the time domain, resources in the frequency domain, resources in the spatial domain, resources in the code domain, or any other combination of time, frequency, spatial, and / or code domain resources capable of enabling communication. In the following, unless explicitly stated otherwise, resources in both the frequency and time domains will be used as examples of transmission resources used to describe some exemplary embodiments of this disclosure. Note that the exemplary embodiments of this disclosure are equally applicable to other resources in other domains.

[0028] As used in this paper, the term "model" refers to the relationship between inputs and outputs learned from training data, thus enabling the generation of corresponding outputs for a given input after training. Specifically, an AI / ML model can refer to a data-driven algorithm that applies AI / ML techniques to generate a set of outputs based on a set of inputs. Model generation can be based on ML techniques, which can also be called AI techniques. Typically, ML models can be built that receive input information and make predictions based on that information. As used in this paper, "model" is equivalent to an AI / ML model, or a data-driven / data processing algorithm / process.

[0029] As used herein, the term "AI / ML model delivery" is a generic term that refers to the delivery of an AI / ML model from one entity to another in any manner. It should be noted that an entity can refer to a network node / function (e.g., gNB, Location Management Function (LMF), etc.), UE, proprietary server, etc. The term "AI / ML model inference" refers to the process of using a trained AI / ML model to produce an output set based on an input set.

[0030] The term "AI / ML model testing" refers to the training sub-process used to evaluate the performance of the final AI / ML model using a different dataset than that used for model training and validation. Unlike AI / ML model validation, testing does not assume subsequent tuning of the model.

[0031] The term "AI / ML model training" refers to the process of training an AI / ML model, such as learning input / output relationships in a data-driven manner and obtaining a trained AI / ML model for inference.

[0032] The term "AI / ML model delivery" refers to the delivery of an AI / ML model over the air interface in a manner opaque to 3GPP signaling. This delivery may consist of either parameters of a model structure known at the receiving end or a new model with parameters. The delivery may contain a complete model or a partial model.

[0033] The term "AI / ML model validation" refers to the training sub-process of evaluating the quality of an AI / ML model using a different dataset than the one used for model training. This helps in selecting model parameters that generalize to datasets other than those used for model training.

[0034] The term "data collection" refers to the process by which network nodes, management entities, or UEs collect data for the purposes of AI / ML model training, data analysis, and inference.

[0035] The term "functional identification" refers to the process / method for identifying AI / ML functions used for mutual understanding between the NW and UE. It should be noted that information regarding AI / ML functions can be shared during functional identification. These AI / ML functions reside depending on specific use cases and sub-use cases.

[0036] The term "model activation" refers to enabling an AI / ML model for a specific function. The term "model deactivation" refers to disabling an AI / ML model for a specific function. The term "model download" refers to the transfer of a model from the network to the UE.

[0037] The term "model identification" refers to the process or method of identifying an AI / ML model for mutual understanding between the NW and the UE. It should be noted that the process or method of model identification may or may not be applicable. It should also be noted that information about the AI / ML model may be shared during model identification.

[0038] The term "model monitoring" refers to the process of monitoring the inference performance of an AI / ML model. The term "model parameter update" refers to the process of updating the model's parameters.

[0039] The term "model selection" refers to the process of choosing an AI / ML model among multiple models to activate the same AI / ML-enabled features. It should be noted that model selection may or may not be performed concurrently with model activation.

[0040] The term "model switching" refers to activating the currently active AI / ML model and activating a different AI / ML model for a specific function. The term "model update" refers to the process of updating the model parameters and / or model structure. The term "model upload" refers to the transfer of the model from the UE to the network.

[0041] The term "network-side (AI / ML) model" refers to an AI / ML model whose inference is performed entirely at the network. The term "UE-side (AI / ML) model" refers to an AI / ML model whose inference is performed entirely at the UE. The term "dual-side (AI / ML) model" refers to (multiple) paired AI / ML models for which joint inference is performed, where joint inference includes AI / ML inference jointly performed across the UE and network; that is, the first part of the inference is first performed by the UE, and then the remaining part is performed by the gNB, or vice versa.

[0042] The term "offline field data" refers to data collected from a field and used for offline training of AI / ML models. The term "online field data" refers to data collected from a field and used for online training of AI / ML models.

[0043] The term "offline training" refers to the AI / ML training process in which a model is trained based on a collected dataset, and where the trained model is later used or delivered for inference. It should be noted that this definition is for guidance only. There may be cases that do not perfectly fit this definition but can still be classified as offline training by generally accepted conventions.

[0044] The term "online training" refers to the AI / ML training process where the model used for inference is trained (typically continuously) as new training samples arrive (nearly real-time). Note: The concepts of (near) real-time and non-real-time are context-dependent and relative to the inference timescale. It should be noted that this definition is for guidance only. There may be cases that do not perfectly fit this definition but can still be classified as online training by generally accepted conventions. It should be noted that fine-tuning / retraining can be done via online or offline training. This note can be removed for fine-tuning.

[0045] The term "reinforcement learning (RL)" refers to the process of training an AI / ML model from the model's outputs (also called actions) and feedback signals (also called rewards) generated from the model's interactions with its environment. The term "semi-supervised learning" refers to the process of training a model using a mixture of labeled and unlabeled data. The term "supervised learning" refers to the process of training a model from inputs and their corresponding labels. The term "unsupervised learning" refers to the process of training a model without labeled data.

[0046] Figure 1 An example communication environment 100 in which exemplary embodiments of the present disclosure may be implemented is shown. In the communication environment 100, a plurality of devices, including a first device 110 and a second device 120, communicate with each other.

[0047] In some example embodiments, if the first device 110 is a terminal device and the second device 120 is a network device serving the terminal device, the transmission direction from the second device 120 to the first device 110 is referred to as the downlink (DL), and the transmission direction from the first device 110 to the second device 120 is referred to as the uplink (UL). In the DL, the second device 120 is a transmitting (TX) device (or transmitter), and the first device 110 is a receiving (RX) device (or receiver). In the UL, the first device 110 is a TX device (or transmitter), and the second device 120 is an RX device (or receiver).

[0048] exist Figure 1 In the example, the second device 120 has a specific coverage area, which may be referred to as a service area or cell (not shown). The first device 110 is located in the cell covered by the second device 120. In the communication environment 100, the second device 120 can transmit data and control information to the first device 110, and the first device 110 can also transmit data and control information to the second device 120.

[0049] The coverage area or cell can be covered by one or more beams provided by one or more Transmit or Receive Points (TRPs) (e.g., TRP#1, ..., TRP#X). Each beam can carry an identifier that enables the first device 110 to identify the beam and perform measurements (e.g., received power, reference signal received power (RSRP)) and other relevant measurements associated with the specific identifier. Each SSB can be identified based on an identifier carried by a Synchronization Signal Block (SSB). Furthermore, downlink measurement signals for the SSB beams used for beam management can be used for training.

[0050] In some example embodiments, model functionality, such as AI / ML-based features, may be provided to the first device 110 and / or the second device 120. For example, the model functionality may be CSI compression based on AI / ML. For illustrative purposes, some example embodiments will be described below using CSI compression as a model functionality.

[0051] In some example embodiments, one or more models can derive results such as compressed CSI. One or more models can be implemented at a first device 110 and / or a second device 120 (not shown) or both (not shown). As an example, for an AI / ML-based CSI compression function, a first model at the first device 110 can perform compression on the CSI, while a second model at the second device 120 can perform decompression on the compressed CSI. The first and / or second models can perform the corresponding function by running inference or performing training. Such a function can be referred to as a two-sided model-based function or a two-sided model-enabled function.

[0052] It should be understood that Figure 1 The number of devices and their connections shown is for illustrative purposes only and does not imply any limitation. The communication environment 100 may include any suitable number of devices configured to implement the exemplary embodiments of this disclosure.

[0053] In the following description, for illustrative purposes, some exemplary embodiments are described in which the first device 110 operates as a terminal device and the second device 120 operates as a network device. However, in some exemplary embodiments, the operations described in connection with the terminal device can be implemented at the network device or other devices, and the operations described in connection with the network device can be implemented at the terminal device or other devices.

[0054] Communication in communication environment 100 may be implemented according to any suitable communication protocol(s), including but not limited to cellular communication protocols, wireless local area network communication protocols (such as IEEE 802.11, etc.), and / or any other currently known or to be developed in the future. Furthermore, communication may utilize any suitable wireless communication technology, including but not limited to: Code Division Multiple Access (CDMA), Frequency Division Multiple Access (FDMA), Time Division Multiple Access (TDMA), Frequency Division Duplex (FDD), Time Division Duplex (TDD), Multiple Input Multiple Output (MIMO), Orthogonal Frequency Division Multiple Access (OFDM), Discrete Fourier Transform Extended OFDM (DFT-s-OFDM), and / or any other currently known or to be developed in the future.

[0055] In some mechanisms, CSI reporting is a key component of 5G or NR and future generations of radio communication systems. Since the gNB may not be able to directly know the radio conditions the UE is experiencing, optimal transmission of DL signals depends on UE feedback. This allows the gNB to adjust its transmission parameters for better performance.

[0056] In the implicit CSI feedback method, the UE attempts to feed back a representation of the channel state observed by the UE on a certain number of antenna ports, regardless of how the gNB processes the reported CSI when transmitting data to the UE. Similarly, the gNB is unaware of how the UE processes this hypothetical transmission at the receiver. The channel portion of the feedback report may include the channel matrix. Quantization coefficients, TX-side correlation matrix Or its eigenvectors or derived quantities, such as the measured reference signal received power (RSRP).

[0057] Typically, CSI feedback can include the following reporting quantities: Link Indicator (LI), Channel Quality Indicator (CQI), Precoding Matrix Indicator (PMI), or Rank Indicator (RI). PMI can have sub-band frequency granularity and be used for multiple layers.

[0058] The currently specified CSI feedback scheme is based on a model or algorithm. If PMI reports are applied, configure the PMI codebook to be used in the CSI report settings. The following codebook types are defined for NR: Type I, typically in practice, uses only a single-panel codebook to indicate beams from a predefined set.

[0059] The Type II codebook is used to report more detailed channel information feedback as a linear combination of Direct Fourier Transform (DFT) beams. Enhanced Type II (eType II) extends the Type II codebook by supporting up to rank 4 and utilizing frequency domain compression techniques. In 3GPP Release 18, eType II Doppler CSI feedback was introduced as an enhancement to improve the performance of wireless communication systems, particularly in high-mobility scenarios. This feedback mechanism is designed to provide more accurate and timely information about channel conditions, including the Doppler effect.

[0060] CSI report parameters can be encoded in uplink control information (UCI) and mapped to the Physical Uplink Shared Channel (PUSCH) and / or Physical Uplink Control Channel (PUCCH). Compared to PUCCH, PUSCH has a much larger payload capacity and therefore can carry heavier subband CSI reports, or even multiple CSI reports, when using carrier aggregation (CA) operation. The longer PUCCH formats 3 & 4 also have capacity for subband CSI reports, but at a much lower level than PUSCH. The encoding format of the CSI report depends on the channel and frequency granularity of the CSI report. The payload size of the CSI varies with CRI and RI.

[0061] For PUCCH and PUSCH with sub-band frequency granularity, worst-case reporting (i.e., filling to the maximum possible report size) is too costly. Therefore, the CSI content is divided into two parts: CSI Part 1 and CSI Part 2. CSI Part 1 has a fixed payload size (decoded at gNB without prior information). CSI Part 2 has a variable payload size based on information in Part 1.

[0062] If UCI including CSI feedback is transmitted on the PUSCH, the modulation and coding scheme can be flexible and adaptive. The modulation and coding scheme (MCS) for the PUSCH can be dynamically adjusted based on channel conditions and quality of service (QoS) requirements.

[0063] In Random Access Network 4 (RAN4), the Base Station (BS) Technical Specification (TS) 38.104 defines the reliability requirements for CSI reports sent from the UE to the BS in the uplink (UL). For example, it is assumed that the UCI multiplexed on the PUSCH contains only CSI Part 1 and CSI Part 2 information; that is, there are no Hybrid Automatic Repeat Request (HARQ) or Acknowledgment (ACK) messages transmitted in the same UCI. The CSI Part 1 / 2 Block Error Probability (BLER) is defined as the probability or ratio of incorrectly decoding CSI Part 1 / 2 information when it is transmitted. The number of UCI information bits in the payload per time slot is defined for two cases: a first case with 5 bits in CSI Part 1 and 2 bits in CSI Part 2; and a second case with 20 bits in CSI Part 1 and 20 bits in CSI Part 2. In both tests, PUSCH data, CSI Part 1, and CSI Part 2 information are transmitted simultaneously. The minimum requirement is defined as a signal-to-noise ratio (SNR) level where the error probability of the first CSI block must not exceed 0.1% and the error probability of the second CSI block must not exceed 1%.

[0064] Similar requirements are also defined for PUCCH format 2-3 regarding UCI BLER performance. However, the CSI part 2 is not included in the requirements for UCI information transmitted on PUCCH.

[0065] AI and ML in the Air Interface (NR_AIML_Air) research project (SI) have already begun in 3GPP Rel-18. AI-based enhancements to CSI reporting (e.g., reduced overhead, improved accuracy, predictions in the time domain) were included in the initial use case set. The research continues in Rel-19 work item NR_AIML_air with the following objectives.

[0066] One research objective is CSI feedback enhancement. For CSI compression (two-sided models), further research may be expected to improve the trade-off between performance and complexity or overhead. For example, extending spatial or frequency compression to spatial or temporal or frequency compression, cell / site-specific models, CSI compression plus prediction (compared to Rel-18-based non-AI / ML methods), etc., are also considered. Mitigation or resolution of issues related to inter-vendor training collaboration is also discussed. While the necessary canonical impact analysis has been addressed, other aspects requiring further research or conclusions need to be discussed.

[0067] On the RAN4 side, the goal is to define the corresponding requirements for AI / ML-based CSI feedback, including the aforementioned UE CSI feedback requirements. Core requirements for the two use cases mentioned above—AI / ML Lifecycle Management (LCM) procedures and UE characteristics—are specified. For example, the required RAN4 core requirements for the three use cases mentioned above are specified. This includes test setup and methodologies for the beam management use case. Necessary RAN4 core requirements for LCM procedures, including performance monitoring, are specified.

[0068] Complete the testing framework and process for the one-sided model, and collaborate with RAN1 to further analyze various testing options for the two-sided model, including at least: • Relationship with traditional requirements • Consider performance monitoring and LCM aspects in detail of use cases. In terms of generalization, • Static / non-static scenarios / conditions and propagation conditions used for testing (e.g., CDL, field data, etc.). • UE processing capabilities and limitations • Post-deployment validation due to model changes / drift • The RAN5 aspects related to testability and interoperability will be addressed on a request-based basis.

[0069] The main idea behind CSI compression use cases is to use deep neural network AI / ML encoders and decoders, which may be trained together, or at least have some degree of awareness and coordination between them. This approach should leverage AI / ML solutions to replace traditional (e.g., enhanced Type II) CSI feedback, where channel information (e.g., the channel matrix per subband k) can be reconstructed (decoded) at the gNB based on the compressed (encoded) information transmitted from the UE. or channel matrix eigenvalues The advantage is that AI / ML solutions can learn (and encode) channel information more efficiently than traditional algorithmic methods. Several architectures for AI / ML encoders and AI / ML encoder designs have been considered, but transformer-based solutions have proven to be the most efficient to date.

[0070] Example of AI / ML encoder and AI / ML decoder feature vector compression design (200 examples) Figure 2A As shown in the figure, the AI ​​encoder (also known as the compressor) at UE 210 can compress things like the channel matrix. Channel information, in which , Represents the eigenvector. Represents the eigenvalues. It can be selected to correspond to each subband. The vector corresponding to the largest eigenvalue to obtain The compressed channel information can be quantized to obtain bitstream 215. Bitstream 215 is transmitted to gNB 220. Dequantization can be performed on the bitstream. The AI ​​decoder (also called the decompressor) at gNB can decompress the (dequantized) bitstream to obtain, for example... Channel information. It can be based on... and Determine the cosine similarity loss function. As an example, the cosine similarity loss function could be... .

[0071] The reliability aspects of compressed CSI feedback have not yet been discussed, as the type of reporting to be used has not been agreed upon. Only potential canonical impacts are captured in the Rel-18 study. Potential canonical enhancements may be based on aspects including CSI-RS configuration (excluding CSI-RS mode design enhancements), CSI configuration for the network to instruct CSI reporting-related information, such as one or more of the following gNB indications to the UE: information indicating the CSI payload size, information indicating the quantization method / granularity, rank limitations, other payload-related aspects, CSI reporting configuration, information for the UE to determine / report the actual CSI payload size, UE reporting-related information configured by the NW, CSI reporting UCI mapping / priority / omission, and the CSI processing procedure.

[0072] Research on AI / ML-based CSI compression can be based on the traditional CSI feedback signaling framework. Further research into potential specification enhancements could focus on the following aspects: CSI-RS configuration (without discussing enhancements to CSI-RS mode design), CSI report configuration, CSI report UCI mapping / priority / omission, and CSI processing procedures. Other aspects are not excluded.

[0073] In CSI compression using a two-sided model use case, for UE-side monitoring, further investigation is conducted into the impact of potential specifications on the components that trigger and are used to report monitoring metrics, including periodic / semi-persistent and non-periodic reporting, as well as other reports initiated from the UE.

[0074] In CSI compression using two-sided model use cases, the traditional CSI reporting principle with CSI Part 1 and Part 2 can be considered as the starting point for the study of UCI format, where Part 1 has a fixed size for network configuration and Part 2 has a dynamic size determined by the information in Part 1.

[0075] In the original Rel-18 study, AIML-based CSI feedback was split into two separate use cases: CSI compression (primarily discussed above) and CSI prediction. In the latter use case, (multiple) historical channel measurements were used to predict future (i.e., as a specific time range) CSI feedback.

[0076] Additionally, in the research section of the Rel-19 work item, the spatial-frequency-time (SFT) CSI compression technique is considered. In general, when multiple historical or previous channel or feature values ​​or CSI feedback implementations are used as inputs to the AI / ML encoder and AI / ML decoder, both CSI compression in the spatial-frequency domain and prediction in the time domain are combined.

[0077] CSI compression follows the design principles of autoencoders, meaning that the AI / ML encoder and decoder can learn from each other; that is, the output of the AI / ML encoder is used at the input of the AI / ML decoder, and the training of the two models can be done jointly or sequentially. Additionally, complementary outputs from the AI / ML encoder, AI / ML decoder, or additional models can be added to optimize the compression and decompression processes.

[0078] Among some mechanisms, adversarial autoencoders (AAEs) have been proposed, which can transform autoencoders into generative models. The autoencoder is trained using a dual objective—a traditional reconstruction error criterion and an adversarial training criterion that matches the aggregated posterior distribution of the autoencoder's latent representation to an arbitrary prior distribution. The training results in an AI / ML encoder learning to transform the data distribution into a prior distribution, while the AI / ML decoder learns a deep generative model that maps the applied prior distribution to the data distribution.

[0079] Figure 2B An example architecture 230 adversarial to autoencoders is shown. Figure 2B The bottom row shows a trained second network that discriminates between samples generated by the autoencoder's hidden code and those generated by a user-specified sampling distribution. The discriminator model is a neural network that computes the probability that a point x in the data space is a sample from the data distribution it attempts to model (positive samples), rather than a sample from the generative model (negative samples). This method can be used for anomaly detection.

[0080] Another example is the use of an autoencoder and a self-supervised classifier for anomaly detection in strawberries using hyperspectral imaging. In this work, the classifier takes a compressed image (in the latent space) from the autoencoder as input to an additional self-supervised classifier that can detect defects in strawberries. The architecture of the solution is 240. Figure 2CThe diagram illustrates the evaluation of the classifier when detecting quality defects in strawberries, including bruising, fungal infection, and soil contamination. In architecture 240, the self-supervised classification task is designed to distinguish between low-dimensional representations of normal data extracted from the autoencoder and synthetic anomalous data, thereby inducing the autoencoder to learn low-dimensional representations with more discriminative power.

[0081] Figure 2D An example architecture 260 of a classifier autoencoder is shown. Gene expression profiles (GEPs) are encoded or compressed into cell identification codes (CICs). The CICs are connected to two parts: an AI / ML decoder part of the replication expression profile (REP), and a classifier that identifies cell types.

[0082] In some mechanisms, an adaptive CSI feedback scheme for precoding has been proposed, which adjusts the feedback overhead. Specifically, a performance evaluator is developed to predict the reconstruction quality of each image, enabling the scheme to adaptively reduce the CSI feedback overhead of transmitted images with high predicted reconstruction quality in Joint Source Channel Coding (JSCC) systems. The core of this scheme lies in adapting the CSI feedback size to achieve a certain level of reliability for the transmitted data / image (i.e., encoding or compression will be based on this feedback), without considering the recovery or reliability of the CSI feedback itself or its transmission on non-ideal links.

[0083] Current 3GPP operating procedures for CSI compression assume an ideal feedback link. This means that the decompression, reconstruction, or decoding of CSI feedback, or the training of AI / ML encoder or decoder models, does not take into account the potential corruption of UCI elements or codewords received at the gNB or AI / ML decoder. This corruption specifically refers to bit errors, which can occur due to time-varying channel conditions, noise, and interference. Figure 2E The diagram shows a reference scheme for a current autoencoder architecture presenting AIML-enabled CSI feedback, illustrating an example diagram of a simplified architecture 270 for autoencoder-based CSI compression. Figure 2E In this configuration, the AI / ML encoder 272 (such as a compressor on the UE side) compresses the CSI feedback, while the AI / ML decoder 274 at the gNB (such as a decompressor or a decompressor on the BS side) decompresses the compressed CSI feedback. Therefore, considering the ideal UL feedback link between the UE and the BS, the performance evaluation of CSI feedback is performed in 3GPP.

[0084] Currently, in 5G or NR, several mechanisms, such as channel coding, link adaptation (i.e., selecting an appropriate MCS), and retransmission (e.g., HARQ), are used to mitigate potential errors in data or control information transmission. However, the transmission of AI / ML-based compressed control information introduces the following new aspects and challenges that need to be addressed: • The transmission of AI / ML CSI feedback on non-ideal / fading channels is not considered in any particular way (i.e., unlike in the conventional way) at either the transmitter (AI / ML encoder) or the receiver (AI / ML decoder).

[0085] Currently, if the Cyclic Redundancy Check (CRC) of the codeword fails at the receiver, it is assumed that the CSI feedback is completely destroyed—that is, the CSI feedback is completely lost—or it needs to be retransmitted. In both cases, the information reported by the UE will not be received by the BS in a timely manner. However, even if the CRC fails, is it possible to use the CSI feedback decoded at the UE? One of the key performance characteristics of AIML-enabled CSI feedback is the ability to reconstruct or decompress CSI / PMI, meaning it also depends on the quality of the AI / ML decoder implemented on the BS side. The main key performance indicators (KPIs) characterizing the quality of CSI reconstruction or decompression are generally considered to be the generalized squared cosine similarity (SGCS) or the root mean square error (RMSE).

[0086] However, the reconstruction quality of AI / ML-based CSI feedback has not yet been tested, as Type I or Type II CSI feedback is based on closed-form expressions, for example. Therefore, it is sufficient to test that the feedback is correctly decoded by the channel decoder. If conventional CSI feedback contains errors after channel decoding, it may not be used at the gNB and can be retransmitted; that is, such CSI feedback transmission will be considered lost or used for reassembly after retransmission.

[0087] Existing requirements for UCI transmission exist in RAN4 demodulation, but they are not designed for the typical payload size of AI / ML compressed CSI feedback. Based on the RAN1 study and assumptions in TR 38.843, the payload can be in the range of X: <= 80 bits, Y: 100-140 bits, Z: >= 230 bits. Furthermore, the requirements for UCI transmission in RAN4 demodulation do not consider any real-world payload. That is, the bits used in the test are predefined and do not represent any real PMI feedback; i.e., real AI / ML CSI feedback is not considered. Additionally, the reconstruction quality of AI / ML-enabled CSI / PMI feedback is not tested.

[0088] One of the main benefits of AI / ML-enabled algorithms is their ability to generalize to data that is not present in the training dataset. This unknown data covers samples outside the training dataset, including samples that have been modified or corrupted by errors. Introducing some errors into the model's input (e.g., by adding variations / errors) or interfering with the model's inference (e.g., by using dropout layers) is a common practice. This non-ideality improves the robustness of models to unknown data and so-called overfitting, without causing a significant degradation in prediction / inference quality.

[0089] To demonstrate the reconstructing capability of AI / ML-enabled CSI feedback considering channel conditions, in Figures 2F to 2G The simulation results are presented in Figures 280 and 290, which show example correlation and NMSE degradation for a 64-length codeword for a UE with two receive antennas. Specifically, the simulation shows an increase in the number of erroneous bits in the latent space (i.e., in the compressed CSI report) and illustrates the impact on the reconstructed CSI. Experiments were conducted for a UE with two receive antennas and a time frame of 100 time slots. The percentage of erroneous bits was calculated by dividing the total number of erroneous bits by the total number of bits in the CSI feedback transmitted over the 100 time slots. Furthermore, the entire experiment was repeated 100 times to ensure statistical reliability. In a time-division duplex (TDD) configuration, an AI / ML encoder and AI / ML decoder for this experiment were learned on DL channel estimation, but without considering any channel effects on the UL feedback. The correlation and NMSE metric between the decoded CSI and the ideal CSI (i.e., before encoding or compression) were then averaged across receive antenna dimensions, time slots, and independent trials (i.e., over 20,000 samples).

[0090] from Figure 2F and Figure 2G It can be seen that in 3GPP AI / ML-based CSI feedback, the 2-bit quantization configuration suffers most significantly from coarse quantization, even at a bit error rate of 1e-6, and performance degradation begins above 1% of the bit error rate. Furthermore, 4-bit quantized codewords show similar degradation, while 8-bit quantized codewords are even more robust at high bit error rates. Even without considering the channel impact on AI / ML-based compressed CSI during autoencoder training, the model generalizes well and overcomes bit errors, especially when using higher-order quantization in the latent space.

[0091] To address at least some of the aforementioned or other potential problems, several solutions for AI / ML-based CSI compression are proposed. According to an example embodiment of this disclosure, a solution for AI / ML-enabled CSI feedback training, inference, and testing is proposed. In this solution, the non-zero probability of errors due to feedback transmission over real-world (i.e., error-prone, fading, noisy, etc.) channels is considered. In the solution, novel testing procedures can be designed that consider not only conventional BLER for CSI feedback but also actual compression reliability; i.e., different UL CSI or PMI reporting tests and metrics can be designed in RAN4. The objective is to evaluate the difference (e.g., squared generalized cosine similarity (SGCS) or root mean square error (RMSE)) between the CSI feedback available at the BS after propagation over error-prone or fading channels and AI / ML decoding utilizing the original uncompressed CSI feedback.

[0092] In another solution, a first device (such as a terminal device) determines a CSI report based on channel state information and according to a first model at the first device. The CSI report includes a compressed CSI. The first device transmits the CSI report to a second device (such as a network device). The second device 120 correspondingly receives the CSI report. The first device performs an operation if at least one condition is met. For example, the operation could be to retransmit the CSI report to the second device. Another example is that the operation could be to update the channel state information and the CSI report and transmit the updated CSI report. The at least one condition includes a first condition where a message from the second device 120 indicates a failure in channel decoding and / or decompression of the CSI report. If the decoding performance metric for decompressing the CSI report is below a threshold, the second device 120 transmits a first message to the first device 110 indicating a failure in receiving or decoding the CSI report. The at least one condition includes a second condition where an evaluated performance metric for channel decoding and / or decompression of the CSI report at the second device 120 is below a threshold. The evaluated performance metric is evaluated by the first model.

[0093] In this way, the first device can decide whether to retransmit the CSI report to the second device. The first and / or second device can evaluate the successful channel decoding and / or decompression of the CSI report and trigger the retransmission of the CSI report based on the evaluation. Therefore, unnecessary retransmission of CSI reports can be avoided.

[0094] The principles and implementation of this disclosure will be described in detail below. Note that... Figures 3 to 7 The sequence of actions shown is merely an example and not a limitation. Actions can be performed in any suitable manner. Reference Figures 3 to 7The described example embodiments can be implemented individually or in any combination. For example, one or more example embodiments shown in a single figure can be combined with one or more example embodiments shown in one or more other figures.

[0095] Figure 3 An example signaling flow 300 for AI / ML-based CSI compression is shown according to some example embodiments of this disclosure. Signaling flow 300 relates to... Figure 1 The first device 110 and the second device 120 in the middle.

[0096] In operation, the first device 110 determines (310) a CSI report based on channel state information and according to a first model at the first device 110. The first model may refer to a compressor or AI / ML encoder, such as a CSI encoder. The CSI report (also known as CSI feedback or CSI feedback information) may include a precoding matrix indicator (PMI) report. The PMI report may include PMIs with subband frequency granularity and for multiple layers. It should be understood that the CSI report may also include other reporting quantities, such as LI, CQI, RI, etc. For the purposes of discussion, some example embodiments will be described with respect to the PMI report.

[0097] The first device 110 transmits (320) a report to the second device 120. For example, the first device 110 may perform channel coding on a compressed CSI to obtain a CSI report and transmit the CSI report to the second device 120. The second device 120 correspondingly receives (325) the CSI report. The second device 120 performs (330) channel decoding and decompression on the CSI report. For example, channel decoding may be performed on the CSI report, followed by decompression using a second model at the second device 120. The second model may refer to a decompressor or AI / ML decoder, such as a CSI decoder. If the performance metric for channel decoding and / or decompression of the CSI report is below a threshold, the second device 120 may transmit (350) a first message to the first device 110. For example, the second device 120 may determine the performance metric for channel decoding of the CSI, or the performance metric for decompressing the CSI (also referred to as the performance metric for reconstructing or reconstructing the CSI), or determine the performance metric for joint channel decoding and decompression of the CSI. The first message indicates a failure to decode and / or decompress the channel reported by CSI. The first threshold may be predefined or configured. The first device 110 may receive this first message.

[0098] The first device 110 determines (350) whether at least one condition is met. The at least one condition includes a first condition where a message from the second device 120 (e.g., a first message) indicates that a performance metric for channel decoding and / or decompression of the CSI report is less than a threshold. For example, the first message may indicate low demodulation quality or low decompression or reconstruction quality. In an example, the first message may indicate failure of CSI report reception or channel decoding and / or decompression. As used herein, channel decoding and decompression of the CSI report may be collectively referred to as decoding the CSI report. The at least one condition includes a second condition where an evaluated performance metric for channel decoding and / or decompression of the CSI report at the second device 120 is less than a threshold. As used herein, a performance metric for channel decoding and / or decompression of the CSI report may be referred to as a “performance metric for decoding the CSI report” or a “performance metric for decoding the CSI report” or a “performance metric for the CSI report”. The performance metric may be a performance metric for channel decoding, a performance metric for decompression, or a performance metric for both channel decoding and decompression. The performance of channel decoding and decompression may be based on a performance metric for channel decoding. The evaluated performance metrics are assessed by a first model and / or a separate evaluator (or classifier) ​​at the first device 110. Thresholds may be predefined or configured. If a second condition is met, the first device 110 may proactively retransmit the CSI report. It should be understood that these conditions are for illustrative purposes only and do not imply any limitations. These conditions and any other suitable conditions may be applied individually or in any combination. The scope of the invention is not limited thereto.

[0099] If at least one condition is met, the first device 110 performs operation (360). For example, this operation could be to retransmit (370) the CSI report to the second device 120. The second device 120 could receive (375) the retransmitted CSI report. For another example, the operation could be to update the channel state information and the CSI report, and (re)transmit (370) the updated CSI report. The second device 120 could receive (375) the (re)transmitted updated CSI report. Potentially, taking into account retransmissions and / or previous transmissions or (multiple) retransmission failures, a first model (such as an AI / ML encoder or a CSI compressor (or a separate model)) could provide the updated CSI.

[0100] In some example embodiments, the first device 110 may use a first model to compress the channel state information to obtain compressed channel state information. The first device 110 may perform channel coding and modulation on the compressed channel state information to obtain a CSI report. That is, compression and channel coding and modulation may be performed separately.

[0101] Alternatively, the first device 110 can perform joint compression and channel coding on the channel state information using the first model to obtain a CSI report. Joint compression and channel coding can also be referred to as integrated source channel coding. For example, the most efficient channel coding for CSI feedback can be learned along with compression. The compression ratio and payload size can be adapted to UL channel conditions. For example, information or messages can be compressed optimally while maintaining the same payload size and compression ratio. Because UL channel conditions are favorable and allow for carrying more information in the information or message, the payload can be increased and the compression ratio can be reduced. Information or messages can be compressed flexibly without sacrificing payload, for example, by maintaining a sufficiently low compression ratio. In the case of joint compression and channel coding, the first device 110 can evaluate the performance of the evaluated channel decoding and decompression.

[0102] In some example embodiments, the channel state information includes at least one feature vector or channel matrix of the channel. The first device 110 may compress the channel state information using a first model based on reference information. The reference information may include, but is not limited to, one or more of the following: at least one feature value of the channel, compressed channel state information feedback, signal-to-interference-plus-noise ratio (SINR), MCS, or the power of the transmitter at the first device 110.

[0103] In some example embodiments, the first device 110 may determine an evaluated decoding performance metric based on a configuration of reference information, compressed channel state information, and at least one previous CSI and / or CSI report. The first device 110 may transmit the evaluated decoding performance metric to the second device 120. As an example, the decoding performance metric may be decoding probability, decoding accuracy, etc.

[0104] In some example embodiments, the second device 120 can perform channel decoding on the CSI report to obtain the decoded codewords of the CSI report. The second device 120 can decompress the decoded codewords using a second model to obtain the decompressed CSI. That is, channel decoding and decompression can be performed separately. In some cases, CSI feedback information (e.g., UCI) is transmitted over a wireless channel after channel coding and modulation (i.e., using a specific MCS). Similarly, at the receiver (i.e., the second device 120), CSI feedback decompression / reconstruction at the AI / ML decoder can be completed after channel decoding. If channel decoding is deemed unsuccessful (e.g., after CRC check), HARQ feedback (non-ACK, NACK) can be used to indicate the need for retransmission.

[0105] In some example embodiments, the second device 120 may determine the number of bit errors in the decoded codeword using CRC. If the number of bit errors exceeds a threshold, the second device 120 may transmit a first message to the first device 110 to indicate low-quality decoding of the CSI report. The threshold may be predefined or configured.

[0106] If the number of bit errors is below a threshold, the second device 120 can transmit a second message to the first device 110. The second message, such as an ACK message, indicates successful channel decoding of the CSI report. The second device 120 can decompress the decoded codeword using a second model to obtain decompressed channel state information. In this way, the CSI feedback at the receiver (i.e., the second device 120) can be used or recovered, even if it contains bit errors due to transmissions on a non-ideal UL channel. The CSI feedback can be used even if there is a CRC checksum error after channel decoding.

[0107] Alternatively, in some example embodiments, the second device 120 may utilize the second model to perform joint channel decoding and CSI decompression on the CSI report to obtain decompressed channel state information. The CSI feedback at the output of the AI / ML decoder already includes (e.g., AI / ML-based) channel coding; that is, the output of the AI / ML decoder is (multiple) codewords that are directly mapped to constellation points (e.g., a Quadrature Amplitude Modulation (QAM) constellation). Similarly, at the receiver, the CSI feedback can be directly reconstructed after demodulation (without explicit channel decoding). Information regarding the success of CSI reconstruction or decompression can be indicated to the transmitter, for example, for retransmission. In the case of joint channel decoding and decompression, the second device 120 may determine a performance metric for joint channel decoding and decompression of the CSI report.

[0108] In some example embodiments, the second device 120 may obtain channel state information using a second model based on the CSI report and reference information. The reference information may include at least one characteristic of the channel, compressed channel state information feedback, SINR, MCS, transmitter power at the first device 110, etc. The second device 120 may determine a performance metric for the CSI report based on the CSI report and reference information, according to the second model or a separate evaluator (or classifier).

[0109] In some example embodiments, the second device 120 may receive an evaluated decoding performance metric of the CSI report from the first device 110. The second device 120 may also determine a decoding performance metric of the CSI report based on the evaluated decoding performance metric.

[0110] In some example embodiments, the second device 120 may receive one or more retransmissions of CSI reports from the first device 110. The second device 120 may decompress multiple consecutive retransmissions of CSI reports to obtain decompressed CSIs.

[0111] Using these embodiments, AI / ML encoders and / or AI / ML decoders can evaluate the success of reconstructing or decompressing feedback using selected or current transmission parameters and / or under expected channel conditions. The following will discuss... Figures 4 to 6 Other embodiments of AI / ML-based CSI compression are described.

[0112] In some example embodiments, trained UL channel awareness or robust CSI feedback can be used in the field of the device, i.e., during model inference. The second device 120 can configure the AI / ML model configuration for the first device 110. For example, this configuration may include thresholds for evaluating decoding performance. Figure 4 Example signaling flow 400 for configuring and / or selecting options for an AI / ML model according to some example embodiments of this disclosure is illustrated. Signaling flow 400 involves... Figure 1 The first device 110 and the second device 120 in the middle.

[0113] In some example embodiments, the first device 110 may transmit capability information of the first device 110 to the second device 120. For example, the capability information may include, but is not limited to, one or more of the following: joint compression and channel coding and modulation for CSI reports, support for retransmission of CSI reports in the absence of a trigger, or support for evaluation of decoding performance metrics of CSI reports at the first device 110. The second device 120 may receive the capability information of the first device 110.

[0114] In some example embodiments, in step 4, the second device 120 may transmit a UE capability information request to the first device 110. In response to receiving this request, in step 5, the first device 110 may transmit a UE capability information response to the second device 120. The configuration of the new capability may be on RRC (i.e., when the configuration is relatively static), or in a more dynamic manner, i.e., together with CSI feedback configuration.

[0115] Support for new UE capabilities and / or NW features may be required to provide the new functionalities described below. For example, coordination on whether to support UL channel-aware models may be necessary. Another capability requiring support for CSI compression is joint channel coding and compression. Another capability is support for proactive retransmission based on successful decoding assessment on the UE side.

[0116] In some example embodiments, the second device 120 may transmit to the first device 110 a configuration including at least one of the following: at least one threshold for a decoding performance metric of the CSI report, a period for retransmitting the CSI report, or an indication to perform an operation based on at least one condition being met. Accordingly, the first device 110 may receive the configuration. As an example, the configuration may be included in a Radio Resource Control (RRC) message, or a CSI report configuration, or any other suitable message or signaling.

[0117] Based on the capabilities or features reported or supported by the first device 110 and the second device 120, at step 6, the second device 120 may configure the first device 110 to use those, for example, via RRC(re-) configuration.

[0118] In one example, when the UE AI / ML encoder can assess the probability of CSI reconstruction or decompression, the network can establish a threshold for acceptable decoded CSI accuracy or probability. If the assessed value is below the threshold, the first device 110 allows automatic retransmission. Multiple thresholds can be configured, for example, in a mapping manner (e.g., according to MCS and / or SINR or SNR), and exchanged between the second device 120 and the first device 110.

[0119] In some example embodiments, the first device 110 may select a target threshold from at least one threshold. If the evaluated decoding performance metric is below the target threshold, the first device 110 may retransmit (270) the CSI report to the second device 120.

[0120] In some example embodiments, in step 7, interoperability between the first device 110 and the second device 120 can be ensured. This step may include updating or obtaining (multiple) matching AI / ML models on the UE or NW side or model or function selection, based on applicable / associated IDs, etc., and depending on whether a UL channel-aware model is required.

[0121] At step 9, the second device 120 may transmit CSI feedback or report configuration to the first device 110. This configuration may include known parameters such as scheduling authority, UCI size, payload, modulation and coding, as well as some new parameters, such as active retransmission. Periodic (less dynamic) configuration may also be performed in step 6. At step 10, the second device 120 may transmit a reference signal or symbol, such as CSI-RS, to the first device 110. At step 11, the first device 110 may use the reference signal for channel estimation at its location.

[0122] Using these embodiments, the generalization capability of AI / ML-enabled CSI compression can be used to overcome bit errors in the feedback to avoid unnecessary retransmissions.

[0123] Figure 5 Example signaling flows for channel-aware CSI feedback are shown according to some example embodiments of this disclosure. Signaling flow 500 relates to... Figure 1 The first device 110 and the second device 120 are described in the signaling stream 500. It is assumed that in the signaling stream 500, the CSI encoder or CSI decoder and the channel encoder or channel decoder are implemented separately from each other. That is, the signaling stream 500 illustrates channel-aware CSI feedback with separate channel coding.

[0124] At step 1, CSI feedback, such as eigenvectors or channel matrices, is transmitted to the CSI encoder at the first device 110. At step 2, if the CSI coding is contributed using an assessment of reconstruction or decompression probability or quality, and if it is determined that the reconstruction or decompression accuracy will be lower due to errors in the UL channel feedback, then an active retransmission of the CSI feedback can be initiated by the first device 110. At step 3, the channel encoder at the first device 110 performs channel coding. At step 4, the first device 110 transmits a CSI report to the second device 120.

[0125] At step 5, the second device 120 can decide regarding ACK / NACK. For example, the NACK condition can be relaxed because decoding can still be accurate even if bit errors exist in the compressed CSI (i.e., after transmission). Even if any error is detected by CRC, it does not mean the received CSI is useless. A new standard for CSINACK can be defined based on the generalization capability of the AI / ML decoder, for example, if the number of bit errors identified by CRC is less than... If the bit is set, an ACK is sent; this can be predefined or configured. .

[0126] At step 6, for option 1, the second device 120 transmits HARQ feedback for the compressed CSI to the first device 110, wherein ACK / NACK is determined after channel decoding.

[0127] At step 8, a classifier or evaluator can be used, which provides information on the accuracy, probability, chance, or (pass / fail) of reconstructing or decompressing the AI / ML CSI feedback.

[0128] At step 9, the alternative to step 6 is option 2, where HARQ feedback is sent based on an assessment of the decoding quality of the received CSI feedback. At step 10C, following the result of step 9, it is decided whether the received CSI feedback can be used. In one alternative, some information from the new CSI feedback can also be used and fused with the new but inaccurate feedback, for example, when the channel does not change very dynamically.

[0129] At step 11, depending on the availability of information regarding the reconstruction quality of the compressed CSI on the NW side, the first device 110 may proactively retransmit the CSI to the NW or wait for HARQ feedback.

[0130] If the reception of the CSI feedback (with many bit errors) or decoding (assuming the decoded feedback is inaccurate) is unsuccessful, steps 14-17 repeat the corresponding steps 5-10. Using these embodiments, the AI / ML encoder and / or AI / ML decoder can also evaluate the success of decoding the feedback using selected / current transmission parameters and / or under expected channel conditions.

[0131] Figure 6 Another example signaling flow for channel-aware CSI feedback is shown according to some example embodiments of this disclosure. Signaling flow 600 relates to Figure 1 The first device 110 and the second device 120 are described in the signaling stream 600. It is assumed that CSI compression and channel coding at the first device 110 and decompression and channel decoding at the second device 120 are jointly performed in the signaling stream 600.

[0132] At step 1, CSI feedback, such as feature vectors or channel matrices, is transmitted to the CSI encoder at the first device 110. At step 2, joint CSI encoding is performed on the CSI feedback to obtain a CSI report. In some embodiments, the AI / ML encoder (or an additional classifier or evaluator model) can assess the need for retransmission. For example, the encoder can output the assessed decoding probabilities. At step 3, the first device 110 transmits the CSI report to the second device 120. If the AI / ML encoder (or the additional classifier / evaluator model) is able to assess the need for retransmission, then, in an option, at step 7, an active retransmission can occur before HARQ feedback is received from the second device 120.

[0133] At step 4, the joint AI / ML CSI decoder at the second device 120 can decode or decompress the CSI report. At step 5, the second device 120 can determine whether the CSI decoding was successful based on the decoding probability. If decoding is successful, the decoded or decompressed CSI can be used at step 6. If decoding is unsuccessful, in an option, at step 8, the second device 120 can transmit an ACK or NACK to the first device 120 based on the decoding probability. The ACK / NACK can be determined based on the additional AI / ML decoder output or according to the classifier / evaluator model. In an alternative, the CSI feedback can still contain some known bits that are known to the AI / ML decoder (also used during training), which help determine the accuracy / decoding probability of the CSI feedback.

[0134] At step 9, in response to receiving a NACK, the first device 110 may retransmit the CSI report. At step 10, if the previous CSI decoding failed, the first device 110 may decode or decompress the combined CSI report. Steps 11-12 can be similar to steps 5-6, and will not be repeated here.

[0135] Using signaling streams 500 and 600, CSI feedback at the receiver can be used or recovered even if errors or CRC checksum errors are present. Retransmission of CSI feedback can be controlled.

[0136] It should be understood that although several example embodiments of this disclosure have been described with respect to CSI feedback or CSI compression use cases, CSI feedback or CSI compression is for illustrative purposes only and does not imply any limitation. Similar methods can be used for any other two-sided model that assumes reporting from a first model (e.g., on the UE side) to a second model (e.g., on the gNB side) via an error-prone radio channel.

[0137] In the solution, a first device (such as a terminal device) determines compressed feedback information based on a first model at the first device and feedback information associated with a second device (such as a network device). For example, the feedback information may be CSI feedback information. The first device determines an evaluated performance metric for channel decoding and / or decompression of the compressed feedback information at the second device, based on the first model and the feedback information and additional input information associated with the feedback information. The first device 110 transmits at least one of the following to the second device: the compressed feedback information or the evaluated performance metric. The second device performs channel decoding and decompression of the feedback information. For example, decompressing the compressed feedback information is based on a second model at the second device. The second device determines a performance metric for channel decoding and / or decompression of the compressed feedback information based on the second model and additional input information associated with the feedback information.

[0138] In this way, the compression and / or decompression of the models at the first and second devices can be evaluated. For example, the success of decoding the feedback using selected / current transmission parameters and / or under expected channel conditions can be evaluated.

[0139] Figure 7 An example signaling flow 700 for AI / ML-based feedback information compression evaluation is illustrated according to some example embodiments of this disclosure. Signaling flow 700 relates to... Figure 1 The first device 110 and the second device 120 in the middle.

[0140] In operation, the first device 110 determines (710) compressed feedback information based on a first model at the first device 110 and based on feedback information associated with the second device 120. For example, the feedback information may be CSI feedback information. The first device 110 determines (720) an evaluated performance metric for channel decoding and / or decompression of the compressed feedback information at the second device 120 based on the first model and based on the feedback information and additional input information associated with the feedback information. The first device 110 transmits (730) the compressed feedback information and / or the evaluated performance metric to the second device 120. The first device 110 may perform channel coding on the compressed feedback information and transmit channel-coded compressed feedback information to the second device 120. Accordingly, the second device 120 receives (735) the (channel-coded) compressed feedback information and / or the evaluated performance metric. As used herein, the performance metrics for channel decoding and / or decompression of compressed feedback information may be referred to as "performance metrics for decoding compressed feedback information", "performance metrics for decoding compressed feedback information", or "performance metrics for compressed feedback information".

[0141] In some example embodiments, the feedback information may include at least one feature vector or channel matrix of the channel. Additional input information may include, but is not limited to, one or more of the following: at least one feature value of the channel, compressed feedback information, SINR, MCS, or the power of the transmitter at the first device 110.

[0142] The first model may include a compressor (also known as an AI / ML encoder) and a classifier. The compressed feedback information can be determined by the compressor based on the feedback information and the additional input information. The evaluated performance metric can be determined by the classifier based on the additional input information and the compressed feedback information.

[0143] In some example embodiments, the first device 110 may also determine the evaluated performance metric based on at least one of the following: at least one previous feedback information, or the configuration of the feedback information.

[0144] In some example embodiments, the first device 110 may perform channel coding and modulation on the compressed feedback information. The first device 110 may then transmit (730) the compressed feedback information to the second device 120 after channel coding and modulation.

[0145] In some example embodiments, if the evaluated performance metric is below a threshold, the first device 110 may proactively retransmit the compressed feedback information to the second device 120. The second device 120 may receive the retransmitted compressed feedback information. Alternatively or additionally, in some example embodiments, if the evaluated performance metric is below a threshold, the first device 110 may update the feedback information and the compressed feedback information. The first device 110 may (re)transmit the updated compressed feedback information to the second device 120. The second device 120 may receive the updated compressed feedback information.

[0146] It should be understood that the thresholds used herein, and any other thresholds used herein, may be predefined, configured by the second device 120, or adaptively determined by the first device 110 and / or the second device 120. The scope of the invention is not limited thereto. For example, the second device 120 may transmit a configuration of multiple candidate thresholds to the first device 110. The first device 110 may select a threshold from these multiple candidate thresholds based on SINR, SNR, and / or MCS, etc.

[0147] In some example embodiments, the second device 120 may transmit to the first device 110 a configuration including at least one threshold for an evaluated performance metric, and / or an instruction to retransmit compressed feedback information based on the evaluated performance metric. In this way, the first device 110 may proactively retransmit compressed feedback information based on the configuration.

[0148] In response to receiving (735) compressed feedback information, the second device 120 performs (740) channel decoding and decompression on the compressed feedback information. For example, decompression is performed using a second model at the second device 120. Similar to uncompressed feedback information, decompressed feedback information may include at least one eigenvector or channel matrix of the channel.

[0149] The second device 120 determines (750) a performance metric for channel decoding and / or decompression of the compressed feedback information based on the second model and additional input information associated with the feedback information. The performance metric may be a performance metric for channel decoding of the compressed feedback information, a performance metric for decompressing the compressed feedback information, or a performance metric for both channel decoding and decompression of the compressed feedback information. The additional input information may include, but is not limited to, one or more of the following: at least one characteristic value of the channel, the compressed feedback information, SINR, MCS, or the power of the transmitter at the first device 110.

[0150] In some example embodiments, the second model may include a decompressor and a classifier. The decompressed feedback information may be determined by the decompressor based on the compressed feedback information and additional input information. The performance metric may be determined by the classifier based on the additional input information and the decompressed feedback information. In some example embodiments, the second device 120 may further determine (750) the performance metric based on at least one previously received feedback information and / or the configuration of the feedback information.

[0151] In some example embodiments, if the performance metric is below a threshold, the second device 120 may transmit a first message to the first device 110 indicating a failure to receive or decode the feedback information. The first message may be a HARQ message such as a NACK message. In response to receiving the first message, the first device 110 may retransmit the compressed feedback information. In this way, the retransmission of the compressed feedback information may be triggered by the second device 120. Otherwise, if the performance metric exceeds the threshold, the second device 120 may transmit a second message to the first device 110. The second message indicates successful channel decoding or decompression of the CSI report. For example, the second message may be an ACK message.

[0152] In some example embodiments, the second device 120 can perform channel decoding on the compressed feedback information to obtain the decoded codewords of the compressed feedback information. The second device 120 can decompress the decoded codewords using a second model to obtain decompressed feedback information. The second device 120 can determine the number of bit errors in the decoded codewords using CRC. If the number of bit errors exceeds a threshold number, the second device 120 can transmit a first message (such as a NACK message) to the first device 110. If the number of bit errors is less than the threshold number, the second device 120 can decompress the decoded codewords using the second model to obtain decompressed feedback information. The second device 120 can transmit a third message to the first device 110 indicating successful channel decoding of the compressed feedback information. In this way, even if the CRC check identifies errors in the decompressed feedback information, the decompressed feedback information can still be used or recovered.

[0153] In some example embodiments, the second device 120 may receive multiple consecutive transmissions or retransmissions of compressed feedback information from the first device 110. The second device 120 may decompress the multiple consecutive retransmissions of compressed feedback information to obtain decompressed feedback information.

[0154] In some example embodiments, the first device 110 may transmit capability information of the first device 110 to the second device 120. The capability information may include at least one of the following: joint compression and channel coding and modulation of feedback information, support for retransmission of feedback information in the absence of triggering, or support for evaluation of performance metrics of compressed feedback information at the first device 110.

[0155] about Figure 7 Several example embodiments are described regarding the performance evaluation of decoding compressed feedback information at the first device 110 and / or the second device 120. These embodiments can be related to... Figures 3 to 6 Any of the described embodiments can be applied in combination. For example, the evaluation of performance metrics for CSI feedback can be performed in a similar manner.

[0156] As mentioned, in some example embodiments, the first device 110 may use separate models or modules for CSI compression and channel coding and modulation. Figure 8An example diagram 800 illustrates AI / ML CSI feedback with conventional channel coding and modulation according to some example embodiments of this disclosure. As shown, an AI / ML encoder 810 and a channel encoder and modulation module 820 are located at a first device 110 (i.e., the encoder side or the UE side). On the encoder side, additional inputs (such as eigenvalues ​​(if the eigenvectors rather than the entire channel matrix are compressed), SINR, MCS, TX power, etc.) can be input to the AI / ML encoder 810. The additional inputs can be used to evaluate the success of CSI reconstruction or decompression (additional output) on the gNB side, or to increase the robustness of the CSI feedback to the current channel conditions.

[0157] The AI / ML decoder 840 and the demodulation and channel decoder 830 are located at the second device 120 (i.e., the decoder side or gNB side). The AI / ML decoder side can use additional inputs (such as eigenvalues ​​(if the eigenvectors rather than the entire channel matrix are compressed), SINR, MCS, TX power, etc.) to evaluate the success of CSI reconstruction or decompression quality or probability (additional output) on the gNB side, or to increase the robustness of the CSI feedback to current channel conditions, or to improve the decoding efficiency of codewords. The additional output can evaluate the reconstruction or decompression quality or probability, or can be used to distinguish between CSI feedback or codewords with or without errors after decompression or reconstruction. It should be noted that the additional inputs or outputs can be processed in the same joint model performing the encoding / decoding of the CSI feedback or by separate models.

[0158] Figure 9 Example Figure 900 illustrates AI / ML CSI feedback with integrated channel coding according to some example embodiments of this disclosure. Figure 9 In the example, the AI / ML encoder 910 of the first device 110 includes a compressed CSI and a channel coding module. The compressed and channel-coded CSI feedback can be input to the modulation module 920 to obtain a CSI report to be transmitted to the demodulation module of the second device 120. The demodulated CSI report can be decoded by the AI / ML decoder 940, which acts as an integrated CSI decompressor and channel decoder.

[0159] Similar to Figure 8 ,exist Figure 9In the AI / ML encoder 910, additional inputs (such as eigenvalues ​​(if the eigenvectors, rather than the entire channel matrix, are compressed), SINR, MCS, TX power, etc.) can be fed into the AI / ML decoder 940. These additional inputs can be used to evaluate the success of CSI decompression or reconstruction (additional output) on the gNB side, or to increase the robustness of CSI feedback to current channel conditions. On the AI / ML decoder side, additional inputs (such as eigenvalues ​​(if the eigenvectors, rather than the entire channel matrix, are compressed), SINR, MCS, TX power, etc.) can be used to evaluate the success of CSI decoding (additional output) on the gNB side, or to increase the robustness of CSI feedback to current channel conditions, or to improve the decoding efficiency of codewords. The additional outputs can evaluate the reconstruction or decompression quality or probability, or can be used to distinguish between CSI feedback or codewords with and without errors after decompression or reconstruction.

[0160] In some example embodiments, AI / ML-based decoders may be able to assess the probability of correct reconstruction or decompression of compressed CSI feedback in the current channel conditions and / or CSI feedback configuration (or any other relevant metric, such as SGCS, MSE, presence or number of errors, etc.). Figure 10 Example Figure 1000 illustrates an AI / ML decoder design with evaluation of decoding probabilities using compressed CSI feedback, according to some example embodiments of this disclosure. As shown, an AI / ML decoder (such as a CSI decompressor) 1010 and a classifier 1020 (also referred to as an evaluator) are located at a first device 110. The classifier can evaluate performance metrics, such as decoding or decompression accuracy or probability (referred to as "prob."), ​​based on compressed CSI and optionally previous CSI, CSI configuration, etc. The received compressed CSI is after channel decoding. It should be understood that in Figure 10 In this model, a channel decoder may exist that performs channel decoding on the received compressed CSI. In the case of separate channel coding, the CSI feedback prior to channel decoding can also be used as additional input to the AI / ML decoder and / or AI / ML classifier. A similar operation can be applied for joint channel coding (decoding) and CSI (decompression). Although the AI / ML decoder 1010 and classifier 1020 are shown as separate modules, it should be understood that the classifier 1020 and AI / ML decoder 1010 may also be implemented in a single model (i.e., with additional inputs and outputs).

[0161] Similarly, AI / ML-based encoders can be able to evaluate the decoding probability (i.e., decompression probability or any other relevant metric) of compressed CSI feedback in the current channel conditions and / or CSI feedback configuration. Figure 11Example Figure 1100 illustrates an AI / ML encoder design with an evaluation of the decoding probability of compressed CSI feedback according to some example embodiments of the present disclosure. As shown, the AI / ML encoder (such as a CSI compressor 1110) and classifier 1120 (also referred to as an evaluator) are located at a second device 120.

[0162] Classifier 1120 can evaluate the quality or accuracy of decompressing or reconstructing CSI at the decoder or receiver, or the presence or number of errors. The classifier can be implemented as a standalone model or as part of an AI / ML decoder that takes additional inputs and provides additional outputs. The classifier can evaluate performance metrics, such as decompression accuracy or probability, based on the encoded CSI, channel information, Tx parameters, etc. Channel information can include the entire channel matrix or a matrix (e.g., per subband) and CSI TX parameters (e.g., MCS, scheduled resources, etc.). Optionally, additional inputs (shown for classifier 1120) or their processed derivatives can also be used to provide UL channel awareness for CSI encoding. For channel decoding of the received CSI, a channel decoding probability can also be determined. The channel decoding probability can include accuracy based on channel-related SINR, etc.

[0163] By using, for example Figures 8 to 11 The AI / ML-based encoder and decoder shown can be used to evaluate the decompression probability (or any other relevant metric) of compressed CSI feedback in the current channel conditions and / or CSI feedback configuration. These models, AI / ML encoders, and / or AI / ML decoders can be applied to… Figures 3 to 7 These are some of the example embodiments.

[0164] These embodiments enhance the architecture of the AI / ML CSI decoder on the NW side. Specifically, CSI feedback (additionally, also based on UL channel condition preprocessing) before or after channel decoding can be used as additional input to the AI / ML decoder or decompressor. Furthermore, the additional output of the AI / ML decoder (which can be implemented using one or more models) provides a confidence metric for the CSI feedback or an assessment of the decompression or reconstruction probability. Such a new output can be used to indicate the retransmission of the CSI feedback. The output of the AI / ML decoder can correspond to quality or performance metrics for both channel decoding and AI / ML compression.

[0165] Furthermore, the architecture of the AI / ML CSI encoder on the UE side can be enhanced. In the example, the additional output of the AI / ML encoder (which can be implemented using a single or multiple models) provides an assessment of the probability of decompression or reconstruction of the CSI feedback on the NW side. If the probability of accurate decoding of the CSI feedback is low (e.g., the expected SGCS is the circuit breaker threshold), proactive retransmission of the CSI feedback can be performed before receiving HARQ feedback. In another example, compressed or channel-coded CSI feedback affected by the real UL channel can be considered as additional input to the AI / ML encoder (e.g., recursively). The output of the AI / ML encoder can correspond to a quality or performance metric for both channel decoding and AI / ML compression.

[0166] In some mechanisms, tests are performed on the frequency reference channel configuration, as shown in Table 1 below. It should be understood that these parameters or values ​​in Table 1 are for illustrative purposes only and do not imply any limitations.

[0167] Table 1

[0168] However, the current CSI feedback testing process only focuses on the demodulation performance of UCI information carrying CSI feedback with some payload (i.e., channel decoding performance metrics, such as the ratio of incorrectly decoded CSI Part 1 / 2 reports / blocks to the total number of transmitted CSI Part 1 / 2 reports).

[0169] The CSI feedback use case is used as an illustrative example. However, a similar approach can be used for other two-sided models that assume reporting from the first model (e.g., on the UE side) to the second model (e.g., on the gNB side) via an error-prone radio channel.

[0170] In some example implementations, the new aspects of AI / ML-based CSI feedback described above can be considered. The following enhancements (new components) can be made in the testing: • The CSI payload is adjusted (or increased) during testing to match typical AI / ML-based CSI reporting payloads in the typical 60-bit to 200-bit or higher range.

[0171] • The payload transmitted in UCI is not just random or fixed meaningless bits, but real / realistic AIML-based compressed CSI feedback that can be decoded on the BS side.

[0172] √ Realistic / replicable compressed CSI feedback can be generated by an AI / ML encoder implemented in a test apparatus (TE). For example, if agreed upon in 3GPP, the AI / ML encoder can follow a reference encoder design. The AI / ML encoder can be provided by an NW vendor. The AI / ML encoder can be provided by a TE vendor.

[0173] √ Realistic, compressed CSI feedback can be obtained from pre-generated or available datasets. Datasets can be standardized within 3GPP or generated by the NW or TE vendor before testing begins.

[0174] • Introduce new test metrics (e.g., in addition to the existing error probabilities of the UCI).

[0175] √ The test metric defines the extent to which CSI feedback is received and decoded (i.e., decompressed or reconstructed) by the AI / ML decoder. SGCS or RMSE or a similar metric can be calculated between the known transmitted CSI feedback (before compression) and the decoded CSI feedback.

[0176] The above tests validated the decoding capabilities of the AI / ML decoder (i.e., on the BS side). However, if the UE AI / ML encoder also supports some new capabilities / outputs, such as decoding probabilities (e.g., for proactive retransmission), this capability can also be tested at RAN4. These retransmissions are traditionally triggered by channel decoding errors, rather than by processing stages after the channel decoder.

[0177] In this case, the testing method could be as follows: the confidence metric exported from the AI / ML encoder ( The accuracy of CSI is compared with that derived from the output of the (test) AI / ML decoder.

[0178] In some example embodiments, the proposed test procedure can be applied to... Figures 3 to 11 The AI / ML encoders (such as CSI compressors) and / or AI / ML decoders (such as CSI decompressors) used in the described embodiments can also be applied to any suitable model, such as a two-sided model for information compression or any other suitable use case. Channel decoding performance metrics and decompression performance metrics can be used individually or in combination in some example embodiments. The embodiments disclosed herein are not limited.

[0179] According to embodiments of this disclosure, the generalization capability of AI / ML-enabled CSI compression can be used to overcome channel decoding and / or decompression errors and avoid unnecessary retransmissions. The AI / ML decoder can tolerate bit errors if they exist in the received CSI feedback. The CSI feedback can occur after channel decoding or demodulation. Furthermore, this functionality will be tested in RAN4 or RAN5 after the proposed test procedure or any other suitable test procedure.

[0180] It should be understood that the above provides some example specifications, signaling flows, and embodiments, and the detailed description may vary. It should be understood that these signaling flows 300, 400, 500, 600, and / or 700 can be used individually or in any suitable combination. Some example embodiments, operations, or features described with respect to one of these signaling flows 300, 400, 500, 600, and / or 700 can be applied to another signaling flow within these signaling flows. A portion of one of these signaling flows 300, 400, 500, 600, and / or 700 can be applied in combination with a portion of another signaling flow. It should be understood that these signaling flows 300, 400, 500, 600, and / or 700 can involve any other suitable operations or signaling not shown. Utilizing these signaling flows and similar signaling flows, AI / ML-based CSI compression can be enhanced.

[0181] Figure 12 A flowchart of an example method 1200 implemented at a first device according to some example embodiments of the present disclosure is shown. For the purposes of discussion, [the following will be discussed]. Figure 1 Method 1200 is described by the angle of the first device 110 in the middle.

[0182] At box 1210, the first device 110 determines a CSI report based on the CSI and according to a first model at the first device, the CSI report including compressed CSI.

[0183] At frame 1220, the first device 110 transmits a CSI report to the second device.

[0184] At block 1230, based on at least one condition being met, the first device 110 performs at least one of the following: retransmits the CSI report to the second device, or updates the CSI and the CSI report and transmits the updated CSI report.

[0185] The at least one condition includes at least one of the following: a first condition, wherein a message from the second device indicates that the performance metric for channel decoding and / or decompression of the CSI report is less than a threshold, or a second condition, wherein an evaluated performance metric for channel decoding and / or decompression of the CSI report at the second device is less than a threshold, the evaluated performance metric being evaluated by at least one of the following: in a first model or evaluator at the first device.

[0186] In some example embodiments, method 1200 further includes: compressing channel state information using a first model to obtain compressed channel state information; and performing channel coding and modulation on the compressed channel state information to obtain a CSI report.

[0187] In some example embodiments, method 1200 further includes performing joint compression and channel coding on the channel state information using a first model to obtain a CSI report.

[0188] In some example embodiments, method 1200 further includes: compressing channel state information using a first model based on reference information, the reference information including at least one of the following: at least one characteristic value of the channel, compressed channel state information feedback, SINR, MCS, or the power of the transmitter at the first device.

[0189] In some example embodiments, method 1200 further includes determining an evaluated performance metric based on at least one of the following: reference information, compressed channel state information, at least one previous CSI, or the configuration of a CSI report.

[0190] In some example embodiments, method 1200 further includes transmitting an evaluated performance metric to a second device.

[0191] In some example embodiments, method 1200 further includes transmitting capability information of the first device to the second device, the capability information including at least one of the following: joint compression and channel coding and modulation for CSI reports, support for retransmission of CSI reports in the absence of triggering, or support for evaluation of performance metrics at the first device.

[0192] In some example embodiments, method 1200 further includes receiving from the second means a configuration including at least one of the following: at least one threshold for a performance metric of channel decoding or decompression of CSI reports, a period for retransmitting CSI reports, or an instruction to perform an operation based on at least one condition being met.

[0193] In some example embodiments, method 1200 further includes: selecting a target threshold from at least one threshold; and retransmitting the CSI report to a second device based on the evaluated performance metric being lower than the target threshold.

[0194] In some example embodiments, the configuration is included in at least one of the following: configuration of radio resource control messages or CSI reports.

[0195] In some example implementations, the CSI report includes a precoded matrix indicator (PMI) report.

[0196] Figure 13 A flowchart of an example method 1300 implemented at a second device according to some example embodiments of the present disclosure is shown. For the purposes of discussion, [the following will be discussed]. Figure 1 Method 1300 is described by the angle of the second device 120 in the middle.

[0197] In block 1310, the second device 120 receives a channel state information (CSI) report from the first device, the CSI report including compressed channel state information.

[0198] In block 1320, the second device 120 performs channel decoding and decompression on the CSI report, at least based on the second model at the second device.

[0199] At block 1330, based on a performance metric for channel decoding or decompression of a CSI report being below a threshold, the second device 120 transmits a first message to the first device, the first message indicating that the performance metric for channel decoding and / or decompression of a CSI report is below a threshold.

[0200] In some example embodiments, method 1300 further includes receiving a retransmission of the CSI report from the first device.

[0201] In some example embodiments, method 1300 further includes: decompressing multiple consecutive retransmissions of the CSI report to obtain a decompressed CSI.

[0202] In some example embodiments, method 1300 further includes: performing channel decoding on the CSI report to obtain the channel decoded codeword of the CSI report; and decompressing the channel decoded codeword using a second model to obtain the decompressed CSI.

[0203] In some example embodiments, method 1300 further includes: determining the number of bit errors in the channel decoded codeword by means of cyclic redundancy check (CRC); and transmitting a first message to the first device based on the determination that the number of bit errors exceeds a threshold.

[0204] In some example embodiments, method 1300 further includes: transmitting a second message to a first device based on determining that the number of bit errors is below a threshold, the second message indicating successful channel decoding of the CSI report; and decompressing the channel decoding codeword using a second model to obtain decompressed channel state information.

[0205] In some example embodiments, method 1300 further includes: performing joint channel decoding and CSI decompression on the CSI report using a second model to obtain decompressed channel state information; and determining a performance metric for joint decompression and channel decoding.

[0206] In some example embodiments, method 1300 further includes: obtaining channel state information based on CSI reports and reference information using a second model, the reference information including at least one of the following: at least one characteristic value of the channel, compressed channel state information feedback, SINR, MCS, or the power of the transmitter at the first device.

[0207] In some example embodiments, method 1300 further includes: determining a performance metric based on the CSI report and reference information according to the second model.

[0208] In some example embodiments, method 1300 further includes: receiving from the first device an evaluated performance metric of channel decoding and / or decompression of the CSI report; and further determining a performance metric based on the evaluated performance metric.

[0209] In some example embodiments, method 1300 further includes receiving capability information of the first device from the first device, the capability information including at least one of the following: joint compression and channel coding and modulation for CSI reports, support for retransmission of CSI reports in the absence of triggering, or evaluation of performance metrics for channel decoding or decompression of CSI reports at the first device.

[0210] In some example embodiments, method 1300 further includes transmitting to the first device a configuration including at least one of the following: at least one threshold for a performance metric of channel decoding and / or decompression of CSI reports, a period for retransmitting CSI reports, or an instruction to perform an operation based on at least one condition being met.

[0211] Figure 14 A flowchart of an example method 1400 implemented at a first device according to some example embodiments of the present disclosure is shown. For the purposes of discussion, [the following will be discussed]. Figure 1 Method 1400 is described by the angle of the first device 110 in the middle.

[0212] At frame 1410, the first device 110 determines compressed feedback information based on the first model at the first device and on feedback information associated with the second device.

[0213] At block 1420, the first device 110 determines, based on a first model or evaluator and on feedback information and additional input information associated with the feedback information, an evaluated performance metric for channel decoding and / or decompression of the compressed feedback information at the second device.

[0214] At frame 1430, the first device 110 transmits at least one of the following to the second device: compressed feedback information or evaluated performance metrics.

[0215] In some example embodiments, method 1400 further includes receiving from the second device a configuration including at least one of the following: at least one threshold for an evaluated performance metric, or an indication to retransmit compressed feedback information based on the evaluated performance metric.

[0216] In some example embodiments, the feedback information includes at least one feature vector or channel matrix of the channel, and the additional input information includes at least one of the following: at least one feature value of the channel, compressed feedback information, SINR, MCS, or the power of the transmitter at the first device.

[0217] In some example embodiments, the first model includes a compressor and a classifier, wherein compressed feedback information is determined by the compressor based on the feedback information and additional input information, and the evaluated performance metric is determined by the classifier based on the additional input information and compressed feedback information.

[0218] In some example embodiments, method 1400 further includes: determining the evaluated performance metric based on at least one of the following: at least one previous feedback message or a configuration of the feedback message.

[0219] In some example embodiments, method 1400 further includes: performing channel coding and modulation on the compressed feedback information; and transmitting the compressed feedback information to a second device after channel coding and modulation.

[0220] In some example embodiments, method 1400 further includes: retransmitting compressed feedback information to a second device based on an evaluated performance metric being below a threshold.

[0221] In some example embodiments, method 1400 further includes: updating the feedback information and the compressed feedback information based on the evaluated performance metric being below a threshold; and transmitting the updated compressed feedback information to a second device.

[0222] In some example embodiments, the threshold is predefined or configured by a second device.

[0223] In some example embodiments, method 1400 further includes: receiving a configuration of multiple candidate thresholds from a second device; and selecting a threshold from the multiple candidate thresholds based on at least one of the following: signal-to-interference plus noise ratio (SINR), signal-to-noise ratio (SNR), or modulation and channel modulation and coding scheme (MCS).

[0224] In some example embodiments, method 1400 further includes transmitting capability information of the first device to the second device, the capability information including at least one of the following: joint compression and channel coding and modulation of feedback information, support for retransmission of feedback information in the absence of triggering, or support for evaluation of performance metrics of compressed feedback information at the first device.

[0225] In some example implementations, the compressed feedback information includes CSI reports.

[0226] In some example implementations, the CSI report includes the PMI report.

[0227] Figure 15 A flowchart of an example method 1500 implemented at a second device according to some example embodiments of the present disclosure is shown. For the purposes of discussion, [the following will be discussed]. Figure 1 Method 1500 is described by the angle of the second device 120 in the middle.

[0228] In frame 1510, the second device 120 receives compressed feedback information from the first device.

[0229] In block 1520, the second device 120 performs channel decoding and decompression on the compressed feedback information, at least based on the second model at the second device.

[0230] At block 1530, the second device 120 determines a performance metric for channel decoding and / or decompression of the compressed feedback information based on the second model or evaluator and additional input information associated with the feedback information.

[0231] In some example embodiments, the second device 120 may transmit to the first device a configuration including at least one of the following: at least one threshold for an evaluated performance metric, or an instruction to retransmit compressed feedback information based on the evaluated performance metric.

[0232] In some example embodiments, the decompressed feedback information includes at least one feature vector or channel matrix of the channel, and wherein the additional input information includes at least one of the following: at least one feature value of the channel, the compressed feedback information, SINR, MCS, or the power of the transmitter at the first device.

[0233] In some example embodiments, the second model includes a decompressor and a classifier, wherein the decompressed feedback information is determined by the decompressor based on the compressed feedback information and additional input information, and the performance metric is determined by the classifier based on the additional input information and the decompressed feedback information.

[0234] In some example embodiments, method 1500 further includes: determining a performance metric based on at least one of the following: at least one previously received feedback message, or a configuration of the feedback message.

[0235] In some example embodiments, method 1500 further includes: transmitting a first message to a first device based on a performance metric being below a threshold, the first message indicating a failure of channel decoding and / or decompression of feedback information.

[0236] In some example embodiments, method 1500 further includes: transmitting a second message to a first device based on a performance metric exceeding a threshold, the second message indicating successful channel decoding and / or decompression of the CSI report.

[0237] In some example embodiments, method 1500 further includes: performing channel decoding on the compressed feedback information to obtain channel decoded codewords of the compressed feedback information; and decompressing the channel decoded codewords using a second model to obtain decompressed feedback information.

[0238] In some example embodiments, method 1500 further includes: determining the number of bit errors in the channel decoded codeword by means of cyclic redundancy check (CRC); and transmitting a first message to the first device based on the determination that the number of bit errors exceeds a threshold.

[0239] In some example embodiments, method 1500 further includes: decompressing the channel decoded codeword using a second model to obtain decompressed feedback information based on determining that the number of bit errors is below a threshold; and transmitting a third message to the first device, the third message indicating successful channel decoding or decompression of the compressed feedback information.

[0240] In some example embodiments, method 1500 further includes: receiving a plurality of successive retransmissions of compressed feedback information from a first device; and performing channel decoding and decompression on the plurality of successive retransmissions of the compressed feedback information to obtain decompressed feedback information.

[0241] In some example embodiments, method 1500 further includes receiving capability information of the first device from the first device, the capability information including at least one of the following: joint compression and channel coding and modulation of feedback information, support for retransmission of feedback information in the absence of triggering, or support for evaluation of performance metrics of compressed feedback information at the first device.

[0242] In some example embodiments, a first device capable of performing any method of method 1200 (e.g., Figure 1 The first device 110 may include components for performing the corresponding operation of method 1200. The device may be implemented in any suitable form. For example, the device may be implemented in a circuit or software module. The first device may be implemented as or included in... Figure 1 In the first device 110.

[0243] In some example embodiments, the first device includes: components for determining a CSI report based on CSI and according to a first model at the first device, the CSI report including compressed CSI; components for transmitting the CSI report to a second device; and components for performing at least one of the following based on at least one condition being met: retransmitting the CSI report to the second device; or updating the CSI and the CSI report and transmitting the updated CSI report, wherein at least one condition includes at least one of the following: a first condition, the first condition being a message from the second device indicating that the performance metric for channel decoding and / or decompression of the CSI report is less than a threshold, or a second condition, the second condition being that the evaluated performance metric for channel decoding and / or decompression of the CSI report at the second device is less than a threshold, the evaluated performance metric being evaluated by at least one of the following: the first model at the first device or an evaluator.

[0244] In some example embodiments, a second means capable of performing any of the methods in method 1300 (e.g., Figure 1 The second device 120 may include components for performing the corresponding operations of method 1300. This device may be implemented in any suitable form. For example, it may be implemented in a circuit or software module. The second device may be implemented as or included in... Figure 1 The second device 120 in the middle.

[0245] In some example embodiments, the second device includes: components for receiving a channel state information (CSI) report from the first device, the CSI report including compressed channel state information; components for performing channel decoding and decompression on the CSI report, at least the decompression being based on a second model at the second device; and components for transmitting a first message to the first device based on a performance metric for channel decoding or decompression of the CSI report being less than a threshold, the first message indicating that the performance metric for channel decoding and / or decompression of the CSI report is less than the threshold.

[0246] In some example embodiments, a first device capable of performing any method of method 1400 (e.g., Figure 1The first device 110 may include components for performing the corresponding operation of method 1400. The device may be implemented in any suitable form. For example, the device may be implemented in a circuit or software module. The first device may be implemented as or included in... Figure 1 In the first device 110.

[0247] In some example embodiments, the first device includes: components for determining compressed feedback information based on a first model at the first device and based on feedback information associated with a second device; components for determining an evaluated performance metric for channel decoding and / or decompression of the compressed feedback information at the second device based on the first model or an evaluator and based on the feedback information and additional input information associated with the feedback information; and components for transmitting at least one of the following to the second device: the compressed feedback information or the evaluated performance metric.

[0248] In some example embodiments, a second means capable of performing any of the methods in method 1500 (e.g., Figure 1 The second device 120 may include components for performing the corresponding operation of method 1500. This device may be implemented in any suitable form. For example, it may be implemented in a circuit or software module. The second device may be implemented as or included in... Figure 1 The second device 120 in the middle.

[0249] In some example embodiments, the second device includes: components for receiving compressed feedback information from the first device; components for performing channel decoding and decompression on the compressed feedback information, wherein at least the decompression is based on a second model at the second device; and components for determining a performance metric for channel decoding and / or decompression of the compressed feedback information based on the second model or an evaluator and on additional input information associated with the feedback information.

[0250] Figure 12 A flowchart of an example method 1200 implemented at a first device according to some example embodiments of the present disclosure is shown. For the purposes of discussion, [the following will be discussed]. Figure 1 Method 1200 is described by the angle of the first device 110 in the middle.

[0251] At box 1210, the first device 110 determines a CSI report based on the CSI and according to a first model at the first device, the CSI report including compressed CSI.

[0252] At frame 1220, the first device 110 transmits a CSI report to the second device.

[0253] At block 1230, based on at least one condition being met, the first device 110 performs at least one of the following: retransmits the CSI report to the second device, or updates the CSI and the CSI report and transmits the updated CSI report.

[0254] The at least one condition includes at least one of the following: a first condition, wherein a message from the second device indicates that the performance metric for channel decoding and / or decompression of the CSI report is less than a threshold, or a second condition, wherein an evaluated performance metric for channel decoding and / or decompression of the CSI report at the second device is less than a threshold, the evaluated performance metric being evaluated by at least one of the following: in a first model or evaluator at the first device.

[0255] In some example embodiments, method 1200 further includes: compressing channel state information using a first model to obtain compressed channel state information; and performing channel coding and modulation on the compressed channel state information to obtain a CSI report.

[0256] In some example embodiments, method 1200 further includes performing joint compression and channel coding on the channel state information using a first model to obtain a CSI report.

[0257] In some example embodiments, method 1200 further includes: compressing channel state information using a first model based on reference information, the reference information including at least one of the following: at least one characteristic value of the channel, compressed channel state information feedback, SINR, MCS, or the power of the transmitter at the first device.

[0258] In some example embodiments, method 1200 further includes determining an evaluated performance metric based on at least one of the following: reference information, compressed channel state information, at least one previous CSI, or the configuration of a CSI report.

[0259] In some example embodiments, method 1200 further includes transmitting an evaluated performance metric to a second device.

[0260] In some example embodiments, method 1200 further includes transmitting capability information of the first device to the second device, the capability information including at least one of the following: joint compression and channel coding and modulation for CSI reports, support for retransmission of CSI reports in the absence of triggering, or support for evaluation of performance metrics at the first device.

[0261] In some example embodiments, method 1200 further includes receiving from the second means a configuration including at least one of the following: at least one threshold for a performance metric of channel decoding or decompression of CSI reports, a period for retransmitting CSI reports, or an instruction to perform an operation based on at least one condition being met.

[0262] In some example embodiments, method 1200 further includes: selecting a target threshold from at least one threshold; and retransmitting the CSI report to a second device based on the evaluated performance metric being lower than the target threshold.

[0263] In some example embodiments, the configuration is included in at least one of the following: configuration of radio resource control messages or CSI reports.

[0264] In some example implementations, the CSI report includes a precoded matrix indicator (PMI) report.

[0265] Figure 13 A flowchart of an example method 1300 implemented at a second device according to some example embodiments of the present disclosure is shown. For the purposes of discussion, [the following will be discussed]. Figure 1 Method 1300 is described by the angle of the second device 120 in the middle.

[0266] In block 1310, the second device 120 receives a channel state information (CSI) report from the first device, the CSI report including compressed channel state information.

[0267] In block 1320, the second device 120 performs channel decoding and decompression on the CSI report, at least based on the second model at the second device.

[0268] At block 1330, based on a performance metric for channel decoding or decompression of the CSI report being below a threshold, the second device 120 transmits a first message to the first device, the first message indicating a failure of channel decoding and / or decompression of the CSI report.

[0269] In some example embodiments, method 1300 further includes receiving a retransmission of the CSI report from the first device.

[0270] In some example embodiments, method 1300 further includes: decompressing multiple consecutive retransmissions of the CSI report to obtain a decompressed CSI.

[0271] In some example embodiments, method 1300 further includes: performing channel decoding on the CSI report to obtain the channel decoded codeword of the CSI report; and decompressing the channel decoded codeword using a second model to obtain the decompressed CSI.

[0272] In some example embodiments, method 1300 further includes: determining the number of bit errors in the channel decoded codeword by means of cyclic redundancy check (CRC); and transmitting a first message to the first device based on the determination that the number of bit errors exceeds a threshold.

[0273] In some example embodiments, method 1300 further includes: transmitting a second message to a first device based on determining that the number of bit errors is below a threshold, the second message indicating successful channel decoding of the CSI report; and decompressing the channel decoding codeword using a second model to obtain decompressed channel state information.

[0274] In some example embodiments, method 1300 further includes: performing joint channel decoding and CSI decompression on the CSI report using a second model to obtain decompressed channel state information; and determining a performance metric for joint decompression and channel decoding.

[0275] In some example embodiments, method 1300 further includes: obtaining channel state information based on CSI reports and reference information using a second model, the reference information including at least one of the following: at least one characteristic value of the channel, compressed channel state information feedback, SINR, MCS, or the power of the transmitter at the first device.

[0276] In some example embodiments, method 1300 further includes: determining a performance metric based on the CSI report and reference information according to the second model.

[0277] In some example embodiments, method 1300 further includes: receiving from the first device an evaluated performance metric of channel decoding and / or decompression of the CSI report; and further determining a performance metric based on the evaluated performance metric.

[0278] In some example embodiments, method 1300 further includes receiving capability information of the first device from the first device, the capability information including at least one of the following: joint compression and channel coding and modulation for CSI reports, support for retransmission of CSI reports in the absence of triggering, or evaluation of performance metrics for channel decoding or decompression of CSI reports at the first device.

[0279] In some example embodiments, method 1300 further includes transmitting to the first device a configuration including at least one of the following: at least one threshold for a performance metric of channel decoding and / or decompression of CSI reports, a period for retransmitting CSI reports, or an instruction to perform an operation based on at least one condition being met.

[0280] Figure 14A flowchart of an example method 1400 implemented at a first device according to some example embodiments of the present disclosure is shown. For the purposes of discussion, [the following will be discussed]. Figure 1 Method 1400 is described by the angle of the first device 110 in the middle.

[0281] At frame 1410, the first device 110 determines compressed feedback information based on the first model at the first device and on feedback information associated with the second device.

[0282] At block 1420, the first device 110 determines, based on the first model and on feedback information and additional input information associated with the feedback information, an evaluated performance metric for channel decoding and / or decompression of the compressed feedback information at the second device.

[0283] At frame 1430, the first device 110 transmits at least one of the following to the second device: compressed feedback information or evaluated performance metrics.

[0284] In some example embodiments, method 1400 further includes receiving from the second device a configuration including at least one of the following: at least one threshold for an evaluated performance metric, or an indication to retransmit compressed feedback information based on the evaluated performance metric.

[0285] In some example embodiments, the feedback information includes at least one feature vector or channel matrix of the channel, and the additional input information includes at least one of the following: at least one feature value of the channel, compressed feedback information, SINR, MCS, or the power of the transmitter at the first device.

[0286] In some example embodiments, the first model includes a compressor and a classifier, wherein compressed feedback information is determined by the compressor based on the feedback information and additional input information, and the evaluated performance metric is determined by the classifier based on the additional input information and compressed feedback information.

[0287] In some example embodiments, method 1400 further includes: determining the evaluated performance metric based on at least one of the following: at least one previous feedback message or a configuration of the feedback message.

[0288] In some example embodiments, method 1400 further includes: performing channel coding and modulation on the compressed feedback information; and transmitting the compressed feedback information to a second device after channel coding and modulation.

[0289] In some example embodiments, method 1400 further includes: retransmitting compressed feedback information to a second device based on an evaluated performance metric being below a threshold.

[0290] In some example embodiments, method 1400 further includes: updating the feedback information and the compressed feedback information based on the evaluated performance metric being below a threshold; and transmitting the updated compressed feedback information to a second device.

[0291] In some example embodiments, the threshold is predefined or configured by a second device.

[0292] In some example embodiments, method 1400 further includes: receiving a configuration of multiple candidate thresholds from a second device; and selecting a threshold from the multiple candidate thresholds based on at least one of the following: signal-to-interference plus noise ratio (SINR), signal-to-noise ratio (SNR), or modulation and channel modulation and coding scheme (MCS).

[0293] In some example embodiments, method 1400 further includes transmitting capability information of the first device to the second device, the capability information including at least one of the following: joint compression and channel coding and modulation of feedback information, support for retransmission of feedback information in the absence of triggering, or support for evaluation of performance metrics of compressed feedback information at the first device.

[0294] In some example implementations, the compressed feedback information includes CSI reports.

[0295] In some example implementations, the CSI report includes the PMI report.

[0296] Figure 15 A flowchart of an example method 1500 implemented at a second device according to some example embodiments of the present disclosure is shown. For the purposes of discussion, [the following will be discussed]. Figure 1 Method 1500 is described by the angle of the second device 120 in the middle.

[0297] In frame 1510, the second device 120 receives compressed feedback information from the first device.

[0298] In block 1520, the second device 120 performs channel decoding and decompression on the compressed feedback information, at least based on the second model at the second device.

[0299] At block 1530, the second device 120 determines a performance metric for channel decoding and / or decompression of the compressed feedback information based on the second model and additional input information associated with the feedback information.

[0300] In some example embodiments, the second device is also caused to transmit to the first device a configuration including at least one of the following: at least one threshold for an evaluated performance metric, or an instruction to retransmit compressed feedback information based on the evaluated performance metric.

[0301] In some example embodiments, the decompressed feedback information includes at least one feature vector or channel matrix of the channel, and wherein the additional input information includes at least one of the following: at least one feature value of the channel, the compressed feedback information, SINR, MCS, or the power of the transmitter at the first device.

[0302] In some example embodiments, the second model includes a decompressor and a classifier, wherein the decompressed feedback information is determined by the decompressor based on the compressed feedback information and additional input information, and the performance metric is determined by the classifier based on the additional input information and the decompressed feedback information.

[0303] In some example embodiments, method 1500 further includes: determining a performance metric based on at least one of the following: at least one previously received feedback message, or a configuration of the feedback message.

[0304] In some example embodiments, method 1500 further includes: transmitting a first message to a first device based on a performance metric being below a threshold, the first message indicating a failure of channel decoding and / or decompression of feedback information.

[0305] In some example embodiments, method 1500 further includes: transmitting a second message to a first device based on a performance metric exceeding a threshold, the second message indicating successful channel decoding and / or decompression of the CSI report.

[0306] In some example embodiments, method 1500 further includes: performing channel decoding on the compressed feedback information to obtain channel decoded codewords of the compressed feedback information; and decompressing the channel decoded codewords using a second model to obtain decompressed feedback information.

[0307] In some example embodiments, method 1500 further includes: determining the number of bit errors in the channel decoded codeword by means of cyclic redundancy check (CRC); and transmitting a first message to the first device based on the determination that the number of bit errors exceeds a threshold.

[0308] In some example embodiments, method 1500 further includes: decompressing the channel decoded codeword using a second model to obtain decompressed feedback information based on determining that the number of bit errors is below a threshold; and transmitting a third message to the first device, the third message indicating successful channel decoding or decompression of the compressed feedback information.

[0309] In some example embodiments, method 1500 further includes: receiving a plurality of successive retransmissions of compressed feedback information from a first device; and performing channel decoding and decompression on the plurality of successive retransmissions of the compressed feedback information to obtain decompressed feedback information.

[0310] In some example embodiments, method 1500 further includes receiving capability information of the first device from the first device, the capability information including at least one of the following: joint compression and channel coding and modulation of feedback information, support for retransmission of feedback information in the absence of triggering, or support for evaluation of performance metrics of compressed feedback information at the first device.

[0311] In some example embodiments, a first device capable of performing any method of method 1200 (e.g., Figure 1 The first device 110 may include components for performing the corresponding operation of method 1200. The device may be implemented in any suitable form. For example, the device may be implemented in a circuit or software module. The first device may be implemented as or included in... Figure 1 In the first device 110.

[0312] In some example embodiments, the first device includes: components for determining a CSI report based on CSI and according to a first model at the first device, the CSI report including compressed CSI; components for transmitting the CSI report to a second device; and components for performing at least one of the following based on at least one condition being met: retransmitting the CSI report to the second device; or updating the CSI and the CSI report and transmitting the updated CSI report, wherein at least one condition includes at least one of the following: a first condition, the first condition being a message from the second device indicating a failure of channel decoding and / or decompression of the CSI report, or a second condition, the second condition being that an evaluated performance metric for channel decoding and / or decompression of the CSI report at the second device is below a threshold, the evaluated performance metric being evaluated by the first model or evaluator at the first device.

[0313] In some example embodiments, a second means capable of performing any of the methods in method 1300 (e.g., Figure 1 The second device 120 may include components for performing the corresponding operations of method 1300. This device may be implemented in any suitable form. For example, it may be implemented in a circuit or software module. The second device may be implemented as or included in... Figure 1 The second device 120 in the middle.

[0314] In some example embodiments, the second device includes: components for receiving a channel state information (CSI) report from the first device, the CSI report including compressed channel state information; components for performing channel decoding and decompression on the CSI report, at least the decompression being based on a second model at the second device; and components for transmitting a first message to the first device based on a performance metric for channel decoding or decompression of the CSI report being below a threshold, the first message indicating a failure of channel decoding and / or decompression of the CSI report.

[0315] In some example embodiments, a first device capable of performing any method of method 1400 (e.g., Figure 1 The first device 110 may include components for performing the corresponding operation of method 1400. The device may be implemented in any suitable form. For example, the device may be implemented in a circuit or software module. The first device may be implemented as or included in... Figure 1 In the first device 110.

[0316] In some example embodiments, the first device includes: components for determining compressed feedback information based on a first model at the first device and based on feedback information associated with a second device; components for determining an evaluated performance metric for channel decoding and / or decompression of the compressed feedback information at the second device based on the first model and based on the feedback information and additional input information associated with the feedback information; and components for transmitting at least one of the following to the second device: the compressed feedback information or the evaluated performance metric.

[0317] In some example embodiments, a second means capable of performing any of the methods in method 1500 (e.g., Figure 1 The second device 120 may include components for performing the corresponding operation of method 1500. This device may be implemented in any suitable form. For example, it may be implemented in a circuit or software module. The second device may be implemented as or included in... Figure 1 The second device 120 in the middle.

[0318] In some example embodiments, the second device includes: components for receiving compressed feedback information from the first device; components for performing channel decoding and decompression on the compressed feedback information, wherein at least the decompression is based on a second model at the second device; and components for determining a performance metric for channel decoding and / or decompression of the compressed feedback information based on the second model and on additional input information associated with the feedback information.

[0319] Figure 16 This is a simplified block diagram of a device 1600 suitable for implementing an example embodiment of the present disclosure. The device 1600 can be provided to implement a communication device, such as... Figure 1 The first device 110 or the second device 120 shown. As shown, the device 1600 includes one or more processors 1610, one or more memories 1620 coupled to the processors 1610, and one or more communication modules 1640 coupled to the processors 1610.

[0320] Communication module 1640 is used for bidirectional communication. Communication module 1640 has one or more communication interfaces to facilitate communication with one or more other modules or devices. The communication interface can represent any interface necessary for communication with other network elements. In some example embodiments, communication module 1640 may include at least one antenna.

[0321] As a non-limiting example, processor 1610 can be any type suitable for a local technology network and can include one or more of the following: general-purpose computer, special-purpose computer, microprocessor, digital signal processor (DSP), and processor based on a multi-core processor architecture. Device 1600 can have multiple processors, such as application-specific integrated circuit chips that are time-dependent on a clock synchronized with the main processor.

[0322] Memory 1620 may include one or more non-volatile memories and one or more volatile memories. Examples of non-volatile memories include, but are not limited to, read-only memory (ROM) 1624, electrically programmable read-only memory (EPROM), flash memory, hard disk, miniature optical disc (CD), digital video disc (DVD), optical disc, laser disc, and other magnetic and / or optical storage. Examples of volatile memories include, but are not limited to, random access memory (RAM) 1622 and other volatile memories that will not be maintained during power outages.

[0323] Computer program 1630 includes computer-executable instructions that are executed by an associated processor 1610. The instructions of program 1630 may include instructions for performing operations / actions of some example embodiments of this disclosure. Program 1630 may be stored in memory (e.g., ROM 1624). Processor 1610 can perform any suitable actions and processes by loading program 1630 into RAM 1622.

[0324] Example embodiments of this disclosure can be implemented by program 1630, thereby enabling device 1600 to execute as described in the reference. Figures 3 to 15 Any process discussed in this disclosure. Exemplary embodiments of this disclosure may also be implemented by hardware, or by a combination of software and hardware.

[0325] In some example embodiments, program 1630 may be tangibly contained in a computer-readable medium, which may be included in device 1600 (such as in memory 1620) or other storage devices accessible to device 1600. Device 1600 may load program 1630 from the computer-readable medium into RAM 1622 for execution. In some example embodiments, the computer-readable medium may include any type of non-transitory storage medium, such as ROM, EPROM, flash memory, hard disk, CD, DVD, etc. As used herein, the term "non-transitory" refers to a limitation on the medium itself (i.e., tangible, not tactile) rather than a limitation on data storage persistence (e.g., RAM vs. ROM).

[0326] Figure 17 An example of a computer-readable medium 1700 is shown, which may be in the form of a CD, DVD, or other optical storage disc. The computer-readable medium 1700 has a program 1630 stored thereon.

[0327] In general, the various embodiments of this disclosure can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. Although various aspects of the embodiments of this disclosure are shown and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that the blocks, apparatuses, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof, as non-limiting examples.

[0328] Some exemplary embodiments of this disclosure also provide at least one computer program product tangibly stored on a computer-readable medium (such as a non-transitory computer-readable medium). The computer program product includes computer-executable instructions (such as those included in a program module) that are executed in a device on a target physical or virtual processor to perform any of the methods described above. Typically, a program module includes routines, programs, libraries, objects, classes, components, data structures, etc., that perform a particular task or implement a particular abstract data type. The functionality of the program module can be combined or split among program modules as needed in various embodiments. The machine-executable instructions for the program module can execute in a local device or a distributed device. In a distributed device, the program module can reside in both local storage media and remote storage media.

[0329] Program code used to perform the methods of this disclosure may be written in any combination of one or more programming languages. The program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that, when executed by the processor or controller, the program code enables the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a stand-alone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0330] In the context of this disclosure, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, etc.

[0331] Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable media can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any suitable combination thereof. More specific examples of computer-readable storage media include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable optical disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0332] Furthermore, although operations are described in a specific order, this should not be construed as requiring that such operations be performed in the specific order shown or sequentially, or requiring that all shown operations be performed to achieve the desired result. In some cases, multitasking and parallel processing can be advantageous. Similarly, although several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure, but rather as a description of features that may be specific to particular embodiments. Unless explicitly stated otherwise, certain features described in the context of a single embodiment may also be implemented in combination in a single embodiment. Conversely, unless explicitly stated otherwise, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0333] Although this disclosure has been described in language specific to structural features and / or methodological actions, it should be understood that the disclosure as defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are disclosed as exemplary forms of implementing the claims.

[0334] Furthermore, the various implementations of this disclosure can be described with reference to the following terms, and their features can be combined in any reasonable manner.

[0335] Clause 1. A first means for communication, comprising: at least one processor; and at least one memory, the at least one memory storing instructions that, when executed by the at least one processor, cause the first means to: determine a CSI report based on channel state information (CSI) and according to a first model at the first means, the CSI report including compressed CSI; transmit the CSI report to a second means; and, based on at least one condition being met, perform an operation including at least one of: retransmitting the CSI report to the second means; or updating the CSI and the CSI report and transmitting the updated CSI report, wherein at least one condition includes at least one of: a first condition, the first condition being that a message from the second means indicates that a performance metric for channel decoding and / or decompression of the CSI report is less than a threshold, or a second condition, the second condition being that an evaluated performance metric for channel decoding and / or decompression of the CSI report at the second means is less than a threshold, the evaluated performance metric being evaluated by at least one of: the first model at the first means or an evaluator.

[0336] Clause 2. The first means pursuant to Clause 1, wherein the first means is further caused to: perform joint compression and channel coding on the channel state information using the first model to obtain a CSI report.

[0337] Clause 3. The first apparatus according to Clause 1, wherein the first apparatus is further caused to: compress the channel state information using the first model to obtain compressed channel state information; and perform channel coding and modulation on the compressed channel state information to obtain a CSI report.

[0338] Clause 4. The first apparatus according to Clause 3, wherein the channel state information includes at least one eigenvector or channel matrix of the channel, and the first apparatus is further caused to: compress the channel state information using a first model based on reference information, the reference information including at least one of the following: at least one eigenvalue of the channel, compressed channel state information feedback, signal-to-interference-plus-noise ratio (SINR), modulation and coding scheme (MCS), or the power of the transmitter at the first apparatus.

[0339] Clause 5. The first device pursuant to Clause 4, wherein the first device is further caused to: determine the evaluated performance metric based on at least one of the following: reference information, compressed channel state information, at least one previous CSI, or the configuration of a CSI report.

[0340] Clause 6. The first device pursuant to Clause 1, wherein the first device is also caused to transmit an evaluated performance metric to the second device.

[0341] Clause 7. The first device pursuant to Clause 1, wherein the first device is further caused to: transmit capability information of the first device to the second device, the capability information including at least one of the following: joint compression and channel coding and modulation for CSI reports, support for retransmission of CSI reports in the absence of triggering, or support for evaluation of performance metrics at the first device.

[0342] Clause 8. A first means pursuant to any one of Clauses 1-7, wherein the first means is further caused to: receive from the second means a configuration including at least one of the following: at least one threshold for a performance metric of channel decoding or decompression of CSI reports, a period for retransmitting CSI reports, or an instruction to perform an operation based on at least one condition being met.

[0343] Clause 9. The first device pursuant to Clause 8, wherein the first device is further caused to: select a target threshold from at least one threshold; and retransmit the CSI report to the second device based on an evaluated performance metric being below the target threshold.

[0344] Clause 10. A second means for communication, comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the second means to: receive a channel state information (CSI) report from a first means, the CSI report including compressed channel state information; perform channel decoding and decompression on the CSI report, at least the decompression being based on a second model at the second means; and transmit a first message to the first means based on a performance metric for channel decoding and / or decompression of the CSI report being less than a threshold, the first message indicating that the performance metric for channel decoding and / or decompression of the CSI report is less than the threshold.

[0345] Clause 11. A method for communication, comprising: at a first device, determining a CSI report based on channel state information (CSI) and according to a first model at the first device, the CSI report including compressed CSI; transmitting the CSI report to a second device; and, based on at least one condition being satisfied, performing an operation including at least one of: retransmitting the CSI report to the second device; or updating the channel state information and the CSI report and transmitting the updated CSI report, wherein at least one condition includes at least one of: a first condition, the first condition being that a message from the second device indicates that a performance metric for channel decoding and / or decompression of the CSI report is less than a threshold, or a second condition, the second condition being that an evaluated performance metric for channel decoding and / or decompression of the CSI report at the second device is less than a threshold, the evaluated performance metric being evaluated by at least one of: the first model at the first device or an evaluator.

Claims

1. A first device for communication, comprising: At least one processor; as well as At least one memory, the at least one memory storing instructions, the instructions, when executed by the at least one processor, cause the first device to: Based on Channel State Information (CSI) and a first model at the first device, a CSI report is determined, the CSI report including compressed CSI; The CSI report is transmitted to the second device; as well as If at least one condition is met, perform an operation that includes at least one of the following: Retransmit the CSI report to the second device; or Update the CSI and the CSI report, and transmit the updated CSI report. The at least one of the following conditions includes at least one of the following: The first condition is that a message from the second device indicates that the performance metric for channel decoding and / or decompression of the CSI report is less than a threshold, or The second condition is that the evaluated performance metric for channel decoding and / or decompression of the CSI report at the second device is below a threshold, and the evaluated performance metric is evaluated by at least one of the following: the first model or evaluator at the first device.

2. The first device according to claim 1, wherein the first device is further caused to: The first model is used to perform joint compression and channel coding on the channel state information to obtain the CSI report.

3. The first device according to claim 1, wherein the first device is further caused to: The channel state information is compressed using the first model to obtain compressed channel state information; and Channel coding and modulation are performed on the compressed channel state information to obtain the CSI report.

4. The first apparatus of claim 3, wherein the channel state information includes at least one feature vector or channel matrix of the channel, and the first apparatus is further caused to: Based on reference information, the channel state information is compressed using the first model, wherein the reference information includes at least one of the following: At least one characteristic value of the channel, Compressed channel state information feedback, Signal-to-Interference-plus-Noise Ratio (SINR) Modulation and coding scheme (MCS), or The power of the transmitter at the first device.

5. The first apparatus of claim 4, wherein the first apparatus is further caused to: determine the evaluated performance metric by the first apparatus or the evaluator based on at least one of the following: The reference information, The compressed channel state information, At least one previous CSI, or CSI report configuration.

6. The first device according to claim 1, wherein the first device is further caused to: The evaluated performance metric is transmitted to the second device.

7. The first device according to claim 1, wherein the first device is further caused to: Transmit capability information of the first device to the second device, the capability information including at least one of the following: Regarding the joint compression and channel coding and modulation in the aforementioned CSI report, Support for retransmission of the CSI report in the absence of a trigger, or Support for the evaluation of the performance metric at the first device.

8. The first device according to any one of claims 1-7, wherein the first device is further caused to: Receive a configuration from the second device including at least one of the following: At least one threshold for a performance metric of channel decoding or decompression of CSI reports. The period used for retransmitting the CSI report, or An instruction to perform the operation based on the satisfaction of at least one of the conditions.

9. The first device according to claim 8, wherein the first device is further caused to: Select a target threshold from the at least one threshold; and The CSI report is retransmitted to the second device if the evaluated performance metric is below the target threshold.

10. A second means for communication, comprising: At least one processor; as well as At least one memory, the at least one memory storing instructions, the instructions, when executed by the at least one processor, cause the second device to: The first device receives a Channel State Information (CSI) report, the CSI report including compressed channel state information; The CSI report is subjected to channel decoding and decompression, at least the decompression is based on the second model at the second device; and Based on a performance metric for channel decoding and / or decompression of the CSI report being lower than a threshold, a first message is transmitted to the first device, the first message indicating that the performance metric for channel decoding and / or decompression of the CSI report is lower than the threshold.

Images