Method and apparatus for wireless communication

Abstract

A method and apparatus for wireless communication are provided. The method comprises: a terminal device obtains a target CSI; the terminal equipment maps the target CSI into a first vector based on a first model, wherein the first vector corresponds to a first bit sequence; the terminal device sends the first bit sequence to a network device; wherein the pairing identifier corresponding to the first model is used for the network equipment to determine a second model, the second model is used for performing CSI reconstruction through the first bit sequence, and the length of the first bit sequence is determined according to single-layer payloads of one or more given layers; the size of the single layer payload is determined from a plurality of levels associated with the size of the load.

Description

Technical Field

[0001] This application relates to the field of communication technology, and more specifically, to a method and apparatus for wireless communication. Background Technology

[0002] In communication systems, channel state information (CSI) feedback is used by network devices to understand the wireless channel state, thereby optimizing transmission parameters to improve system performance. With the introduction of artificial intelligence (AI) technology, determining the effective CSI payload to achieve enhanced CSI feedback is a key technical issue that needs to be considered. Summary of the Invention

[0003] This application provides a method and apparatus for wireless communication. The various aspects related to the embodiments of this application are described below.

[0004] In a first aspect, a method for wireless communication is provided, comprising: a terminal device obtaining a target CSI; the terminal device mapping the target CSI to a first vector based on a first model, the first vector corresponding to a first bit sequence; the terminal device sending the first bit sequence to a network device; wherein, a pairing identifier corresponding to the first model is used by the network device to determine a second model, the second model being used to reconstruct the CSI using the first bit sequence, the length of the first bit sequence being determined based on a single-layer payload of one or more given layers, and the size of the single-layer payload being determined based on multiple levels related to the payload size.

[0005] In a second aspect, a method for wireless communication is provided, comprising: a network device receiving a first bit sequence from a terminal device, the first bit sequence corresponding to a first vector, the first vector being obtained through a mapping of a target CSI based on a first model; the network device determining a second model based on a pairing identifier of the first model; the network device reconstructing a CSI based on the first bit sequence and the second model; wherein the length of the first bit sequence is determined based on a single-layer payload of one or more given layers, and the size of the single-layer payload is determined based on multiple levels related to the payload size.

[0006] Thirdly, an apparatus for wireless communication is provided, the apparatus being a terminal device, the terminal device including a transceiver, a memory, and a processor, the memory for storing a program, the processor for calling the program in the memory, and controlling the transceiver to receive or send signals, so that the terminal device performs the method as described in the first aspect.

[0007] Fourthly, an apparatus for wireless communication is provided, the apparatus being a network device, the network device including a transceiver, a memory, and a processor, the memory for storing a program, the processor for calling the program in the memory, and controlling the transceiver to receive or transmit signals, so that the network device performs the method as described in the second aspect.

[0008] Fifthly, a communication apparatus is provided, comprising a unit or module for performing the method as described in the first or second aspect.

[0009] A sixth aspect provides an apparatus including a processor for calling a program from memory to perform the method as described in the first or second aspect.

[0010] A seventh aspect provides a chip including a processor for calling a program from memory, causing a device on which the chip is mounted to perform the method as described in the first or second aspect.

[0011] Eighthly, a computer-readable storage medium is provided having a program stored thereon that causes a computer to perform the method as described in the first or second aspect.

[0012] Ninth aspect, a computer program product is provided, including a program that causes a computer to perform the method as described in the first or second aspect.

[0013] In a tenth aspect, a computer program is provided that causes a computer to perform the method as described in the first or second aspect.

[0014] In this embodiment, the terminal device determines a first vector based on a first model and a target CSI, and sends the first bit sequence corresponding to the first vector to the network device. The length of the first bit sequence is determined based on the single-layer payload of one or more given layers. The size of the single-layer payload is determined based on multiple levels related to the payload size. The length of the first bit sequence is the size of the total CSI payload. Therefore, determining the total payload based on multiple specified levels not only adapts to coverage differences but also minimizes the number of payload options to a reasonable level, thereby reducing the workload of model training and development. Attached Figure Description

[0015] Figure 1 This is a system architecture example diagram of a wireless communication system to which the embodiments of this application are applicable.

[0016] Figure 2 This is a schematic diagram of the network architecture applicable to the embodiments of this application.

[0017] Figure 3A and Figure 3BThis is a schematic diagram of the structure of the wireless protocol stack applicable to the embodiments of this application.

[0018] Figure 4 This is a schematic diagram of a neuron in a neural network to which the embodiments of this application apply.

[0019] Figure 5 This is a schematic diagram of the neural network to which the embodiments of this application apply.

[0020] Figure 6 This is a schematic diagram of a convolutional neural network applicable to the embodiments of this application.

[0021] Figure 7 This is a schematic diagram of the AI ​​architecture used for CSI compression and decompression.

[0022] Figure 8 This is a comparative diagram of traditional CSI feedback and AI-based CSI feedback.

[0023] Figure 9 This is a flowchart illustrating the CSI compression and decompression process.

[0024] Figure 10 This is a schematic diagram of AI-based CSI compression.

[0025] Figure 11 This is a schematic diagram of encoding and decoding based on cumulative CSI.

[0026] Figure 12 This is a flowchart illustrating a method for wireless communication proposed in an embodiment of this application.

[0027] Figure 13 This is a schematic diagram of quantizing the first vector.

[0028] Figure 14 This is a schematic diagram of a device for wireless communication provided in an embodiment of this application.

[0029] Figure 15 This is a schematic diagram of another device for wireless communication provided in an embodiment of this application.

[0030] Figure 16 This is a schematic diagram of the structure of a communication device provided in an embodiment of this application. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0032] Communication system architecture Figure 1This is a system architecture example diagram of a wireless communication system 100 applicable to embodiments of this application. The wireless communication system 100 may include a network device 110 and a terminal device 120. The network device 110 may be a device that communicates with the terminal device 120. The network device 110 may provide communication coverage for a specific geographical area and may communicate with the terminal device 120 located within that coverage area.

[0033] Figure 1 An exemplary diagram illustrates a network device and multiple terminal devices, for example, Figure 1 Terminal devices 120a to 120j are included. Optionally, the wireless communication system 100 may include multiple network devices, and each network device may include other numbers of terminal devices within its coverage area; this embodiment does not limit this.

[0034] Optionally, the wireless communication system 100 may also include other network entities such as a network controller and a mobility management entity, which is not limited in this embodiment.

[0035] It should be understood that the technical solutions of the embodiments of this application can be applied to various communication systems, such as: 5th-generation (5G) systems or new radio (NR) systems, long-term evolution (LTE) systems, LTE frequency division duplex (FDD) systems, LTE time division duplex (TDD) systems, advanced long-term evolution (LTE-A) systems, enhanced 5G (5G advanced) systems, etc. The technical solutions provided in this application can also be applied to future communication systems, such as 6th-generation (6G) mobile communication systems, satellite communication systems, and so on.

[0036] The communication system in this application embodiment can be applied to carrier aggregation (CA) scenarios, dual connectivity (DC) scenarios, and standalone (SA) network deployment scenarios.

[0037] The terminal device in this application embodiment can also be referred to as user equipment (UE), access terminal, user unit, user station, mobile station, mobile station (MS), mobile terminal (MT), remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent, or user device. The terminal device in this application embodiment can be a device that provides voice and / or data connectivity to a user, and can be used to connect people, objects, and machines, such as a handheld device with wireless connectivity, vehicle-mounted device, etc. The terminal device in the embodiments of this application may be a mobile phone, tablet computer, laptop computer, handheld computer, camera equipment, mobile internet device (MID), wearable device, virtual reality (VR) device, augmented reality (AR) device, wireless terminal in industrial control, wireless terminal in self-driving, wireless terminal in remote medical surgery, wireless terminal in smart grid, wireless terminal in transportation safety, wireless terminal in smart city, wireless terminal in smart home, etc. Optionally, the terminal device may be used to act as a base station. For example, the terminal device may act as a scheduling entity, providing sidelink signals between UEs in vehicle-to-everything (V2X) or device-to-device (D2D) connections. For example, cellular phones and cars communicate with each other using sidelink signals. Cellular phones and smart home devices can communicate without relaying communication signals through base stations.

[0038] The network device in this application embodiment can be a device for communicating with terminal devices. This network device can also be called an access network device or a radio access network device, such as a base station (BS). In this application embodiment, the network device can refer to a radio access network (RAN) node or a next-generation RAN (NG-RAN) node (or device) that connects user equipment to a wireless network. A base station can broadly encompass, or be replaced by, various names including: NodeB, evolved NodeB (eNB), next-generation NodeB (gNB), relay station, transmitting and receiving point (TRP), transmitting point (TP), master station (MeNB), secondary station (SeNB), multi-mode radio (MSR) node, home base station, network controller, access node, wireless node, access point (AP), transmission node, transceiver node, baseband unit (BBU), remote radio unit (RRU), active antenna unit (AAU), remote radio head (RRH), central unit (CU), distributed unit (DU), positioning node, etc. A base station can be a macro base station, micro base station, relay node, donor node, or similar, or a combination thereof. A base station can also refer to a communication module, modem, or chip installed within the aforementioned equipment or apparatus. Base stations can also be mobile switching centers, devices that perform base station functions in D2D, V2X, and machine-to-machine (M2M) communications, network-side devices in 6G networks, and devices that perform base station functions in future communication systems. Base stations can support networks using the same or different access technologies. The embodiments of this application do not limit the specific technologies or device forms used in the network equipment.

[0039] Base stations can be fixed or mobile. For example, a helicopter or drone can be configured to act as a mobile base station, and one or more cells can move depending on the location of the mobile base station. In other examples, a helicopter or drone can be configured as a device to communicate with another base station.

[0040] In some deployments, the network device in this application embodiment may refer to a CU or a DU, or the network device may include both a CU and a DU. The gNB may also include an AAU.

[0041] Network devices and terminal devices can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; they can also be deployed on water; and they can also be deployed in the air on airplanes, balloons, and satellites. This application does not limit the scenario in which the network devices and terminal devices are located.

[0042] It should be understood that all or part of the functions of the communication device in this application can also be implemented by software functions running on hardware, or by virtualization functions instantiated on a platform (such as a cloud platform).

[0043] Figure 2 A schematic diagram of a network architecture 200 according to an embodiment of this application is illustrated. This network architecture 200 describes the network architecture of a 5G NR / LTE / LTE-A system, which can also be referred to as a 5G system (5GS) / evolved packet system (EPS) network architecture. The network architecture 200 includes at least one of the following: network device 110, terminal device 120, 5G core network (5GC) / evolved packet core (EPC) 210, home subscriber server (HSS) / unified data management (UDM) 220, and Internet service 230. Figure 2 The network devices and terminal devices in the diagram are illustrated using RAN and UE as examples, respectively.

[0044] like Figure 2As shown, network device 110 provides user plane and control plane protocol termination to terminal device 120. Network device 110 is connected to 5GC / EPC210 via an S1 / NG interface. 5GC / EPC210 includes a mobility management entity (MME) / authentication management field (AMF) / session management function (SMF) 211, other MMEs / AMFs / SMFs 214, a service gateway (S-GW) / user plane function (UPF) 212, and a packet data network gateway (P-GW) / UPF 213. MME / AMF / SMF 211 is the control node that handles signaling between terminal device 120 and 5GC / EPC210. ​​Generally, MME / AMF / SMF 211 provides bearer and connection management. All user Internet Protocol (IP) packets are transmitted through the S-GW / UPF212, which is itself connected to the P-GW / UPF213. The P-GW provides UE IP address allocation and other functions. The P-GW / UPF213 is connected to Internet service 230. Internet service 230 includes operator-compliant Internet Protocol services, specifically including the Internet, intranet, IP multimedia subsystem (IMS), and packet-switched streaming services. It is evident that network architecture 200 provides packet-switched services; however, those skilled in the art will readily understand that the various concepts presented herein can be extended to networks providing circuit-switched services or other cellular networks.

[0045] Figure 3A and Figure 3B The following are schematic diagrams of the wireless protocol stack structure of one embodiment of this application. Figure 3A and Figure 3B This introduction uses the 5G wireless protocol stack as an example. The 5G wireless protocol stack is divided into two planes: the user plane (UP) protocol stack and the control plane (CP) protocol stack. The user plane protocol stack contains the protocol suite used for user data transmission, while the control plane protocol stack contains the protocol suite used for control signaling transmission in the 5G system. The specific names of each protocol stack layer are as follows: like Figure 3AAs shown, the user plane protocol stack, from top to bottom, includes: the Service Data Adaptation Protocol (SDAP) layer, the Packet Data Convergence Protocol (PDCP) layer, the Radio Link Control (RLC) layer, the Medium Access Control (MAC) layer, and the Physical (PHY) layer.

[0046] like Figure 3B As shown, the control plane protocol stack, from top to bottom, includes: non-access stratum (NAS); radio resource control (RRC) layer, PDCP layer, RLC layer, MAC layer, and PHY layer.

[0047] It should be understood that the different layers in the above protocol stack have different functions, and they work together through inter-layer interaction to achieve communication between terminal devices and network devices. With the development of artificial intelligence technology, AI-assisted computing has permeated the processing implementation methods of the above protocol stack. For example, the scheduling algorithm of the MAC layer and the encoding / decoding algorithm of the PHY layer can apply artificial intelligence algorithms to improve the performance of communication algorithms.

[0048] As an example, Figure 3A and Figure 3B The wireless protocol architecture described herein is applicable to the terminal devices used in this application, such as UEs.

[0049] As an example, Figure 3A and Figure 3B The wireless protocol architecture described herein is applicable to network devices used in this application, such as gNBs.

[0050] It should be understood that the interpretation of the terminology in the embodiments of this application may refer to the TS36, TS37 and TS38 series of specifications of the 3rd generation partnership project (3GPP), but may also refer to the specifications of the Institute of Electrical and Electronics Engineers (IEEE).

[0051] To facilitate understanding, some related technical knowledge involved in the embodiments of this application is first introduced. The following related technologies are optional solutions and can be arbitrarily combined with the technical solutions of the embodiments of this application, all of which fall within the protection scope of the embodiments of this application. The embodiments of this application include at least some of the following contents.

[0052] Neural Networks AI research, exemplified by neural networks, has achieved significant results in many fields and will continue to impact people's lives and work for a long time to come. A neural network can be understood as a computational model composed of multiple interconnected neurons. In a neural network, the connection strength between nodes can be represented as the weighted values ​​corresponding to the input signals, also known as parameters. Each neuron performs a weighted summation of different input signals and outputs the result through a specific activation function. Neurons can achieve nonlinear mappings depending on the activation function.

[0053] by Figure 4 Taking the neuron shown as an example, the input of the neuron can be denoted as A, and each dimension of the input can be denoted as a. j The corresponding weighted value is denoted as w. j Where j takes values ​​of 1, 2, ..., n. The neuron's input can also be configured with a bias term to adjust the output, such as... Figure 4 The constant 1 in the input (corresponding to the weighting value denoted as b) is used. The weighting value, along with the summation units (SUs), enhances or weakens the input. The output of the SU can be input into the activation function f to obtain the output t.

[0054] Common neural networks include convolutional neural networks (CNN), recurrent neural networks (RNN), and deep neural networks (DNN).

[0055] The following text combines Figure 5 This application describes the neural network to which the embodiments are applicable. Figure 5 The neural network shown can be divided into three categories according to the position of different layers: input layer 510, hidden layer 520, and output layer 530. Generally speaking, the first layer is the input layer 510, the last layer is the output layer 530, and the intermediate layers between the first and last layers are hidden layers 520.

[0056] The input layer 510 is used to input data, which may be, for example, a received signal received by a receiver. The hidden layer 520 is used to process the input data, for example, to decompress the received signal. The hidden layer may also be called an intermediate layer. The output layer 530 is used to output the processed output data, for example, to output the decompressed signal.

[0057] See Figure 5 A neural network consists of multiple layers, each containing multiple neurons. Neurons between layers can be fully connected or partially connected. For connected neurons, the output of a neuron in one layer can serve as the input to a neuron in the next layer.

[0058] To facilitate understanding, we will use CNN as an example below, combined with... Figure 6 Examples of multiple layers in a neural network are provided. A CNN is a deep neural network with convolutional structures. For example... Figure 6 As shown, the structure of a CNN may include an input layer 610, a convolutional layer 620, a pooling layer 630, a fully connected layer 640, and an output layer 650. The convolutional layer 620, pooling layer 630, and fully connected layer 640 are the intermediate layers of this CNN.

[0059] It should be noted that, as Figure 6 The CNN shown is only an example of a convolutional neural network. In specific applications, convolutional neural networks can also exist in the form of other network models, and this application does not limit them.

[0060] CSI Feedback CSI feedback is a crucial component of wireless communication systems. CSI feedback from terminal devices helps network equipment accurately understand the wireless channel state, thereby optimizing transmission parameters to improve system performance. The following section uses 5G NR as an example to introduce CSI feedback.

[0061] In 5G NR, two types of codebooks are defined: Type 1 and Type 2, and a set of precoding matrices is used to describe channel state information. These precoding matrices are obtained based on measurements of the Channel State Information Reference Signal (CSI-RS). Type 1 codebooks are of conventional precision and are primarily used to support single-user multiple-input multiple-output (MIMO) transmission; Type 2 codebooks are of high precision and are primarily used to support multi-user MIMO transmission to improve system spectral efficiency. Terminal devices can transmit this information to network devices through a feedback mechanism so that the network devices can perform scheduling and precoding.

[0062] As an example, terminal devices can measure channel quality by analyzing the reference signal (CSI-RS) transmitted by the base station.

[0063] As an example, CSI parameters may include CQI, rank indicator (RI), and pre-coding matrix indicator (PMI), which will be explained in detail below.

[0064] CQI is a quantized value reported by the terminal device to the network device to indicate the channel quality of the downlink (DL). CQI reflects the maximum modulation and coding scheme that the terminal device can receive under the current channel conditions to ensure a certain bit error rate.

[0065] RI represents the number of parallel data streams transmitted in a MIMO system. RI reflects the multipath propagation characteristics of the channel and the rank of the channel matrix, and is usually related to the spatial degrees of freedom of the channel.

[0066] The Precoding Interface (PMI) is a feedback from the terminal device regarding the downlink channel state, instructing the network device which precoding matrix to select for signal transmission. The precoding matrix is ​​a linear transformation matrix used to process signals in a multi-antenna system. The network device determines the transmit precoding for the Physical Downlink Shared Channel (PDSCH), Physical Downlink Shared Channel (PDCCH), and CSI-RS, etc., based on the PMI reported by the terminal device.

[0067] Due to the complexity of wireless channels and the massive MIMO and high-frequency, high-bandwidth communication supported by 5G, CSI information typically has high dimensionality and rich detail, resulting in a very large amount of bandwidth resources required for feedback.

[0068] In 5G systems, physical layer and CSI feedback information play a crucial role in wireless communication. Relevant regulations mandate the compression of physical layer and CSI feedback information to reduce wireless resource consumption and improve system efficiency; this is known as physical layer and CSI feedback compression. Given the critical role of feedback information, effective compression is a key method for enhancing system performance.

[0069] To improve the efficiency and accuracy of CSI feedback, AI technology has been introduced to drive the development and application expansion of wireless communication technology. As an example, AI / machine learning (ML) based CSI compression technology offers greater flexibility, lower distortion, and stronger adaptability. Therefore, leveraging the powerful capabilities of AI technology, more efficient and lower-distortion CSI compression can be achieved.

[0070] The following is combined with Figure 7 and Figure 8 This paper explains the CSI feedback process based on AI / ML models. Figure 7 The AI / ML architecture shown is a model for compressing / decompressing the precoding matrix of a single MIMO layer. (By...) Figure 7 It can be seen that the model has an output / input linear layer and a quantization / de-quantization layer.

[0071] See Figure 7 On the encoding side, the input parameters of the embedding layer include the number of tokens. N token And the dimension of each token. d token Based on positional encoding, the output parameters of the embedding layer, N token And model dimension (dim) d model Perform the calculation and input the result repeatedly. N TF-Enc The next processing layer. Within this layer, normalization and multi-head attention are performed sequentially. After some computation, normalization and multi-layer perceptron (MLP) processing are performed again. This process is repeated multiple times to obtain a new... N token and d model The output parameters (dimensions) of the output linear layer are then input into the output linear layer. z dim The input is fed into the quantization layer for quantization processing. The total payload size is:N payload = N bit z dim .

[0072] See also Figure 7 On the decoding side, the result from the encoding side is input into the dequantization layer, and dequantization is used to obtain... z dim The data is then input into the input linear layer. Based on positional encoding, the output parameters of the input linear layer are... N token and d model Perform the calculation and input the result repeatedly. N TF-Dec The next processing layer. The processing procedure in the processing layer is the same as that in the encoding layer, and will not be described again. The output of the processing layer includes... N token and d model The output is fed into the output linear layer to obtain the decoded product. N token and d token .

[0073] like Figure 7 As shown, the output / input layer and the dimension are z dim The latent space representation is associated with each entry in the latent space representation, which is represented by... N bit Bit quantization. For a MIMO layer, the CSI payload size is determined by the tuple { z dim , N bit} is given, that is N payload = N bit z dim .

[0074] In MIMO / CSI evaluation, the squared generalized cosine similarity (SGCS) is used to measure the consistency of the precoding subspace. The SGCS for each layer is the average SGCS of the bits in several target CSI formats, corresponding to the feature vectors of layers 1 through 4 with floating-point numbers of 32. Furthermore, to allow for different levels of precision (similar to designing enhancement type 2 (eType II) with different parameter combinations), a limited number (e.g., 4) of { z dim, N bit} For candidate values. More choices may impose unnecessary implementation work on terminal and network devices. Wherein, { z dim , N bit The detailed value of} can be further studied.

[0075] Figure 8 Taking bilateral compression as an example, a comparison was made between traditional CSI feedback and AI / ML-based CSI feedback. (See also...) Figure 8 In both CSI feedback methods, the input is H. The UE implementation in traditional CSI feedback corresponds to the AI / ML encoder in AI / ML-based CSI feedback, while the PMI reconstruction on the network device side corresponds to the AI / ML decoder.

[0076] like Figure 8 As shown, in traditional CSI feedback, the terminal device performs channel and interference measurements based on CSI-RS and CSI interference measurement (CSI-IM) signals, and then uses the PMI search algorithm to search for the optimal precoding matrix (also known as the precoder matrix) from a specified codebook. Subsequently, the terminal device can report the corresponding PMI indices i1, i2, etc., via uplink control information (UCI). The network device looks up the corresponding PMI codewords based on the reported indices, thereby reconstructing the precoding matrix. For example, the terminal device can perform CSI reconstruction according to relevant protocols (e.g., protocol TS 38.214), obtaining the precoding matrix. W It can be used to calculate CQI.

[0077] In AI / ML-based CSI feedback, the terminal device first performs singular value decomposition (SVD). Then, the terminal device uses an encoder to compress the target CSI (e.g., a sub-band granular space-frequency domain precoding matrix) into a latent vector / bit and sends the quantized latent information (z) to the network device. The network device then reconstructs the precoding matrix using a pairwise decoder model. WThe encoder and decoder are a pair of co-trained "paired models," and using pairing IDs ensures consistency between them during the inference phase. Therefore, it is necessary to specify the mapping subject to an ID.

[0078] Depend on Figure 8 It is evident that in AI / ML-based CSI feedback, the AI / ML decoder can reconstruct the CSI from the network device side, rather than relying on input from a traditional PMI search algorithm. This approach explicitly standardizes the exact CSI obtained by the network device and the precoding on which the CQI is computed. Furthermore, the network device can specify the interpretation of the PMI index on the precoding matrix.

[0079] for Figure 8 In AI / ML-based CSI feedback, the interpretation or mapping from CSI reports to reconstructed CSI can use a model on the network device side (also known as a CSI reconstruction sub-model or CSI decoder). This model can explicitly specify whether a mapping is feasible, as model development is a vendor implementation choice. Conversely, since the pairing ID is used in the inter-vendor training collaboration phase to achieve compatibility between the encoder and decoder, it can also indicate the encoder-decoder pair used for inference to maintain consistency between the training and inference phases. The mapping between CSI reports and reconstructed CSI can be specified via the pairing ID. The pairing ID is configured in the CSI report configuration. To simplify inter-vendor training, it is necessary to align the potential message dimensions, structure, and their mapping to the UCI payload. Otherwise, there are too many implementation options for mapping UCI bits (e.g., 128 bits) to potential messages with different lengths or representations. To specify detailed configurations for these aspects, codebooks for quantization can be exchanged during the inter-vendor collaboration phase.

[0080] In the CSI feedback process, AI-based CSI feedback enhancement mainly focuses on the following two directions: 1. Spatial-frequency domain CSI compression based on a bilateral AI model (also known as a two-sided model); 2. Temporal CSI prediction based on a one-sided AI model (also known as a one-sided model).

[0081] Unlike the space-frequency domain CSI compression scheme based on a bilateral AI model, the time-domain CSI prediction based on a unilateral AI model uses a unilateral model on the terminal device side, that is, the CSI prediction model is deployed at the terminal device. This scheme can input historical CSI measurement data into the CSI prediction model and then output the predicted CSI value for future time moments, thereby effectively solving the problem of timeliness of CSI feedback.

[0082] The space-frequency domain CSI compression scheme based on a bilateral AI model involves deploying a CSI generation model on the terminal device side and a corresponding CSI reconstruction model on the network device side. These two models work together to complete the tasks of CSI compression, feedback, and reconstruction. This ensures that while CSI is significantly compressed at the terminal device side, the network device can obtain more accurate CSI data, thereby optimizing resource scheduling and other operations. Specifically, in the bilateral AI model, the terminal device performs CSI compression based on the AI ​​model, while the network device performs CSI reconstruction based on the AI ​​model. The purpose of the network device reconstructing the CSI is to restore the compressed bits reported by the terminal device into a precoding matrix that can be directly used for the air interface, enabling the network device to make consistent and executable physical layer / scheduling decisions simultaneously. The compressed bits include a latent vector, which is a low-dimensional real-valued representation of the target CSI (such as a sub-band granular space-frequency domain precoding matrix) transformed by the encoder of the compression model on the terminal device side. This vector is not equal to the physical channel itself, nor is it the final precoding to be transmitted; it is merely a compact intermediate representation. The latent vector is quantized into bits and placed in Part 2 of the UCI, then reported to the network device. The network device uses paired decoders to reconstruct the precoding matrix. The network device can interpret and decode the reported latent vector according to the pairing ID into a sub-band granular space-frequency domain precoding matrix, which is used as transmit precoding for PDSCH / PDCCH / CSI-RS, etc. With executable precoding, the network device can perform actual scheduling actions such as RI / layer selection, multi-user (MU) MIMO (MU-MIMO) pairing / interference management, and modulation and coding scheme (MCS) selection.

[0083] In the bilateral model, CSI feedback enhancements also include support for CSI spatial / frequency compression without a time dimension, i.e., "Case 0". Case 0 refers to spatial frequency (SF) domain CSI compression, meaning the compression does not include the time domain, such as the beam delay domain. This feedback enhancement includes improvements in the model pairing process, inference aspects, and data collection processes. The model pairing process includes ID and applicability reporting. Data collection includes data collection from both the network device side and the terminal device side for training. Data collection is a key research issue for achieving both spatial-frequency domain CSI compression based on bilateral AI models and time-domain CSI prediction based on unilateral AI models.

[0084] The inference aspects include target CSI type, measurement and reporting configuration, CQI RI determination, payload determination, quantization configuration codebook, UCI mapping, CSI processing standards and timelines, and CSI reporting priority rules. For the inference aspect, the target CSI type is the most important topic because payload determination, data collection formats, and monitoring mechanisms all depend on the determination of the target CSI type.

[0085] The target CSI refers to the CSI that the network (NW) aims to reconstruct, which is the CSI that will ultimately be reconstructed without loss during the CSI compression and reconstruction process. For CSI compression in the space-frequency domain (i.e., the spatial-frequency domain), there are four target CSI types: precoding matrix feedback in the space-frequency domain, pre-decoding matrix feedback in the angular delay domain, channel matrix feedback in the space-frequency domain, and channel matrix feedback in the angular delay domain. The angular delay domain is the angle-delay domain, and the channel matrix can be an explicit channel matrix. Precoding matrix feedback is implicit CSI feedback, while explicit channel matrix feedback is explicit CSI feedback. Explicit CSI feedback is a general CSI feedback scheme, not limited to the scope of AI / ML. However, this scheme requires defining the terminal device's receive (Rx) port, frequency granularity, channel gain, or eigenvalue feedback, etc. These aspects require comprehensive research within the 6G domain.

[0086] For example, in Rel-19 Type-I and Type II codebooks with up to 128 ports, a maximum of four CSI-RS resources can be aggregated, with each CSI-RS resource configured with a maximum of 32 ports, forming a maximum of 128 CSI ports. For CSI with up to 32 ports, a CSI report configuration can only configure one CSI-RS resource, with each CSI-RS resource configured with a maximum of 32 ports. For 48, 64, and 128-port CSI, a CSI report configuration can configure a maximum of four CSI-RS resources, with each CSI-RS resource configured with a maximum of 32 ports, and these CSI-RS resources are aggregated to form a maximum of 128 CSI ports.

[0087] To facilitate understanding, the following will be combined with... Figure 9 and Figure 10 The processing module in the document provides an example of the target CSI type. Figure 9 This is a typical CSI processing module for terminal equipment. Figure 10 This is an example of CSI compressed using AI.

[0088] like Figure 9As shown, for CSI feedback, the UE side includes modules (1) to (4), which are, in order, a processing module for the channel matrices per subband, an SVD / eigenvalue decomposition (EVD) module, an angle-delay projection and basis selection processing module, and an encoder. The NW side module (5) is a decoder, and its output is the target CSI.

[0089] For different target CSI types, the terminal device will... Figure 9 Certain modules in the processing flow may be added or removed. For example, when the target CSI type is an angle-delay domain precoding matrix, the terminal device will apply... Figure 9 All processing modules in the target CSI are skipped. For example, when the target CSI type is an explicit channel matrix in the space-frequency domain, all modules except modules (1) and (4) are skipped. For example, when the target CSI type is a precoding matrix in the space-frequency domain, module (3) is bypassed and the feature vector from module (2) is directly encoded. For example, when the target CSI type is an explicit channel matrix in the angle-delay domain, the SVD processing in module (2) is skipped and the channel matrix is ​​directly projected into the angle-delay domain. Therefore, different target CSI types require the terminal device to make diverse trade-offs between complexity, performance and feedback overhead.

[0090] like Figure 10 As shown, the AI-compressed CSI is related to the quantization scheme. See also... Figure 10 The number of samples per-layer input can be ,in, This indicates the number of transmit antenna ports on the terminal device. This indicates the number of sub-bands. These samples, after being input into the AI ​​encoder, will yield... z dim There are one output. After quantization, the output payload is... z Q .

[0091] Figure 10 The quantization scheme used needs to be known by both network devices and terminal devices. To achieve this, the quantization module can use a predefined quantization codebook or a network-shared quantization codebook. If a predefined quantization codebook is used, the scalar codebook can be the existing eType-II W2 quantization codebook. If a network-shared quantization codebook is used, multiple layers or modules in the entire neural network share the same codebook, thereby improving parameter utilization efficiency and reducing redundancy.

[0092] Figure 10 The quantization method used can be uniform or non-uniform scalar quantization, or vector quantization. The following explains two quantization methods using different quantization codebooks.

[0093] For scalar quantization (SQ), the network side needs to share... z dim The number of bits used by each scalar output in the scalar output is ( N bits ), and the threshold (i.e., the quantization interval) used to map codewords to scalar intervals. Therefore, when Figure 10 When using scalar quantization, z Q = z dim × N bits .

[0094] Vector quantization (VQ) quantizes a set of values ​​(i.e., vectors) as a whole, which typically captures the internal dependencies of the data better and achieves better compression. Therefore, the network may need to share more parameters. These parameters include, for example, the dimensions of the vector. v dim ( v dim ≤ z dim ), the number of bits used by each vector ( N bits ), and a codebook for mapping codewords to vector space. At this point, z Q =( z dim / v dim )× N bits Therefore, scalar quantization can be seen as a special case of vector quantization, where each individual scalar value (such as weights or activation values) is quantized independently. In other words, scalar quantization can be considered as... v dim Vector quantization at =1.

[0095] It should be understood that for network-shared quantization codebooks, the format for such codebook sharing must be defined, such as the parameter format used to represent scalar intervals and vector spaces. Network-shared quantization codebooks can be scalar quantization codebooks, meaning multiple layers or modules use the same fixed scalar codebook for quantization. For example, when deploying on edge devices, it's desirable to use a unified, static codebook to compress the values ​​of all layers to save storage space. Network-shared quantization codebooks can also be vector quantization codebooks, where a set of values ​​(i.e., vectors) is quantized as a whole using a fixed (non-learnable) codebook. This codebook can contain multiple predefined vectors. The input vector is mapped to the codebook vector closest to it. Vector quantization captures the dependencies between elements within a vector. When using vector quantization, all layers in the entire network share the same vector codebook; that is, a common, global discrete latent representation can be shared across multiple components to improve efficiency and consistency.

[0096] The bilateral model of CSI compression can include an encoder and a decoder. For ease of understanding, the following will combine... Figure 11 An illustrative example of a two-sided model for CSI compression is provided. Figure 11 Taking the encoder deployed at the transmitting end and the decoder deployed at the receiving end as examples, the inputs and outputs at times tn-2, tn-1 and tn are introduced respectively.

[0097] like Figure 11 As shown, the inputs at the three time points are V tn-2 V tn-1 V tn The outputs are respectively , , For time tn-1, the inputs to both the encoder and decoder include the accumulated CSI from time tn-2, i.e., the historical CSI. For time tn, the inputs to both the encoder and decoder include the accumulated CSI from time tn-1, i.e., the historical CSI.

[0098] Depend on Figure 11 It is known that the encoder input consists of multiple parts: the CSI of the current time slot and the cumulative CSI of the previous few time slots. Therefore, the compressed CSI contains information about both the current and cumulative CSI. In each time slot, the encoder processes CSI containing information related to both the current and cumulative CSI. Thus, the encoder's main task is to compress the current CSI (… H t ) and past cumulative CSI ( C t Feature fusion is performed to generate a compact compressed CSI ( Zt For the decoder, the input consists of two parts: the compressed CSI of the current time slot and the cumulative CSI of the previous few time slots. The decoder is mainly used to recover the CSI of the current time slot.

[0099] Figure 11 One scenario in CSI compression involves compressing the CSI of the current time slot and the accumulated CSI of the previous few time slots. Another scenario involves the terminal device generating CSI feedback based on a CSI prediction model, while the network device reconstructs the CSI based on a reconstruction model. For example, the CSI generation sub-model on the terminal device side predicts the CSI for time slot n based on an observation window and sends the prediction results to the network device. The CSI reconstruction sub-model on the network device side can then perform CSI reconstruction.

[0100] It should be understood that the above introduction to CSI feedback based on 5G is only an example. The development of 5G mobile communication systems can provide support for subsequent new communication technologies. For example, new waveforms providing coverage in the terahertz band of 6G mobile communication technology. Other new technologies include multi-antenna transmission technologies applying full-dimensional MIMO (FD-MIMO), array antennas, and large antennas; metamaterial-based lenses and antennas for improving the coverage of terahertz band signals; high-dimensional spatial multiplexing technologies using orbital angular momentum (OAM); reconfigurable intelligent surfaces (RIS); full-duplex technologies to increase the frequency efficiency of 6G mobile communication technology and improve system networks; AI-based communication technologies for system optimization by leveraging satellites and AI from the design phase and internalizing end-to-end AI support; and next-generation distributed computing technologies for implementing services at complexity levels exceeding the limits of UE operational capabilities by utilizing ultra-high-performance communication and computing resources. Therefore, CSI prediction and compression for the air interface are important research directions in 6G and subsequent communication system technologies.

[0101] As mentioned earlier, payload determination is a research direction in CSI feedback enhancement, specifically inference process enhancement. Payload determination involves several aspects, including payload size configuration, payload size determination for the reported rank, determining / selecting payload bits from the quantization output, and mapping the determined payload to the UCI. The payload-to-UCI mapping includes mapping to CSI Part 1, mapping to CSI Part 2, and partial discarding of CSI Part 2.

[0102] In research on the scalability of AI / ML model architectures, mixed training of all payload options should be considered. Some end-device (or network device) vendors may employ specific encoder output layers (or specific decoder input layers) to support various payload configurations. However, too many options can lead to memory issues or additional model loading delays.

[0103] Furthermore, taking payload size configuration as an example, AI-CSI (AI-based CSI) requires parameter combinations to support various payload sizes / CSI resolutions. This is because a single feedback size cannot be applied to all situations. For instance, for terminal devices located at the cell edge with limited uplink resources, the network side may configure a smaller payload size to reduce uplink overhead.

[0104] For example, the eType II CSI codebook has eight parameter combinations, each with a specific CSI reporting overhead. Defining multiple payload options is beneficial considering uplink coverage at different cell locations, as terminal devices in the cell center can benefit from the higher spectral efficiency achieved by high-resolution CSI feedback, while low-resolution CSI feedback and low rank may be sufficient for terminal devices at the cell edge. This principle can also be applied to payload determination for AI / ML-based CSI feedback. However, the eight payload options in eT2 (eType II) may be overly granular. For a specific payload configuration, if the payload at each layer also depends on the number of subbands (i.e., the number of frequency domain (FD) basis bases, the number of non-zero coefficients), the number of ports, and the rank, it leads to an increase in granularity. For AIML, highly granular payload options will significantly increase the workload of model training and development.

[0105] In summary, AI / ML-based CSI feedback needs to support a variety of payload options to support different application scenarios, but too many options may place higher demands on model loading or model computation capabilities.

[0106] To address the aforementioned issues, this application proposes a method for wireless communication. In this method, after obtaining a target CSI, the terminal device can map it to a first vector based on a first model and send the first bit sequence corresponding to the first vector to the network device. The length of the first bit sequence is the size of the total payload, and this length is determined based on the single-layer payload of one or more given layers. The size of a single payload is determined based on multiple levels related to the payload size. Based on these multiple specified levels, not only can coverage differences be adapted, but the number of payload options can also be reduced to a reasonable level, thereby reducing the workload of model training and development.

[0107] To facilitate understanding, the following will be combined with... Figure 12 The methods proposed in the embodiments of this application will be described in detail. Figure 12 It is presented from the perspective of the interaction between terminal devices and network devices.

[0108] The terminal device can be any of the terminals or terminal-side devices mentioned above, such as a UE.

[0109] In some embodiments, the terminal device can communicate with a network device. As one embodiment, the terminal device can receive reference signals sent by the network device for channel estimation. For example, the terminal device can receive CSI-RS sent by the network device.

[0110] In some embodiments, the terminal device may be a communication device that deploys the first model. The first model deployed on the terminal device may be used for inference or prediction. When the first model is deployed on a terminal-side device corresponding to the terminal device, the terminal-side device may be an auxiliary device communicating with the terminal device, or it may be another communication device relative to the network device side. For example, the terminal-side device may be a network relay.

[0111] As one embodiment, the terminal-side device can be a server that trains the first model, i.e., the first device. The first device can perform inference based on the deployed first model and send the inference results or prediction results to the terminal device. For example, the first device can be a non-3GPP entity belonging to the terminal device vendor, such as an OTT server.

[0112] As an example, the first model can be a model related to CSI compression. The first model may include a CSI encoder and / or a CSI generation sub-model. For example, the first model may be an AI / ML model. Alternatively, the first model may be a baseline model.

[0113] As an example, based on the pairing ID, the first model can work in conjunction with a second model deployed on the network device side to achieve CSI compression based on a bilateral AI model, such as space-frequency domain CSI compression.

[0114] In some embodiments, the terminal device may be equipped with at least one transmitting antenna and at least one receiving antenna for wireless transmission.

[0115] In some embodiments, the cell where the terminal device is located is the first cell. The terminal device communicates with the network equipment of the first cell.

[0116] The network device can be any of the network devices or network-side devices described above, such as a base station. The network device can be the network device corresponding to the first cell. The network device can provide services to all terminal devices in the first cell.

[0117] In some embodiments, the network device may deploy a second model. The second model may be an AI / ML model. As one example, the second model may work in conjunction with the first model for joint inference. For instance, the first model is used for CSI generation, and the second model is used for CSI reconstruction. When the second model is used for CSI decompression, it may also be referred to as a CSI decoder or a CSI reconstruction sub-model.

[0118] As one example, the network device can collect data to construct a dataset related to an AI / ML model. As another example, the network device can sample the collected data at regular time intervals to obtain multiple samples arranged in a specific time sequence.

[0119] As one example, a network device can train an AI / ML model deployed on a terminal device and then send the trained model to the terminal device. For instance, the network device can train a first model and then send the trained first model to the terminal device.

[0120] It should be noted that in the embodiments of this application, the model can also be replaced by a function, for example, the first model is the first function.

[0121] See Figure 12 , Figure 12 The flowchart shown includes steps S1210 to S1230, which are described below. It should be noted that... Figure 12 The method shown may also include other steps, which are not limited in the embodiments of this application.

[0122] In step S1210, the terminal device obtains the target CSI. The terminal device can obtain the target CSI through the measurement results of CSI-RS or other reference signals, or it can obtain the target CSI through CSI prediction results. This application embodiment does not limit this.

[0123] In some embodiments, the target CSI can characterize at least one of the following: a precoding matrix in the space-frequency domain; a precoding matrix in the angle-delay domain; a channel matrix in the space-frequency domain; and a channel matrix in the angle-delay domain.

[0124] In some embodiments, a target CSI can be one or more subsets of a target CSI set. With a fixed port / array configuration and bandwidth / total number of subbands, a target CSI set can be obtained by collecting / generating a batch of target CSI samples (such as a subband-granular space-frequency domain precoding matrix or its equivalent representation). As an example, the port configuration can be represented as "N1×N2×2", for example, N1 horizontal × N2 vertical × 2 polarization. Here, N1 can represent the number of horizontal antenna ports, and N2 can represent the number of vertical antenna ports. As an example, each target CSI set can contain N4 samples. N4 represents the number of target CSI samples.

[0125] In some embodiments, a target CSI set can map to multiple CSI feedback sets. Each CSI feedback set corresponds to a specific CSI payload and subband configuration, or supports a one-to-many mapping between a target CSI sample and multiple CSI feedback samples. For example, the same target CSI sample, encoded by different feedback sets, can yield M different feedback results (M is a positive integer). Therefore, there is a 1:M mapping between a target CSI sample and multiple CSI feedback samples. Thus, a 1:M mapping from a target CSI set to multiple CSI feedback sets is supported.

[0126] As an example, a CSI feedback set can correspond to an air interface feedback format. The air interface feedback format can be represented by at least one of the following parameters: payload size, subband mask / selection, quantization configuration / codebook.

[0127] As an example, the M CSI feedback sets correspond to M different subbands and payload configurations. Different port configurations or bandwidths may require different target CSI sets. The target CSI set can be associated with the pairing ID corresponding to the model, as detailed in step S1220.

[0128] In step S1220, the terminal device maps the target CSI to a first vector based on the first model. The first model can be an AI / ML-based encoder model, i.e., a first AI / ML encoder model. The first model can also be referred to as an AI / ML-based CSI generation sub-model.

[0129] In some embodiments, the first model is used to implement AI / ML-based CSI compression. Performance targets for CSI compression can be calculated along the following dimensions: per layer (when the target CSI is a precoding matrix), per CSI feedback report configuration, and per group (N1, N2) antenna port configuration. Average SGCS and normalized mean square error (NMSE) are supported as performance targets for exchanged data. For example, network devices can choose to send SGCS values ​​at the X percentile.

[0130] As an example, the NMSE performance metric is used to test the encoder on the end-device side after training. This performance metric also considers the actual operating mode during the inference phase, where the latent message (CSI feedback) undergoes quantization and dequantization before being used as input to the network device-side decoder. The associated target CSI and CSI feedback pair, or data instance, can be represented as... n 4∈{1,…,N4}.

[0131] For example, for a given layer ∈{1,…, v Sub-band n 3∈{1,…,N3} and data examples n 4∈{1,…,N4}, SGCS can be defined as: ; in, The first for network device-side decoder reconstruction Layer, First n 3rd sub-band, the first n 4. Metrics for the precoder of data instances; The PMI representation used to correspond to the real target CSI represents the precoder metric.

[0132] For example, for a given layer ∈{1,…, v} and data examples n 4∈{1,…,N4}, NMSE can be defined as: ; in, This represents the CSI feedback value calculated by the terminal device side using CSI compression (after the first...). Layer, First n 4 data instances after dequantization). This indicates the corresponding CSI feedback value provided by the network device (if the switched CSI feedback is a floating-point value).

[0133] As an example, SGCS can be calculated by averaging over N3 subbands, or by averaging or statistically distributing over N4 data instances in the dataset.

[0134] As an example, the average NMSE can be the average of N4 data instances paired with the associated target CSI and CSI feedback.

[0135] As an example, the performance metrics are calculated on the terminal device side, and the result is not the performance target. The performance metric calculation for each layer only applies when the target CSI is a precoding matrix.

[0136] Mapping the target CSI to a first vector based on the first model can be understood as follows: the target CSI is the input parameter of the first model, and the first vector is the output parameter obtained based on the input parameter.

[0137] The first vector corresponds to the first bit sequence. This first bit sequence carries the CSI report fed back by the terminal device, which is the payload corresponding to the target CSI. Determining the CSI payload includes determining the total payload size, i.e., the length of the first bit sequence. The number of bits in the payload is determined after quantization.

[0138] In some embodiments, the length of the first bit sequence is... Figure 7 In z dim and N bit related. z dim Candidate values ​​can include 32, 64, and 128. N bit The candidate values ​​can include 2, 3, and 4. Based on these candidate values, we can select from the following options: z dim , N bit Candidate values ​​for}.

[0139] As an example. N bit and z dim The design of the values ​​can consider the following factors: avoiding linear scaling of payload size with the reporting level, and limiting the variation in payload size between different levels (e.g., like Rel-16 eType II). Lower tiers (e.g., tiers 1 and 2) are configured with smaller payload sizes than higher tiers (e.g., tiers 3 and 4). A finite number of { z dim , N bitFor example, four candidate combinations of values. CSI report configuration includes measurement resource configuration, subband size, rank limits, etc.

[0140] In some embodiments, the first bit sequence supports codebook indexing / quantization representation based on ports for each subband and each layer. As one example, the first bit sequence supports representing the target CSI type precoding matrix using port subband fields, i.e., the target CSI is expressed using "port × subband" fields. As an example, the first bit sequence may include each layer (e.g., layer...) ) in each sub-band (such as sub-band) On, targeting Precoded vector index of each CSI-RS port.

[0141] For example, layers N sb Individual belt and P The precoding matrix on each CSI-RS port can be ,in, It is a P×1 vector.

[0142] As an example, the first bit sequence may only support sub-band PMI reporting. That is, CSI feedback does not force the reporting of wideband PMI, and the terminal device may only provide the optimal precoding index (or equivalent quantization weight) on a sub-band basis.

[0143] For bidirectional CSI compression use cases, pair IDs (i.e., IDs associated with the exchanged dataset / model parameters) can be used for inference configuration. Pair IDs can be associated with multiple quantization parameters and different CSI payload sizes. For inference configuration, a trade-off between accuracy / payload size and performance can be considered based on multiple quantization configuration candidate values.

[0144] The first bit sequence can be obtained by quantizing the first vector. The quantized first bit sequence is the CSI payload. In some embodiments, the terminal device can quantize the first vector according to the quantization configuration to obtain the first bit sequence. The quantization configuration can be pre-configured by the network device or determined according to relevant signaling.

[0145] Regarding quantization configuration, independent quantization can be applied to each segment. For example, the first vector can be obtained from the first bit sequence based on differential quantization. In eType II and Type II codebooks, differential quantization is applied to entries or subbands within the same polarization, respectively. This differential quantization provides marginal benefits with additional computational cost. Therefore, segment-independent quantization is preferred. Similarly, in MIMO, each layer in the parallel data stream should consider independent quantization. The quantization codebook can be determined by network-side model training in inter-vendor collaborative direction A, and then exchanged from the network device to the terminal device for model training and inference on the terminal device side.

[0146] As an example, the compressed CSI is broken down into measurable bits. Each layer in the MIMO transmission can output a real-valued latent vector, which is then segmented and quantized to obtain the first bit sequence.

[0147] As an example, the quantization of the first vector can be performed based on a determined payload. During CSI payload determination and quantization, the real-valued latent message at a given layer... z l It can be used as its total length / dimension d z To describe each paragraph L seg All entries can be used Q Reporting is done in bits, specifically per quantized bit segment. The following example uses a vector from a given layer, combined with... Figure 13 An example is provided.

[0148] See Figure 13 Dimensions of potential messages for a given layer d z =64, the 64 entries are respectively z 1 to z 64 .based on L seg The 64 entries are divided into segments with a count of 2, meaning each segment contains 2 real-valued entries. For example, the entries... z 1 and z 2 is the first paragraph. Each paragraph consists of... Q =4 bits are used for quantization (quantized by Q =4 bits), meaning that the two entries are jointly quantized by 4 bits.

[0149] exist Figure 13 In the middle, the total number of segments is from S = d z / L seg=32 is given. Therefore, the effective load of each layer is given by... S × Q =128 bits are given. With this setting... L seg =1 means scalar quantization (for real values).

[0150] As an example, the first bit sequence can be determined based on the payload configuration. For example, the length of the first bit sequence (the size of the total payload) can be determined according to the payload configuration. The payload configuration can be achieved through { d z , L seg , Q The payload is described by a combination of these configurations, and there may be multiple specified payload configurations. For example, four payload configurations may be sufficient to cover a variety of uplink coverage areas. Thus, each payload configuration corresponds to a unique payload per layer, such as 64 / 128 / 256 / 512 bits. Therefore, single-layer payloads no longer depend on subband configurations or port configurations.

[0151] In the above embodiments, the network device can configure a combination of parameters in the CSI report configuration to indicate the payload configuration. By combining { d z , L seg , Q When described using the rank assumption, the payload of each layer is universal across all ranks; that is, the same payload configuration applies to all layers across all ranks. In this case, the total payload is: Q × rank × d z / L seg ,in, rank It represents the rank (number of layers).

[0152] In the above embodiments, the network device can configure a parameter combination in the CSI report configuration, which can be used to determine the payload parameter combination (PC) across all layers and ranks, such as { D l,c , Q l ,c}or{ D l,c , rank , Q l,c}. The payload parameter combination is for each layer / each column of the corresponding precoding matrix ( cFor each layer, the corresponding column in the precoding matrix (which can be simplified to the column corresponding to each layer) requires feedback parameters. D l ,c The number of bits related to selection / indexing (e.g., codebook index, group index, direction coarse selection). Q l,c This represents the number of bits used for fine-grained amplitude / phase / coefficient quantization based on the layer and corresponding column.

[0153] As an example, the network device can limit the combination of payload parameters in all layers and all corresponding columns (for each layer) ({ D l,c , Q l,c}or{ D l,c , rank , Q l,c The total number of}). This total number can be less than or equal to 8, for example, 4, 5, 6, 8.

[0154] As an example, a target CSI based on a precoding matrix can have a rank-common and layer-common model to support scalable payload design. For the rank-common and layer-common model, a portion of the bits is shared across all layers (e.g., selecting the rank, selecting a port group / main beam cluster); another portion of the bits is used separately for each layer (e.g., refining port weights within that layer, phase / amplitude quantization). A scalable rank-common and layer-common model can support layer / rank-specific payload parameters, for example, adding more refinement bits (larger) when the feedback budget allows. or When budget is tight, only the core common components should be retained. In some scenarios, the single-layer payload sizes of the middle and / or high layers can be comparable, such as the total payload sizes of rank 2, rank 3, and rank 4 being comparable. For this model, spectral efficiency or system throughput, along with the trade-off with feedback overhead, can serve as performance metrics.

[0155] In the above embodiment, the target CSI is a precoding matrix. To balance performance, overhead, and complexity, it is necessary to configure the payload parameter combination for each layer, i.e., { D l,c , Q l,c}or{ D l,c , rank , Q l,c}

[0156] As an example, { D l,c ,Q l,c}or{ D l,c , rank , Q l,c The parameters in} can include, but are not limited to, the following candidate values: D l,c For example, the candidate values ​​are: 32, 64, 96, 128, 192; Q l,c The candidate values ​​are, for example: 1, 2, 3, 4 (8 and 10 are possible for VQ only); rank Candidate values ​​are, for example: 2, 4, D l,c .

[0157] As one example, multiple payload parameter combinations can form a PC pool for payload parameter combinations for each layer and each corresponding column of each layer. In some scenarios, network devices can also be configured with payload parameter combinations across layers and corresponding columns, i.e., payload parameter combinations composed of different layers and / or different corresponding columns. Therefore, the payload parameter combinations associated with CSI reports can support multiple configurable parameter combinations (across layers and corresponding columns).

[0158] Table 1 shows an example of different parameter combinations (PC1, PC2, ...) with rank 1 to rank 4. Here, Xn, Xn-31, Xn-32, Xn-33, Xn-41, Xn-42, Xn-43, and Xn-44 represent a pair of parameters selected from each layer of the PC pool. D l,c , Q l,c}or{ D l,c , rank , Q l,c As shown in Table 1, Xn represents X1, X2, X3, ... It should be noted that the number of bits quantized for each Xn can be different or the same.

[0159] Table 1 As an example, layer-common and rank-common payloads can be supported for all ranks. Alternatively, layer-common and rank-common payloads can be supported only for rank 1 and rank 2. When selecting each layer's { D l,c , Q l,c}or{ Dl,c , rank , Q l,c After that, the average SGCS of each layer can be regarded as a performance metric. For a specific layer pattern (e.g., {Xn-31, Xn-32, Xn-33} and {Xn-41, Xn-42, Xn-43, Xn-44}), the average SGCS and / or spectral efficiency and / or throughput of that layer are set as performance metrics.

[0160] In some embodiments, the first bit sequence includes a first field carrying the first part of the CSI (i.e., CSI Part 1) and a second field carrying the second part of the CSI (i.e., CSI Part 2). When the first and second fields are included, the first bit sequence is used to carry a two-part CSI report. The first part of the CSI typically consists of a CSI-RS resource indicator (CRI), an RI, and a first CQI. The second part of the CSI typically consists of a layer indicator (LI) and a PMI. The PMI portion of the second part of the CSI can be replaced with compressed CSI. For example, in a CSI compression use case using a two-sided model, a reporting principle starting with the first and second parts of the CSI can be considered.

[0161] As mentioned above, the payload corresponding to CSI is mapped to UCI, which includes the mapping or discarding of different CSI contents. Therefore, the first part of CSI can also be called the first UCI part, and the second part of CSI can also be called the second UCI part.

[0162] The length of the first field can represent the size of the payload corresponding to the first part of the CSI, and the length of the second field can represent the size of the payload corresponding to the second part of the CSI. The first and second fields can satisfy one of the following: the lengths of the first and second fields are determined based on pre-configured parameters or related signaling; or the length of the first field is determined based on pre-configured parameters or related signaling, and the length of the second field is determined based on the first field. When the length is determined based on pre-configured parameters or related signaling, this length is usually a fixed value.

[0163] When the lengths of the first and second fields are determined according to pre-configured parameters or related signaling, both the lengths of the first and second fields are fixed values. Therefore, the first part of the CSI carried by the first field and the second part of the CSI carried by the second field are both fixed payloads.

[0164] When the length of the first field is determined based on pre-configured parameters or related signaling, and the length of the second field is determined based on the first field, the length of the first field is a fixed value, while the length of the second field is a dynamic value. Therefore, the first part of the CSI carried by the first field is a fixed payload, while the second part of the CSI carried by the second field is a dynamic payload. In other words, the first part of the CSI has a fixed size configured by the network, while the size of the second part of the CSI is dynamic. Specifically, the size of the second part of the CSI is determined by the information in the first part of the CSI.

[0165] In some embodiments, the length of the second field is determined based on the first field, including but not limited to the fact that the first part of the CSI carried by the first field can indicate the payload corresponding to the second part of the CSI. As an example, the first part of the CSI may also include payload indication information, such as compression ratio and payload indicator. In this case, the CSI report content will be an integrated CSI combining RI / CQI / PMI. The concept of integrated CSI is more likely to be explicit feedback than traditional implicit feedback. Integrated CSI can consist of multiple accumulated information blocks during decoding. For example, when different information blocks correspond to different compression ratio levels, the reconstruction quality may depend on the number of information blocks that the decoder model (second model) takes as input.

[0166] In the above embodiments, for integrated explicit CSI feedback, the CSI coding scheme can be described as follows: the first UCI part is a fixed payload, including the number of information blocks N (N is a positive integer) and the first information block; the second UCI part is a dynamic payload, including N-1 information blocks (i.e., the second, ..., Nth information blocks). For integrated CSI reporting, the multi-user decoder (CSI reconstruction part) model can be used as a method for network selection of MCS, while considering the inference of terminal devices, without requiring CQI determination / reporting by the terminal devices.

[0167] As an example, the first part of the CSI may carry the RI, the first codeword CQI, and a size indicator for the second part of the CSI. The size indicator for the second part of the CSI may be indicated by fields such as "compression ratio / payload indicator".

[0168] As an example, when the second part of the CSI corresponds to a dynamic payload, it can be used to carry quantized potential segments. If the first part of the CSI indicates RI>4, the second part of the CSI can also carry a second codeword CQI.

[0169] When the length of the second field changes dynamically, the total payload size can be determined jointly by the network device configuration and the information in the first part of the CSI. For example, the network device can specify the available payload levels / limits in its configuration. The terminal device can then dynamically determine the length of the second field based on these levels / limits and the size of the specific second part of the CSI indicated by the RI in the first part of the CSI.

[0170] The length of the first bit sequence is determined based on the single-layer payload of one or more given layers. The one or more given layers can be a single layer or multiple layers in a MIMO transmission. For example, the one or more given layers can be one or more data streams transmitting the first bit sequence. Alternatively, the one or more given layers can be one or more spatial layers used by the terminal device for uplink transmission.

[0171] It should be noted that when the length of the first field is a fixed value, "the length of the first bit sequence is determined based on the single-layer payload of one or more given layers" can be replaced with: the length of the second field is determined based on the single-layer payload of one or more given layers. When the length of the second field is determined based on the single-layer payload of one or more given layers, the terminal device can select the method for determining the single-layer payload size and calculate the payload size corresponding to the second part of the CSI, thereby obtaining the total payload size.

[0172] As an example, the number of one or more given layers is associated with the RI. The terminal device infers the RI and reports it in the first part of the CSI.

[0173] The size of a single-layer payload is determined based on multiple levels related to the payload size. These levels can be the payload ratings / limits mentioned earlier; that is, multiple levels can be replaced by multiple ratings. Multiple levels related to the payload size can correspond to multiple payload selections, thus adapting to coverage differences.

[0174] As an example, multiple levels related to the payload size can correspond to multiple payload options, which can be used in the training / inference process of the first model and / or the second model. When there are a specified number of levels, the difficulty of model development and the workload of training can be effectively controlled. For example, there can be K levels, where K is a positive integer less than 4.

[0175] In some embodiments, the size of a single-layer payload satisfies at least one of the following: configured by the dimension of a given layer, the number of entries per segment, and the number of quantization bits per segment; not configured based on subbands, ports, or rank. As described above, the dimension of a given layer can be determined by... d z To represent the number of entries in each section, you can use... Lseg This can be represented by each quantized bit segment. Q This is represented as follows. The size of a single-layer payload is configured based on sub-bands, ports, or rank, which can be understood as decoupling the size of a single-layer payload from sub-bands / ports / rank. In other words, the determination of the size of a single-layer payload is insensitive (or minimally sensitive) to the number of sub-bands, the number of ports, and the rank assumption.

[0176] As an example, dimension d z Number of entries per paragraph L seg Each quantization bit Q It can form a ( d z , L seg , Q The triplet, namely the dimension-segment-bit triplet, is a general parameterization that can be used for segmented independent quantization. It is simple to implement and has stable performance. When inferring the payload size layer by layer using this triplet, the number of segments... The effective load of a single layer is p layer = S × Q .

[0177] As an example, the size of a single-layer payload should not scale with subband or port configurations or rank assumptions. Based on various payload configurations for different uplink coverages, a minimum and necessary number of payload configurations should be specified to obtain a finite number of single-layer payload sizes. For example, the determination of a single-layer payload is independent of subband or port configurations. Similarly, the determination of a single-layer payload is independent of rank assumptions.

[0178] In some embodiments, the multiple levels associated with the load size may include at least one of the following: the load level corresponding to one or more given layers; the coverage level of the terminal device; and the hierarchical overhead table associated with one or more given layers.

[0179] As an example, the size of a single-layer payload is determined based on the payload class corresponding to one or more given layers. The payload class can refer to the level of payload per layer. The system can set several discrete levels per layer, i.e., payload classes, to cover various uplink capabilities. For example, multiple payload classes can correspond to 64 / 128 / 256 / 512 bits per layer. In this case, the level of payload size for a single layer can be expressed as: p layer ∈{64,128,256,512}.

[0180] In the above embodiments, the size of a single-layer payload is independent of its rank; therefore, this approach can be described as "fixed layer level, rank independent." This approach can approximately decouple subband / port / rank, simplifies training / implementation, and reduces cross-factory collaboration costs. Based on the principle that the single-layer payload does not change with rank / subband / port, network devices can be configured with a single payload level for one or more given layers to share. For example, a network device can select one of multiple payload levels for a given layer and inform the terminal device. Alternatively, a network device can be configured with multiple payload levels, and the terminal device can select one or more payload levels for one or more given layers.

[0181] In the above embodiments, when the length of the first bit sequence (or the second field) is determined based on the single-layer payload of multiple given layers, the payload levels corresponding to the multiple given layers are the same, or at least two of the multiple given layers correspond to different payload levels.

[0182] For example, when multiple given layers correspond to the same payload level, the payload size of the multiple given layers is the same, and the total payload size can be determined by the product of the payload size of a single layer and RI.

[0183] For example, when multiple given layers correspond to different payload levels, the needs of different scenarios can be met. In some scenarios, the single-layer payload size corresponding to a lower layer can be high in bits, while the single-layer payload size corresponding to a higher layer can be low in bits.

[0184] As an example, when the size of a single-layer payload is determined based on the coverage level of the terminal device, coverage layering can be achieved. Based on coverage, the size of a single-layer payload can be divided into multiple tiers, i.e., multiple coverage levels. For example, multiple coverage levels can correspond to low-tier, medium-tier, and high-tier payload sizes, respectively. Alternatively, multiple coverage levels can correspond to light-tier and heavy-tier payload sizes, respectively. The light-tier payload size has fewer bits, saving bits. The heavy-tier payload size has more bits, achieving higher precision.

[0185] In the above embodiments, the coverage level is determined based on the rank of the transmitted first bit sequence; that is, the coverage level changes with the rank. Because the coverage level changes, the size of a single-layer payload scales accordingly. Therefore, this method of determining the total payload size can be called "coverage layering + rank scaling rule." Based on the change in coverage level, the size of a single-layer payload can be gently scaled with the rank to avoid linear explosion and resource waste. For example, the single-layer payload size with a rank of 2 can be approximately twice the size of the single-layer payload with a rank of 1; the single-layer payload size with a rank of 3 / 4 is close to the single-layer payload size with a rank of 2.

[0186] In the above embodiments, the training of the first model and the second model can be carried out according to the traditional principle of eType-II multi-parameter combination and rank expansion nonlinear growth, so as to reduce training complexity.

[0187] In the above embodiments, multiple coverage levels can be dynamically switched based on mobility or subband. For example, when setting the payload size for light and heavy coverage levels within a cycle, the critical subband / beamset can adopt the heavy coverage level, while the rest remain at the light coverage level.

[0188] As an example, the size of a single-layer payload can be determined based on a hierarchical cost table. This hierarchical cost table can also be called a per-layer cost table or a layer-by-layer cost table. Network devices can configure and distribute hierarchical cost tables to meet different rank configuration requirements. For example, the hierarchical cost table can be set based on a rank of 8, thus satisfying any rank from 1 to 8. Based on this method, network devices can allocate different precisions for different layers, achieving maximum flexibility in single-layer payload configuration while maintaining the shape and dynamic determination of the first / second part of the CSI; it also facilitates the development of differentiated strategies by network devices.

[0189] In the above embodiments, the total payload size can be obtained by summing the overhead of the first RI layers based on the hierarchical overhead table and RI. For example, when the terminal device reports RI=v in a certain report, the total payload size can be the sum of the first v layers in the hierarchical overhead table.

[0190] In the above embodiments, the network device can be based on the number of bits corresponding to each layer or ( d z , L seg , Q Triples are used to set the hierarchical cost table, and inter-layer weights can also be set. Therefore, the hierarchical cost table can be used to indicate one of the following: the number of bits per layer; the dimension, number of entries per segment, and number of quantization bits per segment for each layer; or inter-layer weights. For example, in the hierarchical cost table, each layer corresponds to a specific number of bits. Similarly, in the hierarchical cost table, each layer corresponds to a specific (…). d z , L seg , Q )combination.

[0191] As one implementation, the hierarchical cost table supports a rank of 8, and can be represented as: TableID={L1:256, L2:192, L3:96, L4:96, L5:64, L6:64, L7:48, L8:48}. In this implementation, the cost decreases as the number of levels increases.

[0192] As an implementation method, the costs of different layers in the hierarchical cost table can be the same or different, and no restriction is made here.

[0193] As an implementation approach, network devices can set inter-layer weights based on different scenarios. For example, network devices can set inter-layer weights based on signal-to-noise ratio (SNR) / reference signal-received power (RSRP), Doppler / speed, uplink budget, port / bandwidth, whether MU is paired, service priority, etc.

[0194] As an implementation, inter-layer weights can be non-linearly distributed. Lower-level layers have greater weights than higher-level layers. For multiple higher-level layers, the weights can be close. For example, with a rank of 4, the inter-layer weights of different layers can be as follows: Layer 1 > Layer 2 > Layer 3 ≈ Layer 4. The weights corresponding to the layers can be represented by w. l To represent this, the weights of layers 1 to 4 can be w1 > w2 > w3 = w4, respectively.

[0195] The preceding text introduced various methods for determining the single-layer payload size. Terminal devices can decide on the single-layer payload size for one or more given layers based on a first logical strategy. The first logical strategy can be related to one or more of the following: uplink (UL) resources, processing unit (PU) capabilities, application scenario, terminal device mobility characteristics, and rank variations. The first logical strategy can correspond to multiple selection schemes, such as conservative (P0), balanced (P1), and aggressive (P2) schemes.

[0196] As an example, when UL is limited or PU is tight, a conservative (P0) scheme can be chosen for the single-layer payload size. For example, only high bits are allocated to layer 1, medium bits to layer 2, and low bits to layers ≥3; subband clipping can also be performed if necessary.

[0197] As an example, for scenarios such as enhanced mobile broadband (eMBB), high SNR, and expected MU pairing, a relatively aggressive (P2) scheme can be chosen for the single-layer payload size. For example, the first two layers could have high bits, the third and fourth layers could have medium bits, and the rest could have low bits; it is also possible to allow subband A to be overcoded.

[0198] As an example, for high-speed / high-Doppler scenarios, the single-layer payload size can be selected as a relatively balanced (P1) or P0 scheme: to avoid extreme precision and prioritize stability and timeliness.

[0199] As an example, when the rank changes greatly, the "hierarchical decreasing allocation" (L1>L2>L3≈L4) is prioritized and the total number of bits corresponding to the maximum rank is limited.

[0200] As an example, for weak coverage / edge scenarios, we force the use of P0 and retain higher precision for Layer 1 / critical subbands, while downgrading the rest.

[0201] After selecting the single-layer payload size for multiple given layers based on the first logic, a verification window can be set. This verification window can be periodic or event-triggered. Within the verification window, a re-amplifier can be inserted once for calibration / monitoring.

[0202] The length of the first bit sequence can be the sum of the single-layer payload sizes of multiple given layers. For example, the size of a single-layer payload is... p layer At that time, the total effective payload size is: .

[0203] When the single-layer effective load size is the same for multiple given layers P total =RI× p layer .

[0204] In some embodiments, the summation method for multiple given layers can be determined according to the configuration or can be determined automatically. As one embodiment, the summation method can include three methods: direct summation of multiple given layers, weighted summation of multiple given layers, and sub-weighted summation.

[0205] As an example, summing multiple given layers directly is equivalent to summing the first v layers directly, i.e. P total =Σ {l=1..v} Bits[L l (General). Among them, Bits[L l That is the first Spatial layer (layer) lThe number of bits allocated in this CSI compression feedback. Bits[L] l The load can be determined based on the hierarchical overhead table, in bits, and is used to calculate the total load of the second part of the CSI.

[0206] Assuming the reported rank is RI=v, it will involve layers. For each layer L l Network devices can specify an overhead. For example, a hierarchical overhead table can directly specify the number of bits for each layer: Bits[L l ]= For example, the hierarchical expense table can be based on ( d z , L seg , Q The cost is specified using triples. If triples are used, d z It can be different for different floors, or the same for all floors; L seg , Q This affects the quantization granularity and reconstruction quality. As mentioned earlier, the number of segments in each layer is... Bits[L l ]= S l × Q l .

[0207] As another embodiment, multiple given layers can be summed with weights. Taking the weights into account, Bits'[L] l ]=w l ×Bits[L l When there are 4 given layers, the weights can be, for example, w1=1.0, w2=0.9, w3=0.6, w4=0.6).

[0208] As another embodiment, the total payload size can be determined based on subband weighting. For example, critical subbands can be treated with a multiplier γ>1, and non-critical subbands can be treated with a multiplier 1, while ensuring that the sum does not exceed the UL budget.

[0209] The preceding text introduced various implementations where the length of the first bit sequence is determined based on a single-layer payload, and the size of the single-layer payload is related to multiple levels. These multiple levels can be configured by the network device for the terminal device.

[0210] In some embodiments, indication information at multiple levels is carried in at least one of the following: RRC signaling, DCI, and high-level short messages.

[0211] As an example, RRC signaling can be used to implement semi-static configuration. RRC signaling can indicate the identity (ID) of a hierarchical cost table, a set of logical policies, and multiple profiles. For example, RRC signaling can include a cost table template (up to 8 layers). Each layer in the Table ID can correspond to a specified number of bits or triples. As another example, RRC signaling can include a set of multiple logical policy profiles for the terminal device to determine a first logical policy. The policy profile set could be, for example, P0 (conservative) / P1 (equilibrium) / P2 (aggressive), and each profile can provide a mapping rule corresponding to a given layer position / scene label.

[0212] As an example, DCI or higher-level short messages can be used to implement dynamic configuration. DCI or higher-level short messages can indicate at least one of the following: CSI report occasion; configuration file or hierarchical overhead table ID associated with the CSI report occasion; payload level and / or coverage level; whether the length of the second field corresponding to the second part of the CSI is determined by the first field.

[0213] For example, the DCI can indicate the timing of each CSI report or every few CSI report times. For each CSI report time, the terminal device can select a profile using an index or directly select a table ID and a level indicator. Therefore, multiple profiles in the RRC signaling can be used by the terminal device to determine the profile corresponding to the first bit sequence based on the DCI. The level indicator can be different payload levels or levels corresponding to coverage levels (e.g., "low / medium / high"). Specifically, the RRC can provide several profiles, and the terminal device selects the current profile based on the DCI or configuration. After selecting a profile, the terminal device sums the "per-layer overhead" according to the profile to obtain the size of the payload corresponding to the second part of the CSI or the total payload.

[0214] For example, DCI can instruct that the length of the second field corresponding to the second part of CSI is determined by the first field. In other words, the first part of CSI still includes the RI reported by the terminal device and the payload size corresponding to the second part of CSI. The configuration file selected and configured by the network device can determine how the terminal device "sums up" to obtain the target payload size of the second part of CSI, such as direct summation or weighted summation.

[0215] In some embodiments, the length of the first bit sequence can be extended based on the reported rank. That is, the size of the CSI feedback can be extended with the reported rank. In conventional schemes, the feedback size for rank 2 is approximately twice that for rank 1, while the feedback sizes for rank 3 and rank 4 are similar to those for rank 2. This design is based on the premise that rank 2 is the most likely rank in most cases. If the feedback size increases linearly with rank, then the network needs to allocate resources for the maximum possible rank. When the actual reported rank is lower than the maximum rank, this will result in a waste of uplink resources.

[0216] For model-based bilateral CSI compression, the above principle can be followed: the PMI feedback cost roughly doubles as it increases from rank 1 to rank 2, while the feedback costs for rank 3 and rank 4 remain at the same level as rank 2. According to this rule, the parameter combination used to determine the AI-CSI feedback cost can include at least three parameters: the model output dimension (…). z dim ), number of bits per dimension ( N bit ), and the payload size when the rank is 1 (denoted as X1). Assuming that the model structure is standardized using a layer-by-layer inference approach, using... z dim The output of the dimension inference represents the inference result for a specific MIMO layer. When the rank > 1, the terminal device can determine the payload size of each layer based on the configured payload size of rank 1 (X1) and the reported rank value. For example, when the rank is greater than 1, the terminal device can determine the payload size of each layer. When the effective load of the layer is Yl and the reported rank is v, Yl = f(X1, rank = v).

[0217] In step S1230, the terminal device sends the first bit sequence to the network device.

[0218] Network devices can reconstruct the CSI using a first bit sequence. The pairing ID corresponding to the first model can be used by the network device to determine a second model, which is used for CSI reconstruction using the first bit sequence. In other words, network devices can reconstruct the CSI based on both the first bit sequence and the second model. As an example, the second model can decode the first bit sequence to reconstruct the target CSI used for scheduling and / or precoding.

[0219] In some embodiments, the second model can be an AI / ML-based decoder model, i.e., the first AI / ML decoder model. The first model can also be referred to as an AI / ML-based CSI reconstruction sub-model.

[0220] In some embodiments, the pairing ID corresponding to the first model is shared by the network device and the terminal device. The network device can select a second model to be trained in pairs with the first model based on the pairing ID.

[0221] As one example, the network device can send the pairing IDs of the first model and the second model to the terminal device through the configuration information of the pairing ID. In some examples, the terminal device can inform the network device in advance that the deployed model is the first model. In other examples, the terminal device can inform the network device that the first bit sequence is determined based on the first model when uploading the first bit sequence.

[0222] As an example, the use of pairing IDs needs to consider the compatibility of the datasets associated with the pairing ID. For instance, when the network device's CSI reconstruction model is backward compatible with the exchanged dataset, the same pairing ID can be used as the exchanged dataset. In some scenarios, when the network device uses the same pairing ID as the exchanged dataset, a value label or sub-dataset ID can also be appended. Alternatively, the network device can use different pairing IDs and append signals indicating backward compatibility with the exchanged dataset, such as listing a list of pairing IDs for all compatible datasets. Conversely, when the network device's CSI reconstruction model is not backward compatible with the exchanged dataset, different pairing IDs can be used. Pairing IDs in the exchanged dataset possess Public Land Mobile Network (PLMN) uniqueness / global uniqueness.

[0223] As an example, the new dataset can be used for dataset updates and / or other configuration-related updates. For instance, when additional data samples are added to the exchanged dataset, the new samples in the new dataset can be used to update the exchanged dataset. The newly updated network device-side CSI reconstruction model can work with the end devices. The network device side can share additional datasets with the end device side for potential model updates. As another example, the new dataset can be used to update the exchanged dataset with other configurations (e.g., greater bandwidth, more ports, additional payload configurations, etc.). As yet another example, the new dataset can be used to train a new network device-side CSI generation model that is incompatible with the field end devices. As yet another example, a dataset can be split into multiple dataset batches before the dataset exchange. Labeling is required when adding additional data samples to the exchanged dataset or when a dataset is split into multiple subsets before the dataset exchange.

[0224] In the above embodiments, the exchanged dataset refers to a versioned data packet agreed upon and shared / distributed by the network device and the terminal device for training or calibrating paired AI / ML models. This data packet contains a sample set and necessary metadata to define the effective statistical domain of the quantized latent vector z and ensure interoperability between the encoder and decoder during the inference phase. Therefore, under the same pairing ID, X target CSI sets are typically maintained, each target CSI set bound to a port configuration. Different bandwidths / total number of subbands can be further distinguished under the same port configuration.

[0225] For example, the relationship between a pair ID and a related dataset can be represented as: a pair ID corresponds to { S 1,…, S X}, S x For the first x A target CSI set, and S x Binding (N1) x ×N2 x ×2, bandwidth / total number of subbands); where each S x It contains N4 target CSI samples.

[0226] Based on the above formula, the dataset associated with the pairing ID can be described as follows: the pairing ID has X target CSI sets (X is a positive integer), each target CSI set is associated with N1×N2×2 port configurations, and each target CSI set has N4 target CSI samples.

[0227] When a terminal device obtains the first bit sequence based on an encoder model, this encoder model can encode according to given rules. A set of encodeable rules can start with minimal information, such as input features (per terminal device, per cycle). This information can be one or more of the following: SNR_wb (or CRS / CSI-RS estimate), Doppler / velocity class features, UL_budget, MU_ready, QoS_class, RankPred, Port / BW size, and PU_cap. UL_budget, or UL budget, can be used to indicate the upper limit of the UCI bits for CSI; MU_ready can be used to indicate whether there is a potential co-frequency pairing; QoS_class, or quality of service (QoS) level, can correspond to ultra-reliable and low-latency communication (URLLC) / eMBB; RankPred is used by the terminal device to predict the rank to be used or to statistically analyze historical RI; and PU_cap is the PU / inference latency margin declared by the terminal device.

[0228] The above text combined Figure 12 and Figure 13 This paper introduces several implementations where the length of the first bit sequence is determined based on the single-layer payload size associated with multiple levels. As mentioned earlier, these multiple levels are independent of subband / port / rank to simplify model training and reduce cross-factory collaboration costs. To accommodate more scenarios, the length of the first bit sequence can also be extended based on these multiple levels.

[0229] In some embodiments, payload determination and the payload itself are scalable in terms of subband number, ports, and rank. In AI / ML-based CSI feedback payload determination, to reduce the increased workload of model training and development caused by higher granularity, only the three payload configurations described above (low, medium, and high) can be specified to provide a limited number of payload options for cell center / edge differences.

[0230] As an example, the configuration of three single-layer payload sizes can be divided based on port, subband, and rank, as follows: Low-end configuration (64 / 128 bits / layer): Suitable for devices with ≤32 ports, limited bandwidth / subband, scarce UL resources, or edge terminal equipment; Mid-range configuration (256 bits / layer): Suitable for >32 ports or larger bandwidth; High-end configuration (512 bits / layer): Suitable for central terminal equipment under large-scale port / broadband conditions or when fine MU-MIMO pairing is required.

[0231] It should be noted that the setting mechanism (low / medium / high) and scenario-driven switching are intended to switch between different load curves for scenarios such as center / edge, high-speed / stationary, and multi-user / single-user. If there are specific performance / implementation requirements, specific gear / port / subband combinations can be extended during the vendor collaboration phase, and the support sets of both parties can be aligned through an applicability inquiry process.

[0232] In some embodiments, when configured based on multiple levels, the expansion of the payload size may be associated with at least one of the following: the length of the first bit sequence is aligned with the first uplink resource; the single-layer payload is divided into base blocks and enhancement blocks; and the first vector is segmented adaptively quantized based on ports or subbands.

[0233] As one example, aligning the length of the first bit sequence with the first uplink resource can refer to determining the payload based on the uplink resource, i.e., using a complementary approach to determine and extend the payload. For example, based on uplink grant (UL-grant)-aware opportunistic payload selection, the CSI bits (first bit sequence) are aligned with the available uplink resource budget for the current instance to avoid over-reporting / under-reporting.

[0234] In the above embodiment, the network device configures an available set of bit sizes. When the terminal device performs this CSI report, it can obtain the target size of the second part by adding the layers together using the "second part size indication" carried in the first part of the CSI and the RI. When the layers correspond to unequal bits, layer weights (w1≥w2≥w3…) are configured on the layer overhead table to obtain the Bits[L] of each layer. l Then sum them up to get the second part of the size.

[0235] As one implementation, a single-layer payload is divided into basic blocks and enhancement blocks, which are transmitted according to resource availability. For example, when transmitting the first bit sequence, the bit segment corresponding to the basic block is transmitted first to ensure availability; then, the bit segment corresponding to the enhancement block is added according to budget / needs. The transmission of enhancement blocks can be truncated in advance. Alternatively, the bit sequence of each layer payload can be divided into N ordered blocks. The first part of the CSI provides "the number of blocks N and the order of the carried blocks," and the second part of the CSI carries the data sequentially. When congested, network devices only need the basic blocks, or place them sequentially. When idle, basic blocks can be added to enhancement blocks. This method involves accumulating triples into multiple segments.

[0236] As an example, when the first vector performs segmented adaptive quantization based on ports or subbands, bits can be concentrated on more important segments within the same total bit count for a single layer. For example, while keeping the total bit count constant for each layer, different segments can be quantized using different methods. L seg , Q Important paragraphs should be enlarged. L segor smaller Q The opposite applies to ordinary segments. This method avoids the linear scaling of single-layer overhead with port / subband as in traditional eT2, while preventing performance crashes caused by excessively small tiers paired with ultra-large scale.

[0237] In the above embodiments, the segment index / scheme may be implicitly represented by the pairing ID or indicated in a simplified manner in the first part of the CSI.

[0238] The method described in the above embodiments can achieve port / subband "applicability constraint" rather than nonlinear expansion, and provide an applicability matrix (e.g., ≤32 ports can use 64 / 128 bits / layer; >32 ports are recommended to ≥256 bits / layer).

[0239] In some embodiments, segmented adaptive quantization based on ports or subbands can refer to the first vector being grouped based on port groups and / or subband groups. As one implementation, port groups and / or subband groups can be port / group encodings under a pair ID. The pair ID can be used to define a pre-encoded mapping, the consistency of which is guaranteed by configuration. In large-scale port / multi-subband scenarios, a "more relevant group" can be used as a single encoding unit, reducing... d z Without damaging the structure. Under the description of the paired ID, the terminal device can first perform port / subband group mapping, and then unify the grouped vectors ( d z , L seg , Q This reduces training / loading complexity.

[0240] As one example, the indication information for port grouping and / or subband grouping is carried in the configuration information of the pairing ID. For example, RRC can issue port grouping parameters under the configuration to which the pairing ID belongs to implement port grouping. As another example, RRC can issue subband grouping parameters in the pairing ID to implement subband grouping.

[0241] As one example, based on port grouping parameters, the terminal device can map 128 ports to Gp groups, for example, groups of 8 or 16 ports each. During encoding, the terminal device can perform subspace compression on each group, mapping ports within the group to a fixed dimension. d p After concatenating all groups, the port vector after grouping is obtained. d z , L seg , Q The vector of ports × subbands after grouping can be used as the input dimension. dz And use segmented independent quantization.

[0242] As one example, based on subband grouping parameters, the terminal device can map up to 19 subbands to Gs groups, for example, grouping adjacent or related subbands into one group. During encoding, the terminal device first aggregates each subband group to a fixed dimension. d s Replace the subband-by-subband description; and concatenate it with the results of port grouping.

[0243] In relevant specifications, such as the 3GPP standard, subband configuration is implemented using bitmaps and supports flexible configuration. The number of subbands can be up to 19, and subbands can be continuous or discontinuous. For port configuration, to ensure compatibility with future communication systems, the 32-port configuration can be extended to 128 ports. Vendors can design scalable model architectures based on this flexible subband / port configuration. From a performance perspective, a small payload configuration may not be a good choice for a full 100MHz bandwidth or 128 ports. For example, 64 / 128 bits per layer is only suitable for <32 ports, and 256 bits per layer is only suitable for >32 ports. Similarly, a large payload configuration may be excessive for small subband or port configurations. Therefore, specifying supported subband and port configurations for each payload option is more beneficial for performance and implementation.

[0244] Furthermore, during inter-vendor collaboration, the suitability of each payload configuration pair for subband / port configurations can be further determined. Specifically, for a given dataset or model training tagged with a specific pair ID, the network device can choose to develop its model and quantization for a subset of standardized payload, subband, or port configurations. The network device can then replace the reference encoder input / output pairs with the selected configurations. On the end device side, the end device can also choose to develop its encoder for an even subset of the configurations selected by the network device. During the suitability query phase, a complete list of CSI report configurations is provided to the end device, which can respond to the suitability of each configuration based on its actual model training / deployment.

[0245] The previous section introduced the implementation methods for payload expansion and model training based on subband and port configuration. Another issue facing CSI feedback is whether the configurations for inference reports and monitoring reports can be separated. In actual communication, inference measurement resources and monitoring resources can be the same, thus allowing for flexible configuration of dedicated resources for monitoring reports. If dedicated monitoring reports are allowed, one report might be discarded due to collisions. Therefore, when a monitoring report occurs, it's advisable to consider reporting both inference and monitoring results simultaneously. This approach may be beneficial for terminal devices in managing PUs and buffers. Thus, in CSI feedback, when a terminal device needs to send a monitoring report, this report is used to simultaneously report both inference and monitoring results.

[0246] The above text combined Figures 1 to 13 The method embodiments of this application are described in detail below. Figures 14 to 16 The present application provides a detailed description of the apparatus embodiments. It should be understood that the descriptions of the apparatus embodiments correspond to the descriptions of the method embodiments; therefore, any parts not described in detail can be found in the foregoing method embodiments.

[0247] Figure 14 This is a schematic block diagram of a device for wireless communication according to an embodiment of this application. The device 1400 can be any of the terminal devices described above. Figure 14 The apparatus 1400 shown includes a first processing unit 1410, a second processing unit 1420, and a transmission unit 1430.

[0248] The first processing unit 1410 can be used to obtain the target CSI.

[0249] The second processing unit 1420 can be used to map the target CSI into a first vector based on the first model, wherein the first vector corresponds to a first bit sequence.

[0250] The transmitting unit 1430 can be used to transmit a first bit sequence to a network device; wherein, the pairing identifier corresponding to the first model is used by the network device to determine a second model, the second model is used to perform CSI reconstruction through the first bit sequence, the length of the first bit sequence is determined according to the single-layer payload of one or more given layers, and the size of the single-layer payload is determined according to multiple levels related to the payload size.

[0251] Optionally, the first bit sequence includes a first field carrying a first part of the CSI and a second field carrying a second part of the CSI, wherein the first field and the second field satisfy one of the following: the lengths of the first field and the second field are determined according to pre-configured parameters or related signaling; the length of the first field is determined according to pre-configured parameters or related signaling, and the length of the second field is determined according to the first field.

[0252] Optionally, the plurality of levels includes at least one of the following: the payload level corresponding to the one or more given layers; the coverage level of the terminal device; and the hierarchical overhead table associated with the one or more given layers.

[0253] Optionally, the multiple given layers correspond to the same payload level, or at least two of the multiple given layers correspond to different payload levels.

[0254] Optionally, the coverage level is determined based on the rank of the first bit sequence being transmitted.

[0255] Optionally, the hierarchical overhead table is used to indicate one of the following: the number of bits corresponding to each layer; the dimension, the number of entries per segment, and the number of quantization bits per segment corresponding to each layer; and the inter-layer weights.

[0256] Optionally, the size of the single-layer payload satisfies at least one of the following: configured by the dimension of a given layer, the number of entries per segment, and the number of quantization bits per segment; not configured based on subband, port, or rank.

[0257] Optionally, the indication information at the multiple levels is carried in at least one of the following: RRC signaling, DCI, and high-level short messages.

[0258] Optionally, the length of the first bit sequence is extended based on the plurality of levels, and the extension is also related to at least one of the following: the length of the first bit sequence is aligned with a first uplink resource; a single-layer payload is divided into base blocks and enhancement blocks; and the first vector is segmented adaptively quantized based on ports or subbands.

[0259] Optionally, the first vector is grouped based on port groups and / or subband groups, and the indication information of the port groups and / or the subband groups is carried in the configuration information of the pairing identifier.

[0260] Optionally, the first model is an AI / ML-based encoder model, and the second model is an AI / ML-based decoder model.

[0261] Optionally, the first processing unit 1410 and the second processing unit 1420 in the device 1400 can be processor 1610, and the transmitting unit 1430 can be transceiver 1630. The device 1400 may also include memory 1620. See details below. Figure 16 .

[0262] Figure 15 This is a schematic block diagram of another device for wireless communication according to an embodiment of this application. The device 1500 can be any of the network devices described above. Figure 15The apparatus 1500 shown includes a receiving unit 1510, a first processing unit 1520, and a second processing unit 1530.

[0263] The receiving unit 1510 can be used to receive a first bit sequence from the terminal device, the first bit sequence corresponding to a first vector, the first vector being obtained by mapping the target channel state information (CSI) based on a first model.

[0264] The first processing unit 1520 can be used to determine the second model based on the pairing identifier of the first model.

[0265] The second processing unit 1530 can be used to perform CSI reconstruction based on the first bit sequence and the second model; wherein the length of the first bit sequence is determined according to the single-layer payload of one or more given layers, and the size of the single-layer payload is determined according to multiple levels related to the payload size.

[0266] Optionally, the first bit sequence includes a first field carrying a first part of the CSI and a second field carrying a second part of the CSI, wherein the first field and the second field satisfy one of the following: the lengths of the first field and the second field are determined according to pre-configured parameters or related signaling; the length of the first field is determined according to pre-configured parameters or related signaling, and the length of the second field is determined according to the first field.

[0267] Optionally, the plurality of levels includes at least one of the following: the payload level corresponding to the one or more given layers; the coverage level of the terminal device; and the hierarchical overhead table associated with the one or more given layers.

[0268] Optionally, the multiple given layers correspond to the same payload level, or at least two of the multiple given layers correspond to different payload levels.

[0269] Optionally, the coverage level is determined based on the rank of the first bit sequence being transmitted.

[0270] Optionally, the hierarchical overhead table is used to indicate one of the following: the number of bits corresponding to each layer; the dimension, the number of entries per segment, and the number of quantization bits per segment corresponding to each layer; and the inter-layer weights.

[0271] Optionally, the size of the single-layer payload satisfies at least one of the following: configured by the dimension of a given layer, the number of entries per segment, and the number of quantization bits per segment; not configured based on subband, port, or rank.

[0272] Optionally, the indication information at the multiple levels is carried in at least one of the following: RRC signaling, DCI, and high-level short messages.

[0273] Optionally, the length of the first bit sequence is extended based on the plurality of levels, and the extension is also related to at least one of the following: the length of the first bit sequence is aligned with a first uplink resource; a single-layer payload is divided into base blocks and enhancement blocks; and the first vector is segmented adaptively quantized based on ports or subbands.

[0274] Optionally, the first vector is grouped based on port groups and / or subband groups, and the indication information of the port groups and / or the subband groups is carried in the configuration information of the pairing identifier.

[0275] Optionally, the first model is an AI / ML-based encoder model, and the second model is an AI / ML-based decoder model.

[0276] Optionally, the receiving unit 1510 in device 1500 can be a transceiver 1630, and the first processing unit 1520 and the second processing unit 1530 can be processors 1610. Device 1500 may also include a memory 1620. See details below. Figure 16 .

[0277] Figure 16 The diagram shown is a structural schematic of a communication device according to an embodiment of this application. Figure 16 The dashed lines indicate that the unit or module is optional. The device 1600 can be used to implement the methods described in the above method embodiments. The device 1600 can be a chip, a terminal device, or a network device.

[0278] Apparatus 1600 may include one or more processors 1610. The processor 1610 may support apparatus 1600 in implementing the methods described in the preceding method embodiments. The processor 1610 may be a general-purpose processor or a special-purpose processor. For example, the processor may be a central processing unit (CPU). Alternatively, the processor may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0279] The apparatus 1600 may further include one or more memories 1620. The memories 1620 store a program that can be executed by the processor 1610, causing the processor 1610 to perform the methods described in the preceding method embodiments. The memories 1620 may be independent of the processor 1610 or integrated within the processor 1610.

[0280] The device 1600 may also include a transceiver 1630. The processor 1610 can communicate with other devices or chips via the transceiver 1630. For example, the processor 1610 can send and receive data with other devices or chips via the transceiver 1630.

[0281] This application also provides a computer-readable storage medium for storing a program. This computer-readable storage medium can be applied to a terminal device or network device provided in this application embodiment, and the program causes a computer to execute the methods performed by the terminal device or network device in the various embodiments of this application.

[0282] The computer-readable storage medium can be any available medium that a computer can read, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0283] This application also provides a computer program product. The computer program product includes a program. This computer program product can be applied to a terminal device or network device provided in the embodiments of this application, and the program causes a computer to execute the methods performed by the terminal device or network device in the various embodiments of this application.

[0284] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0285] This application also provides a computer program. This computer program can be applied to a terminal device or network device provided in this application, and the computer program causes the computer to execute the methods performed by the terminal or network device in various embodiments of this application.

[0286] In this application, the terms "system" and "network" are used interchangeably. Furthermore, the terminology used in this application is only for explaining specific embodiments of the application and is not intended to limit the application. The terms "first," "second," "third," and "fourth," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. In addition, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0287] In the embodiments of this application, the term "instruction" can be a direct instruction, an indirect instruction, or an indication of a relationship. For example, A instructing B can mean that A directly instructs B, such as B being able to obtain information through A; it can also mean that A indirectly instructs B, such as A instructing C, so B can obtain information through C; or it can mean that there is a relationship between A and B.

[0288] In the embodiments of this application, the term "correspondence" may indicate a direct or indirect correspondence between two things, or an association between two things, or a relationship such as instruction and being instructed, configuration and being configured.

[0289] In the embodiments of this application, "predefined" or "preconfigured" can be implemented by pre-storing corresponding codes, tables, or other means that can be used to indicate relevant information in the device (e.g., including terminal devices and network devices). This application does not limit the specific implementation method. For example, predefined can refer to what is defined in the protocol.

[0290] In the embodiments of this application, the term "protocol" may refer to standard protocols in the field of communications, such as LTE protocols, NR protocols, and related protocols applied in future communication systems. This application does not limit the scope of these protocols.

[0291] In the embodiments of this application, determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0292] In the embodiments of this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0293] In the embodiments of this application, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0294] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0295] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0296] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0297] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for wireless communication, characterized in that, include: The terminal device obtains the target channel state information (CSI). The terminal device maps the target channel state information (CSI) into a first vector based on a first model, and the first vector corresponds to a first bit sequence. The terminal device sends the first bit sequence to the network device; Wherein, the pairing identifier corresponding to the first model is used by the network device to determine the second model, the second model is used to perform CSI reconstruction through the first bit sequence, the length of the first bit sequence is determined according to the single-layer payload of one or more given layers, and the size of the single-layer payload is determined according to multiple levels related to the payload size.

2. The method according to claim 1, characterized in that, The first bit sequence includes a first field carrying a first part of the CSI and a second field carrying a second part of the CSI, wherein the first field and the second field satisfy one of the following: The lengths of the first field and the second field are determined based on pre-configured parameters or related signaling. The length of the first field is determined based on pre-configured parameters or related signaling, and the length of the second field is determined based on the first field.

3. The method according to claim 1 or 2, characterized in that, The plurality of levels includes at least one of the following: The payload level corresponding to the one or more given layers; The coverage level of the terminal device; A hierarchy cost table associated with the one or more given layers.

4. The method according to claim 3, characterized in that, The multiple given layers have the same payload level, or at least two of the multiple given layers have different payload levels.

5. The method according to claim 3, characterized in that, The coverage level is determined based on the rank of the first bit sequence being transmitted.

6. The method according to claim 3, characterized in that, The hierarchical overhead table is used to indicate one of the following: the number of bits per layer; the dimension, number of entries per segment, and number of quantization bits per segment per layer; and the inter-layer weights.

7. The method according to any one of claims 4-6, characterized in that, The magnitude of the single-layer effective load satisfies at least one of the following: Configured by specifying the dimension of the layer, the number of entries per segment, and the quantization bits per segment; Configuration is not based on subband, port, or rank.

8. The method according to claim 1 or 2, characterized in that, The indication information at the multiple levels is carried in at least one of the following: Radio Resource Control (RRC) signaling, Downlink Control Information (DCI), and Higher Layer Short Message Service (HLS).

9. The method according to claim 1 or 2, characterized in that, The length of the first bit sequence is extended based on the plurality of levels, and the extension is also related to at least one of the following: The length of the first bit sequence is aligned with the first uplink resource; The single-layer effective load is divided into a base block and a reinforcement block; The first vector is segmented adaptively quantized based on ports or subbands.

10. The method according to claim 9, characterized in that, The first vector is grouped based on port groups and / or subband groups, and the indication information of the port groups and / or the subband groups is carried in the configuration information of the pairing identifier.

11. The method according to claim 1 or 2, characterized in that, The first model is an encoder model based on artificial intelligence (AI) / machine learning (ML), and the second model is a decoder model based on AI / ML.

12. A method for wireless communication, characterized in that, include: The network device receives a first bit sequence from the terminal device, the first bit sequence corresponding to a first vector, the first vector being obtained through the mapping of target channel state information (CSI) based on a first model; The network device determines the second model based on the pairing identifier of the first model; The network device performs CSI reconstruction based on the first bit sequence and the second model; The length of the first bit sequence is determined based on the single-layer payload of one or more given layers, and the size of the single-layer payload is determined based on multiple levels related to the payload size.

13. The method according to claim 12, characterized in that, The first bit sequence includes a first field carrying a first part of the CSI and a second field carrying a second part of the CSI, wherein the first field and the second field satisfy one of the following: The lengths of the first field and the second field are determined based on pre-configured parameters or related signaling. The length of the first field is determined based on pre-configured parameters or related signaling, and the length of the second field is determined based on the first field.

14. The method according to claim 12 or 13, characterized in that, The plurality of levels includes at least one of the following: The payload level corresponding to the one or more given layers; The coverage level of the terminal device; A hierarchy cost table associated with the one or more given layers.

15. The method according to claim 14, characterized in that, The multiple given layers have the same payload level, or at least two of the multiple given layers have different payload levels.

16. The method according to claim 14, characterized in that, The coverage level is determined based on the rank of the first bit sequence being transmitted.

17. The method according to claim 14, characterized in that, The hierarchical overhead table is used to indicate one of the following: the number of bits per layer; the dimension, number of entries per segment, and number of quantization bits per segment per layer; and the inter-layer weights.

18. The method according to any one of claims 15-17, characterized in that, The magnitude of the single-layer effective load satisfies at least one of the following: Configured by specifying the dimension of the layer, the number of entries per segment, and the quantization bits per segment; Configuration is not based on subband, port, or rank.

19. The method according to claim 12 or 13, characterized in that, The indication information at the multiple levels is carried in at least one of the following: Radio Resource Control (RRC) signaling, Downlink Control Information (DCI), and Higher Layer Short Message Service (HLS).

20. The method according to claim 12 or 13, characterized in that, The length of the first bit sequence is extended based on the plurality of levels, and the extension is also related to at least one of the following: The length of the first bit sequence is aligned with the first uplink resource; The single-layer effective load is divided into a base block and a reinforcement block; The first vector is segmented adaptively quantized based on ports or subbands.

21. The method according to claim 20, characterized in that, The first vector is grouped based on port groups and / or subband groups, and the indication information of the port groups and / or the subband groups is carried in the configuration information of the pairing identifier.

22. The method according to claim 12 or 13, characterized in that, The first model is an encoder model based on artificial intelligence (AI) / machine learning (ML), and the second model is a decoder model based on AI / ML.

23. A device for wireless communication, characterized in that, The device is a terminal device, which includes a transceiver, a memory, and a processor. The memory is used to store programs, and the processor is used to call the programs in the memory and control the transceiver to receive or send signals so that the terminal device performs the method as described in any one of claims 1-11.

24. An apparatus for wireless communication, characterized in that, The device is a network device, which includes a transceiver, a memory, and a processor. The memory is used to store programs, and the processor is used to call the programs in the memory and control the transceiver to receive or send signals so that the network device performs the method as described in any one of claims 12-22.

25. A communication device, characterized in that, Includes units or modules for performing the method as described in any one of claims 1-11 or 12-22.

26. A chip, characterized in that, Includes a processor for calling a program from memory, causing a device on which the chip is mounted to perform the method as described in any one of claims 1-11 or 12-22.

27. A computer-readable storage medium, characterized in that, It contains a program that causes a computer to perform the method as described in any one of claims 1-11 or 12-22.

28. A computer program product, characterized in that, Includes a program that causes a computer to perform the method as described in any one of claims 1-11 or 12-22.

Images