Method and apparatus for wireless communication

Abstract

Provided are a method and device for wireless communication. The method comprises: a terminal device determining a first structure according to first configuration information and / or a capability of the terminal device; and the terminal device performing CSI compression based on a first model to obtain a latent vector having the first structure; wherein the first configuration information is determined according to at least one of reported information of the terminal device, configuration information issued by a network device, and protocol pre-defined information.

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 channel state information (CSI) feedback, CSI compression enhanced by artificial intelligence (AI) / machine learning has higher feedback efficiency and resolution, stronger structural adaptability, and a more flexible objective function.

[0003] However, the latent vectors compressed by CSI are difficult to classify according to different parameters or the importance of codebook indexes, as is the case with codebook-based CSI feedback. When terminal devices lack sufficient resources to report CSI (e.g., preempted by low-latency services), it is difficult to implement a discard strategy based on the priority / importance naturally assigned by physical meaning, as is the case with traditional non-AI-based CSI reporting. Therefore, how to perform CSI feedback based on these latent vectors becomes a technical problem that needs to be considered. Summary of the Invention

[0004] This application provides a method and apparatus for wireless communication. The various aspects related to the embodiments of this application are described below.

[0005] In a first aspect, a method for wireless communication is provided, comprising: a terminal device determining a first structure based on first configuration information and / or the capabilities of the terminal device; the terminal device performing CSI compression based on a first model to obtain a latent vector having the first structure; wherein the first configuration information is determined based on at least one of information reported by the terminal device, configuration information issued by a network device, and protocol predefined information.

[0006] In a second aspect, a method for wireless communication is provided, comprising: a network device determining a first structure based on first configuration information and / or the capabilities of a terminal device; the network device determining a Communication Interface (CSI) based on a second model and the first structure; wherein the first structure is related to a first model of the terminal device, the first model being used by the terminal device to determine a latent vector of the CSI compressed with the first structure, and the first configuration information being determined based on at least one of reported information from the terminal device, configuration information issued by the network device, and protocol predefined information.

[0007] 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.

[0008] 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.

[0009] Fifthly, a communication apparatus is provided, comprising a unit or module for performing the method as described in the first or second aspect.

[0010] 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.

[0011] A seventh aspect provides a chip including a processor for calling a program from memory, causing a device having the chip mounted to perform the method as described in the first or second aspect.

[0012] 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.

[0013] 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.

[0014] 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.

[0015] In this embodiment, the terminal device can determine a first structure related to CSI compression based on first configuration information and / or the capabilities of the terminal device, and obtain a latent vector with the first structure through CSI compression based on the first model. The first configuration information is determined based on the information reported by the terminal device, the configuration information issued by the network device, or the protocol predefined information. Therefore, the terminal device can determine the first structure based on the protocol or negotiation with the network device. The negotiated or specified first structure can organize the latent vector into a divisible bitstream structure that can be sent or discarded according to priority / importance based on CSI compression characteristics, thereby improving the flexibility of CSI feedback based on latent vectors. Attached Figure Description

[0016] Figure 1 This is a system architecture example diagram of a wireless communication system to which the embodiments of this application are applicable.

[0017] Figure 2 This is a schematic diagram of the network architecture applicable to the embodiments of this application.

[0018] Figure 3A and Figure 3B This is a schematic diagram of the structure of the wireless protocol stack applicable to the embodiments of this application.

[0019] Figure 4 This is a schematic diagram of CSI feedback applicable to the embodiments of this application.

[0020] Figure 5 This is a schematic diagram of the architecture of the CSI compression model applicable to the embodiments of this application.

[0021] Figure 6 This is a flowchart illustrating a method for wireless communication proposed in an embodiment of this application.

[0022] Figure 7 This is a schematic diagram illustrating grouping based on the importance of feature blocks.

[0023] Figure 8 This is a schematic diagram of constructing multiple reporting groups based on feature block sorting and revenue threshold.

[0024] Figure 9 This is a flowchart illustrating another method for wireless communication proposed in an embodiment of this application.

[0025] Figure 10 This is a schematic diagram of CSI feedback based on the CSI compressed feature blocks and reporting groups.

[0026] Figure 11 This is a flowchart illustrating a possible implementation method for adding CSI feedback based on the reporting group.

[0027] Figure 12 This is a flowchart illustrating one implementation method of CSI feedback based on group configuration.

[0028] Figure 13 This is a flowchart illustrating one implementation method for triggering CSI feedback based on group configuration.

[0029] Figure 14 This is a schematic diagram of a device for wireless communication provided in an embodiment of this application.

[0030] Figure 15 This is a schematic diagram of another device for wireless communication provided in an embodiment of this application.

[0031] Figure 16This is a schematic diagram of the structure of a communication device provided in an embodiment of this application. Detailed Implementation

[0032] 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.

[0033] Communication system architecture Figure 1 This 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.

[0034] Figure 1 An example is shown of a network device and multiple terminal devices, such as... 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.

[0035] 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.

[0036] 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.

[0037] 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.

[0038] 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.

[0039] 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.

[0040] 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.

[0041] 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.

[0042] 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.

[0043] 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).

[0044] Figure 2 A schematic diagram of a network architecture 200 according to an embodiment of this application is illustrated. This network architecture 200 illustrates 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.

[0045] 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.

[0046] 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.

[0047] 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.

[0048] 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.

[0049] 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.

[0050] 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.

[0051] 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).

[0052] 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.

[0053] CSI Feedback In communication systems (e.g., 5G NR systems), Channel Identity Matrix (CSI) is a crucial basis for network equipment to perform downlink multiple-input multiple-output (MIMO) precoding, beam management, link adaptation, and resource scheduling. As an example, the terminal equipment (e.g., UE) measures the reference signal and estimates the downlink channel matrix H or equivalent channel representations (e.g., precoding vectors, spatial correlation features, etc.), then feeds back the CSI-related indicators to the network equipment (e.g., gNB). Figure 4 As shown. See also Figure 4 The UE can send CSI feedback to the gNB, and the gNB performs downlink precoding based on the CSI feedback.

[0054] As an example, in the codebook-based CSI feedback framework of 5G NR, the terminal device needs to report several fields in a single CSI report, such as rank indicator (RI), channel quality indicator (CQI), precoding matrix indicator (PMI), and layer indicator (LI). These fields together determine the downlink precoding and link adaptation strategies of the network device.

[0055] CSI feedback quality directly impacts spectral efficiency, block error rate (BLER), and system throughput. This is especially true in scenarios involving massive MIMO, high-frequency broadband, and significant multi-user interference, where CSI accuracy is more sensitive to the benefits of beamforming and multi-user scheduling. To improve CSI feedback quality, block segmentation mechanisms and packet dropping concepts have been introduced in NR and other communication systems. The following explanation uses 5G NR as an example.

[0056] For the chunked mechanism, NR introduces a chunked CSI reporting organization method. Specifically, the terminal device splits a single CSI report into two parts: Part 1 and Part 2. These two parts satisfy the following conditions: (1) The number of bits in Part 1 is fixed (or can be fixed under a given configuration), which makes it easier for network devices to complete decoding to determine the length; (2) The information carried in Part 1 is sufficient to determine whether Part 2 exists and its number of bits. That is, after successfully decoding Part 1, the network device can determine the decoding hypothesis of Part 2 and avoid performing a large number of blind tests on Part 2. (3) When uplink control resources are scarce or uplink control information (UCI) conflicts occur (e.g., preempted by low-latency services), the terminal device can send only Part 1 and discard Part 2, so that the system can still maintain basic operation based on the skeleton CSI and use more refined enhancement information as optional increments.

[0057] The aforementioned CSI Part 1 / Part 2 mechanism provides a crucial feasibility constraint at the standardization level: first, a fixed-length Part 1 eliminates bit-count uncertainty, and then Part 1 determines the existence and length of Part 2. Part 1 emphasizes required fields and minimum availability, while Part 2 emphasizes enhanced details and optional reporting. This segmentation mechanism enables the system to have a certain degree of incremental availability: receiving only Part 1 allows for the formation of executable basic decisions; receiving Part 2 further improves accuracy and performance.

[0058] Regarding the packet dropping concept, 5G NR allows terminal devices to discard (omit) portions of the CSI report under certain circumstances. As an example, when the CSI report is carried via the physical uplink shared channel (PUSCH) and uses a segmented reporting structure of CSI Part 1 / Part 2, the standard allows terminal devices to discard a portion of Part 2 content when uplink resources are limited or UCI multiplexing conflicts occur.

[0059] It's important to note that this discarding is not random; rather, it's done tiered based on priority reporting levels. For example, in the specific mechanism described in TS 38.214, when a CSI report consists of two parts on the PUSCH, the terminal device first discards Part 2, corresponding to the lowest priority level, and then progressively affects higher priority levels. This mechanism enables a controllable and predictable performance degradation path when resources decrease, and avoids the receiver falling into highly complex blind detection when the field set / bit length is uncertain.

[0060] Furthermore, the enhanced Type II codebook groups the numerous detailed parameters, primarily PMI, in Part 2 according to their field importance. Specifically, Part 2 is divided into three groups: Group 0, Group 1, and Group 2, and mapped to priority levels. Typically, Group 0 corresponds to the highest priority, followed by Group 1, and then Group 2. When it is necessary to reduce the load on Part 2, terminal devices generally discard Group 2 first, then Group 1, and finally Group 0. Based on this mechanism, even if Part 2 is partially discarded, network devices can still construct a usable precoding decision framework based on the remaining high-priority groups, and gradually supplement low-priority details to improve accuracy when resources permit.

[0061] As an example, the Part 2 PMI grouping and discarding mechanism of the enhanced Type 2 codebook essentially provides a standard built-in, field-level progressive enhancement framework: (1) when resources are sufficient, Group 0, Group 1, and Group 2 are reported to obtain the highest precision; (2) when resources are limited, Group 2 is reduced first according to priority level, then Group 1 is reduced, and in extreme cases, only Group 0 is retained or Part 2 is discarded entirely; (3) this process makes Part 2 structurally interpretable, bit count inferable, and receiver-end controllable while being discardable and progressive, and can serve as a reference for subsequent designs. Subsequent designs, for example, are designs that allow compressed features generated by AI / machine learning (ML) to be groupable, discardable according to priority, and incrementally supplemented.

[0062] AI / ML-based CSI Compression The performance of codebook-based CSI feedback is limited by codebook resolution and structural assumptions, especially with large-scale antenna arrays, wideband frequency selectivity, and high beam dimension, where the feedback volume increases dramatically. Therefore, 6G is gradually proposing a transition from reporting codebook indexes to reporting AI / ML-based compressed CSI representations, i.e., CSI compression. Specifically, downlink channel-related information is compressed and encoded at the terminal device side to obtain a low-dimensional or low-bit compressed feature (e.g., a latent feature). After the terminal device reports this compressed feature, the network device reconstructs the signal based on it or uses it directly for radio decision-making.

[0063] As an example, the objects of CSI compression are not unique, and commonly include: (1) explicit channel matrix H, such as complex matrices organized by subcarrier / subband; (2) equivalent precoding vector / precoding matrix representations, such as optimal or candidate precoding vectors; (3) linear algebraic derivatives, such as eigenvectors / principal eigenvectors obtained from channel covariance or Gram matrices, used to characterize the dominant subspace. There are also two main approaches to the recovery objectives on the network device side: (1) recover the explicit channel matrix or high-dimensional CSI from the compressed features, and then calculate precoding or perform scheduling according to the conventional link; (2) directly recover the key quantities required for precoding from the compressed features, such as equivalent PMI, principal subspace, or directly output the precoding vector, without having to explicitly recover the complete CSI.

[0064] AI / ML-enhanced CSI compression can employ end-to-end trainable encoder-decoder structures, such as autoencoder architectures. For ease of understanding, the following section will combine... Figure 5 The encoder-decoder structure and process are explained.

[0065] See Figure 5 The encoder is deployed on the UE side, and the decoder is deployed on the gNB side. The encoder's input is the explicit channel matrix / precoding vector / feature vector, and its output is the quantized compression feature z. The UE sends this feature z to the gNB as input to the decoder. The decoder's output is the key quantities required to reconstruct the input / precoding. Figure 5 The processing flow shown may include the following steps.

[0066] Step 1, the UE will input x (The input, which can be a channel matrix H, a precoding vector, an eigenvector, etc.), is fed into a neural network encoder E to map the high-dimensional input to low-dimensional continuous features. f The encoder is designed to retain as much of the most critical information for recovery / decision-making on the gNB side as possible, given a limited compression dimension. This process can be represented as: f =E( x ).

[0067] Step 2, for adaptation to air interface transmission, continuous features f Typically, it needs to be quantized into a finite bit representation, i.e., a quantized feature.

[0068] Step 3: Decoding is performed on the gNB side. After receiving and reconstructing the quantization features, the gNB sends them to the decoder network. The output can be a reconstructed channel matrix, an estimate of the precoding vector / matrix, or equivalent parameters directly related to downlink decisions. This process can be represented as: .

[0069] The above text combined Figure 5 This paper introduces the AI / ML-enhanced CSI compression process. Compared with traditional codebook-based CSI, AI / ML-enhanced CSI compression has several advantages. First, AI / ML-enhanced CSI compression has higher feedback efficiency and resolution. With the same or lower bit budget, the compressed features can carry richer channel structure information, making it particularly suitable for high-dimensional antennas and broadband scenarios. Second, AI / ML-enhanced CSI compression has stronger structural adaptability. The model can learn the implicit structure of specific deployment environments, antenna configurations, and propagation statistics, without being strictly limited by fixed codebook configurations. Third, AI / ML-enhanced CSI compression has a more flexible objective function, which can optimize both "reconstruction error" and "task performance," such as objectives more directly related to precoding gain or system throughput.

[0070] As can be seen from the above, existing AI / ML-based CSI compression schemes typically employ an encoder-decoder structure, where the terminal device maps high-dimensional channel information such as the channel matrix and precoding vector into quantized latent feature vectors (or latent vectors) and reports them.

[0071] However, compressed latent vectors are difficult to divide into Part 1 and Part 2 based on different parameters or the importance of codebook indexes, as is the case with traditional codebook-based CSI feedback. It's also difficult to naturally assign priorities / importance and discard strategies based on physical meaning. In other words, the compressed bitstream (latent vectors) obtained from AI / ML-based CSI compression is usually indivisible. When resources are insufficient (e.g., preempted by low-latency services), uncertain behavior of "either the entire sequence is transmitted or the entire sequence fails" can easily occur, increasing implementation complexity and interoperability verification difficulty. If the current 3GPP-discussed scheme adopts a layer-based discarding mechanism based on CSI reporting, some data may be directly discarded, easily leading to data integrity corruption.

[0072] Furthermore, when improved accuracy is required, existing solutions often lack a mechanism for incremental supplementary reporting of the same CSI context as needed, forcing repeated sending of the entire packet features or fallback to other feedback formats, resulting in insufficient utilization of feedback resources.

[0073] To address the aforementioned issues, this application proposes a method for wireless communication. A terminal device can determine a specific structure related to CSI compression, i.e., a first structure, based on first configuration information and / or the capabilities of the terminal device. After performing CSI compression based on a deployed first model, the terminal device can obtain a latent vector with the first structure. The first configuration information is determined based on information reported by the terminal device, configuration information issued by the network device, or protocol predefined information. Therefore, the terminal device can determine the first structure based on a protocol or negotiation with the network device, and obtain a separable latent vector based on the first structure. This organizes the compressed, uninterpretable CSI compression features into a groupable, priority / importance-based selective transmission or discardable, and incrementable bitstream structure, improving the flexibility of CSI feedback based on AI / ML.

[0074] It should be noted that this application proposes an AI / ML-enhanced CSI compression method. This method is applicable to release 20 (Rel20) and future protocols. In some examples, this method can be applied to separated source-channel coding (SSCC), joint source-channel coding (JSCC), or joint source-channel coding and modulation (JSCCM).

[0075] To facilitate understanding, the following will be combined with... Figure 6 The methods proposed in the embodiments of this application will be described in detail. Figure 6 The method shown is executed by a terminal device. The terminal device can be any of the communication terminals or terminal-side devices described above, such as a UE.

[0076] In some implementations, the terminal device may deploy a first model. This first model can be an AI / ML-based encoder. It should be noted that the first model can be deployed on the terminal device, on the terminal device side, on auxiliary devices communicating with the terminal device, or on other communication devices relative to the network device side.

[0077] As one implementation, the first model can be a model related to CSI compression. The first model is the first encoder used for CSI compression. The first encoder can be an AI / ML model. For example, the first encoder is an AI / ML-based CSI compression encoder.

[0078] As one implementation, the first model can partition the latent vectors after CSI compression based on semantics or other methods. For example, the first model can be a semantic encoder. Alternatively, the first model can be trained to determine the benefit or importance of different bitstream segments of the latent vectors. Furthermore, the training process of the first model can directly train which parts of the bitstream of the latent vectors are relatively important.

[0079] As an implementation approach, the first model is the terminal-side model (also known as the UE-side model) in the bilateral model related to CSI feedback.

[0080] As one implementation, the terminal device can train or update the first model after deploying the first model.

[0081] The terminal device can communicate with the network device. In one implementation, the terminal device can receive a reference signal sent by the network device to perform channel estimation. This reference signal is, for example, a CSI reference signal (CSI-RS). In another implementation, the terminal device can send the channel estimation result (CSI feedback) to the network device so that the network device can accurately understand the channel quality.

[0082] In some implementations, the cell where the terminal device is located is designated as the first cell. The network equipment in the first cell provides services to the terminal device.

[0083] The network device can be any of the network-side devices described above, such as a base station. As an example, the network device communicating with the terminal device could be the network device corresponding to the first cell. The network device can provide services to all terminal devices in the first cell.

[0084] In some implementations, the network device may deploy a second model. This second model can be an AI / ML-based decoder. The second model and the first model can form a bilateral model related to CSI feedback. The second model can work in conjunction with the first model for joint inference. For example, the first model is used for CSI compression, and the second model is used for CSI reconstruction or refactoring.

[0085] As one implementation, the second model can be a model related to CSI reconstruction. The second model is the first decoder used for CSI decompression. The first decoder can be an AI / ML model. For example, the first decoder is an AI / ML-based CSI decoder.

[0086] One implementation approach is for the network device to train a two-sided model and then send the trained model to the terminal device. For example, the network device can train a first model and then send the trained first model to the terminal device.

[0087] As one implementation, the first and second models can be a set of bilateral models related to CSI feedback and identified by a pairing identity (pairing ID).

[0088] 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.

[0089] See Figure 6 , Figure 6 The process shown includes steps S610 and S620, which are described below. It should be noted that... Figure 6 The method shown may also include other steps, which are not limited in the embodiments of this application.

[0090] In step S610, the terminal device determines the first structure based on the first configuration information and / or the capabilities of the terminal device. The terminal device determining the first structure may refer to the terminal device determining the type of the first structure and / or the parameter information of the first structure.

[0091] The first structure is a specific structure of the compressed CSI reported by the terminal device. The terminal device can perform CSI feedback based on the first structure to facilitate CSI reconstruction by the network device and determine the channel quality. The rationality of the first structure can directly affect the CSI performance after reconstruction by the network device. The first structure determined based on model training and / or negotiated configuration can effectively improve CSI performance.

[0092] In some implementations, the compressed CSI can be a latent vector obtained based on a first model. The first structure refers to the structure of the latent vector obtained by the terminal device through CSI compression based on the first model. Compared to the currently indivisible latent vector, the latent vector with the first structure is divisible. As mentioned earlier, an indivisible latent vector means that the entire vector needs to be transmitted; if its integrity is compromised, it will directly affect the performance of CSI feedback. In contrast, a divisible latent vector means that only a portion of the vector needs to be transmitted, and transmitting only a portion of the latent vector may reduce the accuracy of CSI feedback without affecting performance.

[0093] As an implementation approach, the first structure can achieve the segmentation of latent vectors based on CSI compression features, thereby aligning with the progressive approach of CSI Part 1 / Part 2 in 5G NR in terms of protocol behavior. That is, basic information is given priority in delivery, and enhanced information is added step by step when resources permit; when resources are insufficient or there are UCI conflicts, information can be discarded step by step according to rules, and incremental retransmission on demand is supported in the same CSI context to improve accuracy.

[0094] As one implementation, the first structure can be indicated based on multiple feature blocks compressed by CSI. Feature blocks can also be called semantic blocks or semantic feature blocks. The first structure can also be called the first semantic structure, representing a specific structure obtained based on semantics.

[0095] As one implementation, the latent vector with the first structure can be divided into multiple parts. These multiple parts can correspond to multiple CSI compression features. The first structure can include the method of dividing the latent vector. The terminal device can send part or all of the latent vector based on different division methods, which will be explained in detail later in conjunction with step S620.

[0096] The first structure may include the latent vector being divided into multiple reporting groups, for example, V reporting groups, where V is a positive integer. The parameter information of the first structure may include the number of multiple reporting groups, i.e., the number of groups.

[0097] Multiple reporting groups can be divided based on CSI-related benefits, meaning that multiple reporting groups can correspond to multiple different CSI-related benefits. CSI-related benefits can be expressed as performance metrics or their proxies directly related to precoding / scheduling. For example, CSI-related benefits can be parameters such as code matching degree, prediction spectral efficiency, effective throughput, and reduction in block error rate. Furthermore, CSI-related benefits can also be metrics related to CSI feedback, such as squared generalized cosine similarity (SGCS) and minimum mean square error (MMSE).

[0098] For example, the CSI-related benefits for each reporting group can also be referred to as group-level importance. This CSI-related benefit can be defined as the marginal benefit of the group's features as a whole to the task, that is, the additional incremental benefit that will be brought after the previous groups have already reached their goals.

[0099] For example, latent vectors can be divided into reporting groups such as basic information group and multiple enhancement groups based on CSI-related benefits.

[0100] For example, latent vectors can be divided into reporting groups such as core foundation group, first-level enhancement group, ..., detail supplement group based on CSI-related benefits.

[0101] As one implementation approach, when multiple reporting groups determine the benefits based on CSI-related benefits, these groups need to satisfy multiple benefit threshold constraints. Furthermore, when ranking the multiple reporting groups, the constraints of monotonically increasing benefits and diminishing marginal returns for each group also need to be satisfied.

[0102] As an example, the grouping configuration satisfies the following three constraints: meeting the threshold constraints for each group, monotonically increasing returns, and diminishing marginal returns for each group. The threshold constraints for each group can be determined based on multiple return thresholds to ensure different threshold requirements. For example, the base group must be able to complete basic decoding, i.e., reach the minimum threshold. The monotonically increasing returns can be understood as the network device enhancing details as it receives more groups, ensuring that the overall task returns do not decrease. The diminishing marginal returns of each group can be understood as ensuring that earlier groups bring greater incremental returns, while later groups only supplement details.

[0103] In the example above, when the latent vectors are divided into V reporting groups, the multiple reward thresholds can be expressed as follows: , … The relationship between the size of multiple revenue thresholds can satisfy... .vice versa.

[0104] The first structure may also include the latent vector being divided into multiple feature blocks, for example, K feature blocks, where K is a positive integer. The parameter information of the first structure may include the number of feature blocks, i.e., the number of features. In other words, the CSI compression output on the terminal device side is no longer a single indivisible bitstream, but is organized into K independently encapsulated feature blocks.

[0105] As one implementation, any one of the K feature blocks can be encapsulated independently. The K feature blocks can be represented as follows: f 1. f 2. f 3、…、 f K Based on independent encapsulation, the K feature blocks can be further mapped to multiple reporting groups with clearly defined importance. Based on this grouping mapping, V reporting groups can be obtained. .

[0106] In the above implementation, the K feature blocks can generate multiple reporting groups based on different feature types, or multiple reporting groups to be sent can be generated based on feature importance. For example, the K feature blocks are arranged in descending order according to feature importance, and any reporting group in the multiple reporting groups includes at least one feature block. The multiple reporting groups are generated based on multiple revenue thresholds and the K feature blocks arranged in descending order.

[0107] As an example, the feature importance of each feature block can be determined in offline evaluation during the model training process.

[0108] It should be noted that this application embodiment does not require feature blocks to be sorted by importance in the index order; it only requires that after mapping to multiple reporting groups, the indexes of each group be sorted by importance, and that there is an ordered partitioning. This ensures that the revenue threshold and marginal revenue properties hold true. However, in some engineering implementations, to reduce calibration complexity and enhance interpretability, a more practical grouping construction special case can be adopted: first, sort the feature blocks, then recursively select the block set for each group according to the threshold, satisfying the three constraints of grouping, and generating the model attributes. .

[0109] To facilitate understanding, the following will be combined with... Figure 7 and Figure 8 An example is provided to illustrate how to generate multiple reporting groups based on K feature blocks arranged in descending order. Figure 7 This is a schematic diagram showing K feature blocks sorted by importance. Figure 8 This diagram illustrates the recursive construction of groups based on multiple thresholds. Multiple reporting groups can be configured by grouping them according to the importance index order of feature blocks. The main steps are as follows.

[0110] First, for K feature blocks, for each feature block Calculate importance This yields a block index sequence sorted in descending order of importance. f 1. f 2、…、 f Kmax ,like Figure 7 As shown.

[0111] Secondly, multiple revenue thresholds can be satisfied. When constructing each reporting group recursively based on multiple revenue thresholds, first select the top-ranked blocks to form group 1. This makes it just meet the minimum threshold in a statistical sense. .

[0112] Next, continue adding subsequent blocks to Group 1 until the cumulative performance meets the threshold. The newly added block forms group 2 ( ). And so on, constructing group 3 level by level. ), ..., group V ( ),like Figure 8 As shown.

[0113] Finally, the result will be Fields such as V are embedded into the group configuration.

[0114] Based on the above grouping configuration method, the constraints of monotonically increasing task rewards and diminishing marginal returns can be satisfied, and the goal of group 1 meeting the minimum threshold required for the task and subsequent groups supplementing detailed information can be achieved. Furthermore, the embedding of this configuration in model attributes or grouping settings can be consistently implemented across terminal devices and network devices. When the grouping strategy needs to be updated, it can also be quickly recalibrated by updating the importance assessment and threshold configuration.

[0115] As an implementation, the K feature blocks can satisfy one of the following: the K feature blocks have the same quantization level; at least two feature blocks among the K feature blocks have different quantization levels; all feature blocks in the same reporting group have the same quantization level, while feature blocks in different reporting groups have different quantization levels; at least two feature blocks in the same reporting group have different quantization levels.

[0116] For example, in separate source channel coding or related implementations, feature blocks need to be quantized into bit representations. Since different feature blocks / groups have varying importance, using a uniform quantization level may result in important blocks being insufficiently refined or unimportant blocks wasting bits. Therefore, a multi-dimensional quantization level configuration strategy can be adopted for K feature blocks to improve the effective information density under the same resource budget. Several optional quantization strategies are as follows.

[0117] Strategy 1: Apply the same quantization level to each feature block, which is easy to implement.

[0118] Strategy 2: Assign different quantization levels to different feature blocks based on their importance to improve the accuracy of key blocks.

[0119] Strategy 3: Use the same quantization level for feature blocks within the same reporting group to reflect group-level priority.

[0120] Strategy 4: Within a group, differentiated quantification levels can still be configured for feature blocks within the group based on the group's importance, such as using group importance as a weight to adjust the importance of blocks within each group.

[0121] Strategy 5 employs a joint source-channel coding and modulation path, where the encoder directly outputs a complex symbol stream without bit quantization. It only needs to meet power constraints and maintain group-level mapping and appending mechanisms.

[0122] Based on multiple quantization strategies, the system can further refine the bit budget allocation without changing the structural configuration. For example, high-importance groups / blocks can obtain higher quantization accuracy, while low-importance groups / blocks can have reduced accuracy or be discarded, thereby maintaining the basic performance threshold when resources are limited and improving the accuracy of the enhancement layer when resources are abundant; at the same time, it provides compatibility options for different link forms (SSCC or JSCCM), improving deployment flexibility.

[0123] Multiple feature blocks can correspond to multiple feature importances; that is, the importance of each feature block can be determined based on its feature importance. Feature importance is associated with CSI-related benefits. For example, feature importance can be based on the wireless task benefit function. This involves measurement to provide a verifiable and practical semantic definition of the importance of feature blocks without being tied to the internal structure of specific AI / ML models, and based on this, to form a group-level progressive structure. Specifically, the profit function... It can reflect the benefits related to CSI, thus measuring the gains that network devices can achieve after acquiring some or all of the feature blocks. Benefit function Combining Figure 10 Please provide a detailed explanation.

[0124] The feature importance corresponding to a feature block can also be called block-level importance. It should be noted that feature importance can also be expressed as necessity, meaning that the absence of this block will cause the task reward to fail to reach the minimum threshold or significantly decrease. This definition does not depend on the interpretability of a specific network structure. In the actual quantization implementation based on feature blocks generated by the reporting group, vendors can calculate feature blocks through offline simulation / sampling evaluation. k Contribution , Or alternative indicators, and statistically analyze their robust ranking under different scenario conditions, thereby providing a basis for grouping construction.

[0125] The importance of a feature block or reporting group reflects its contribution to the CSI reconstruction of the original input. Therefore, the first structure can partition latent vectors based on the contribution of CSI compressed features to CSI reconstruction.

[0126] The first structure may also include latent vectors corresponding to multiple layers, or the latent vectors may be divided into multiple bitstreams. The parameter information of the first structure may include the number of layers corresponding to the latent vectors.

[0127] The first structure may include the latent vector being divided into multiple segments. The parameter information of the first structure may include the number of segments.

[0128] In some implementations, the partitioning method of the first structure indicator can be determined based on compressed features or feature blocks. For example, the latent vectors compressed by CSI can be grouped based on compressed features or feature blocks to generate multiple reporting groups with clearly defined importance.

[0129] When the terminal device determines the first structure based on the first configuration information, the first structure is determined through configuration. This configuration can be pre-configured, statically configured, semi-statically configured, or dynamically configured, and the embodiments of this application do not limit this.

[0130] As one implementation, when the first structure indicates that the latent vector is divided into multiple reporting groups, the first configuration information may include grouping configuration.

[0131] As one implementation, when the first structure indicates that the latent vector is divided into multiple feature blocks, the first configuration information may include block configuration.

[0132] As one implementation, when the first structure indicates that the latent vector corresponds to multiple layers, the first configuration information may include a hierarchical configuration.

[0133] The first configuration information may include parameter information of the first structure, and may also instruct the terminal device on how to perform CSI feedback. As one implementation, the first configuration information may include at least one of the following: the number of reporting groups, the number of feature blocks, or the number of layers corresponding to the first structure; the mapping relationship between multiple feature blocks and multiple reporting groups corresponding to the first structure; multiple benefit thresholds corresponding to the first structure; the discarding rules corresponding to the first structure; the encapsulation rules for at least one reporting group or at least one feature block sent by the terminal device; and the appending strategy corresponding to the first structure.

[0134] As one implementation, the number of reporting groups, feature blocks, or layers corresponding to the first structure can include the maximum number that the first model can output, or it can include the actual number sent. For example, the number of feature blocks corresponding to the first structure can include the maximum number of feature blocks. .

[0135] As one implementation, when latent vectors are divided into multiple feature blocks based on compressed features, these multiple feature blocks can be mapped to multiple reporting groups. The first structure can indicate the mapping relationship between the multiple feature blocks and the multiple reporting groups.

[0136] As an implementation approach, the multiple benefit thresholds corresponding to the first structure can be used to partition latent vectors. These multiple benefit thresholds can define different benefit levels, also known as multi-level task benefit thresholds. By defining multiple benefit thresholds, task benefit objectives can be categorized, such as basic availability, acceptable, good, and preferred. Based on multiple benefit thresholds, the question of whether a group can satisfy the task can be transformed from an empirical judgment into a clear constraint, facilitating performance control on the terminal device side when performing incremental reporting, conflict discarding, and incremental addition. Furthermore, multiple benefit thresholds can serve as progressive enhancement objectives, constraining the performance level to be achieved when resources allow and more groups are received. This allows the basic group to carry the main key semantics, while the enhancement group primarily supplements details and brings smaller but predictable performance improvements.

[0137] For example, multiple reporting groups determined based on multiple benefit thresholds can satisfy group-level asymptotic availability and predictability constraints, enabling the system to operate stably even when only a subset of groups are received, and monotonically increasing as more groups are received. For instance, multiple reporting groups can introduce a prefix-by-group approach. To achieve stable operation, the requirements for task performance on the terminal / network side need to be explicitly parameterized; for example, multiple benefit thresholds can be defined as... , … .

[0138] As one implementation approach, the discarding rule corresponding to the first structure can be determined based on CSI-related benefits or importance. The discarding rule can also be understood as a sending rule. In CSI feedback, this discarding rule needs to guarantee the transmission of basic information. For example, in resource-constrained or conflict-ridden scenarios, the terminal device adopts a sending rule that loads information from highest to lowest importance: at least the basic group is sent, guaranteed by the parameter `minGuaranteedGroup`; if resources allow and the number does not exceed the highest group number (`maxGroupID`), then other groups are added sequentially. The basic group can be Group 1, and other groups can be sorted based on importance, in the following order: Group 2, ..., Group... groupNum .

[0139] In the above implementation, when resources are insufficient or conflicts occur, the terminal device can discard lower-importance enhancement groups in a predefined order of importance, achieving predictable degradation. Since the network device can parse the reported information (e.g., control fields indicating reported content) and determine the packet mapping and actual transmission depth accordingly, uncertainty is reduced and high-complexity blind detection is avoided, making the behavior of packet transmission, resource-based discarding, and on-demand incremental additions deterministic, achievable, and standardizable.

[0140] As one implementation, after obtaining the latent vector with the first structure, the terminal device can send part or all of the latent vector. For example, the terminal device can send at least one reporting group or at least one feature block. The encapsulation rules of the reporting group or feature block sent by the terminal device can indicate how to encapsulate, the encapsulation positions of different fields, etc.

[0141] As one implementation, the appending strategy corresponding to the first structure can indicate whether incremental appending is supported and the range of appending is allowed. The terminal device can perform CSI feedback appending based on this appending strategy. When the first structure partitions latent vectors based on reporting groups, the appending strategy can indicate appending based on the index of the reporting group. When the first structure partitions latent vectors based on feature blocks, the appending strategy can indicate appending based on the index of the feature blocks. When the first structure indicates that a latent vector corresponds to multiple layers, the appending strategy can indicate appending based on the layer index.

[0142] Taking a grouped configuration as an example, each group profile can include at least one or more parameters of the first structure for predictable reporting and parsing in subsequent inference phases. When latent vectors are divided into multiple groups, the above parameters can include one of the following: Maximum number of feature blocks The first model's output feature block limit is the maximum number of feature blocks it can handle in each iteration, i.e., the maximum capability of the first model. The actual number of feature blocks sent by the terminal device can be less than [a certain number]. ; Number of groups V≥2: Map all feature blocks into multiple groups of different importance; Group mapping (NvMapping): This defines which feature blocks each group contains. This mapping can be constructed or indicated by feature block indices, ensuring mutual exclusion between groups and coverage of the entire set. A specific group (e.g., group 1) is the highest-importance base group, statistically satisfying the minimum task threshold. For example, achieving the minimum threshold for basic tasks with a high probability; Group Priority Order: For example, usually group 1 > group 2 > ... > group V, and vice versa; Encapsulation and appending rules: such as whether incremental appending is supported (contextMode) and the range of appendables within the same context (contextReuseRule).

[0143] The first configuration information is determined based on at least one of the following: information reported by the terminal device, configuration information issued by the network device, and protocol predefined information. When the first configuration information is determined based on information reported by the terminal device, the terminal device can determine the first structure itself or based on feedback from the network device regarding the reported information. When the first configuration information is determined based on configuration information issued by the network device, the terminal device can determine the first structure based on the network device's configuration. When the first configuration information is determined based on protocol predefined information, the terminal device can determine the first structure based on the protocol predefined information and / or the actual scenario.

[0144] In some implementations, the information reported by the terminal device may include the terminal device's capability information. This capability information can be used to indicate whether the terminal device supports latent vectors with a first structure. When the terminal device supports latent vectors with a first structure, it can indicate that the first model deployed by the terminal device can generate latent vectors with the first structure.

[0145] As one implementation, the terminal device's capability information can also indicate whether the terminal device supports generalization capabilities related to the first structure. For example, when a latent vector with the first structure is divided into multiple parts, this capability information can indicate whether the terminal device supports partial discarding. In other words, the terminal device's capabilities can include whether the latent vectors generated by the terminal device support partial discarding.

[0146] As one implementation, the capability information of the terminal device can indicate the capabilities of the terminal device. Supporting latent vectors with a specific structure can be an attribute of the terminal device or a capability of the terminal device. The terminal device needs to inform the network device of its capabilities by reporting information, so that the network device can issue configuration information to determine the first structure based on the reported information.

[0147] In some implementations, the information reported by the terminal device may include an application report sent by the terminal device. As one implementation, the terminal device can report parameters of a supported or determined first structure through this application report. Alternatively, the terminal device can report parameters of a specific supported structure through this application report and determine the first structure based on configuration information issued by the network device. The network device can determine the first structure based on the application report and configure it using the issued configuration information.

[0148] As one implementation, the terminal device can report the capabilities or parameters of the first model through the applicability report, and can also report the computing capabilities of the terminal device through the applicability report.

[0149] For example, in some cases, the number of features and / or the number of packets supported by the first model are determined by the terminal device's reporting when the terminal device and network device exchange applicability reports, or by the configuration issued by the network device to the terminal device.

[0150] For example, in some cases, the number of features and / or groups supported by the first model are model attributes. This attribute is determined by the terminal device's reporting when the terminal device and network device exchange suitability reports, or by the configuration issued by the network device to the terminal device.

[0151] In some implementations, the configuration information issued by the network device can be determined based on the information reported by the terminal device, or it can be determined based on the training results of the first model by the network device. The first model deployed by the terminal device can be trained by the network device. In this case, the network device understands the capabilities and characteristics of the first model and can instruct the first structure through the issued configuration information.

[0152] As one implementation, the configuration information issued by the network device may include CSI report configuration. The first configuration information can be determined based on the CSI report configuration and can be directly related to the CSI report configuration. For example, the CSI report configuration may include one or more structural configurations, and the first configuration information may indicate the structural configuration corresponding to the first structure.

[0153] In some implementations, protocol predefined information can refer to explicit specifications of a standard or protocol. This predefined information can indicate multiple architectures, and the terminal device can choose a first architecture from among these based on its own capabilities. As one implementation, the number of features and / or packets supported by the first model can be determined by a standard, for example, a positive integer number of packets.

[0154] The first configuration information may correspond to a first identifier, enabling the terminal device and network device to determine the first structure through the first identifier. For example, the first configuration information may indicate the first identifier, and the terminal device may determine the first structure based on the first identifier. Furthermore, the terminal device may perform CSI feedback and CSI feedback appending based on the first structure. As another example, when the terminal device performs CSI feedback, it may carry the first identifier, enabling the network device to determine the first structure possessed by the latent vector used for CSI feedback.

[0155] In some implementations, the first identifier can be at least one of the following: the identifier of the first configuration information (identity, ID), the identifier of the CSI report configuration corresponding to the first configuration information, the pairing identifier (pairingID) corresponding to the first model, or the identifier corresponding to the first structure. For example, when the first structure divides latent vectors into multiple reporting groups, the identifier corresponding to the first structure can be represented as groupProfileID.

[0156] As one implementation, when a CSI report configuration includes a series of structure configurations, each CSI report configuration ID can correspond to one structure configuration. The first configuration information can correspond to the CSI report configuration ID indicating the first structure. For example, a CSI report configuration can include a series of group configurations. Alternatively, each CSI report configuration ID can correspond to one group configuration.

[0157] As one implementation, the first structure can be selected / negotiated by the terminal device and the network device before or during the first model's activation, and indexed by a unique identifier (first identifier). When the first identifier is groupProfileID, subsequent reports from the sending end only need to carry this identifier to allow the receiving end to recover packet boundaries and resolve configuration rules. In some cases, when there is no groupProfileID, the first identifier can be the CSI report configuration ID corresponding to the first structure.

[0158] As one implementation, when the CSI feedback of the terminal device carries the first identifier, the first identifier can be carried in the first part corresponding to the CSI feedback (e.g., CSI Part 1), and part or all of the latent vector can be carried in the second part corresponding to the CSI feedback.

[0159] As an implementation approach, the pairing identifier corresponding to the first model can be one-to-one with the identifier of the first structure to simplify configuration and save air interface transmission overhead. For example, in CSI compression feature grouping, the pairing identifier corresponding to the first model can be uniquely mapped to the groupProfileID. In CSI compression feedback scenarios, the encoder (first model) is deployed on the terminal device side, and the decoder (second model) is deployed on the network device side. When the network device needs to manage the encoder-decoder model pairs (indicated by pairingID) and their corresponding grouping strategies, complex configurations may arise where the same model pair corresponds to multiple sets of grouping strategies. For example, when the same encoder-decoder model pair needs to handle different services / tasks, adopt different UCI conflict strategies, or allocate different model attributes based on resources (different... (e.g., V). To reduce implementation complexity and overhead of air interface control fields, a one-to-one correspondence can be established between model pairing identifiers and groupProfileID.

[0160] In the above implementation, it is stipulated that each encoder-decoder model pair corresponds to only one fixed grouping strategy, that is, the pairingID uniquely identifies the groupProfileID. Terminal devices or network devices can directly deduce NvMapping through the pairing identifier. V and related strategies. When reporting over the air interface, terminal devices can explicitly carry only the pairing identifier, without needing to carry the groupProfileID separately (or implicitly represent it in a less expensive way), thus simplifying header field and signaling configuration. For example, when there is no groupProfileID, the grouping configuration parameters can be attached to the pairing identifier, or included in the auxiliary parameters of model training, or included in the training dataset.

[0161] The above implementation method can significantly simplify network configuration management and air interface field carrying. Specifically, this implementation method can reduce the number of identification fields, reduce the parsing and interoperability burden, and is especially suitable for scenarios with stable model policies and relatively fixed business types; at the same time, it still retains the core capabilities of progressive group reporting and on-demand addition, resulting in lower complexity and more robust deployment.

[0162] When a terminal device determines the first structure based on its capabilities, the first structure may be related to the terminal device's computing power or determined based on training results. In some implementations, the terminal device's computing power influences the parameters of the first structure. For example, when the terminal device has strong computing power, the partitioning method indicated by the first structure can improve the performance and accuracy of CSI feedback. In other implementations, when the terminal device trains the first model, its capabilities are related to the training results of the first model, thereby determining the first structure.

[0163] The first structure can also be determined based on the training process of the first model and / or the second model. During the training and calibration phase of the first model and / or the second model, the terminal device or network device can statistically analyze the contribution / importance distribution of CSI compression features to CSI feedback, thereby pre-determining a structural configuration that is consistent and usable by both the terminal device and the network device.

[0164] For example, terminal or network devices can employ dual-end model training and extensive simulation / field evaluation to statistically analyze the contribution / importance distribution of CSI compression features on a slow time scale. This allows for the pre-determining of a set of packet configurations that are consistently usable by both the terminal and network devices, forming one or more packet configurations. Evaluation scenarios include typical deployment scenarios, typical service thresholds, and typical channel condition ranges.

[0165] In step S620, the terminal device performs CSI compression based on the first model to obtain a latent vector with a first structure.

[0166] The terminal device performs CSI compression based on the first model, which can be understood as the terminal device performing CSI compression through the first model. As one implementation, when the first model is an encoder, the terminal device can input an explicit CSI matrix or a precoded vector H into the encoder to obtain a latent vector for CSI feedback.

[0167] As one implementation, the latent vector output by the first model is the complete latent vector obtained from the input parameters.

[0168] In some implementations, regardless of whether the first model is trained on the terminal device side or the network device side, it can be trained based on auxiliary information. After the first and second models are paired, during training, the network device can send auxiliary information to the terminal device for model training. This auxiliary information can be used to instruct the training process of the first model. This auxiliary information can also indicate or determine the training result of the first model, that is, the first model can be trained to support the first structure.

[0169] As one implementation, the auxiliary information used to train the first model may include parameter information of the first structure. The first structure can be determined based on model training. For example, the auxiliary information may include the number of multiple reporting groups or the number of multiple feature blocks. Alternatively, the auxiliary information may include the group-level importance ranking of multiple reporting groups.

[0170] In some cases, this auxiliary information may describe the structure of the latent vectors, the generalization method of the first and / or second models, or the length of the latent vectors involved in the generalization. In other cases, this auxiliary information may describe the omission or punching method, or range, or other structural information corresponding to the structure of the latent vectors (e.g., the first structure). In still other cases, this auxiliary information may include whether the terminal device supports latent vectors with a specific structure.

[0171] As one implementation, the terminal device can receive auxiliary information sent by the network device. When the terminal device trains the first model, this auxiliary information can be used to train the first model. When the network device trains the first model, this auxiliary information can be used by the terminal device to determine the first structure supported by the first model. When the first model and the second model are paired models, the network device can train both models simultaneously and inform the terminal device of the auxiliary information.

[0172] For example, when the first model is an encoder and the second model is a decoder, the training processes of the first and second models and the configuration of the first structure can be coordinated. In one implementation, the training process consists of two stages: model training and group configuration construction, and can be iteratively optimized in a closed loop. In stage one, the encoder-decoder is trained until convergence. The encoder on the terminal device side takes CSI-related inputs, such as the channel matrix H, precoding vector, and feature vector / principal subspace representation, as inputs, and outputs a fixed upper limit number of feature blocks. Or equivalent latent representation; followed by quantization / encapsulation; output from the decoder on the network device side. Alternatively, the output can be equivalent to the task. Training loss can be reconstruction error, task loss, or a weighted combination of both. In phase two, importance assessment and group determination are performed based on the reward function. With model parameters frozen or using small-step updates, block-level marginal rewards / necessities are calculated offline to form a group-level importance ranking. Then, iterative optimization is performed using constraints such as thresholds and diminishing marginal returns to construct... To generate a groupProfile.

[0173] Based on the above training process, a standardized group-level importance progressive structure can be further introduced on the basis of AI / ML-enhanced CSI compression model training optimization, providing a foundation for subsequent protocol encapsulation and incremental mechanisms.

[0174] Latent vectors with the first structure can be represented using the parameter information of the first structure. For example, the number of reporting groups corresponding to the first structure can be represented as V. Similarly, the number of feature blocks corresponding to the first structure can be represented as K.

[0175] For example, the parameter information of the first structure may include at least one of the following: the number of reporting groups, the number of feature blocks, or the number of layers corresponding to the first structure; the mapping relationship between multiple feature blocks and multiple reporting groups corresponding to the first structure; multiple revenue thresholds corresponding to the first structure; the discard rule corresponding to the first structure; and the append strategy corresponding to the first structure.

[0176] In step S630 ( Figure 6 (Not shown in the diagram), after the terminal device obtains a latent vector with a first structure, it can send part or all of the latent vector to the network device. The terminal device can send part or all of the latent vector based on the first structure. The part or all of the latent vector sent by the terminal device is used by the network device to determine the CSI. When the terminal device sends the entire latent vector, the network device can parse the entire latent vector and obtain a high-precision CSI. When the terminal device sends a part of the latent vector, the network device can obtain basic CSI information or a CSI with a certain degree of precision based on that part of the latent vector.

[0177] The latent vector portion can refer to a portion of the bitstream determined based on the first structure of the latent vector. For example, when the first structure indicates that the latent vector corresponds to multiple reporting groups, the latent vector portion can be one or more of those reporting groups. Similarly, when the first structure indicates that the latent vector corresponds to multiple layers, the latent vector portion can be one or more of those layers.

[0178] In some implementations, the latent vector sent by the terminal device can be a portion specified by the network device. In some cases, this specified portion corresponds to the compressed content of certain layers of CSI; for example, if CSI has four layers, specifying the first and second layers. In other cases, the network device can specify the quantization method of the reported portion, the payload size of the reported portion, or select the type of payload through RRC signaling.

[0179] In some implementations, the latent vector portion sent by the terminal device can be determined based on the first structure and transmission resource conditions. As an example, when the system experiences uplink resource fluctuations, such as control plane congestion, low-latency service preemption, or UCI conflicts, the terminal device can progressively report and predictably discard data while ensuring that the network device can resolve it.

[0180] As one implementation, when the first structure indicates that a latent vector is divided into multiple reporting groups, a portion of the latent vector can be at least one of the multiple reporting groups. That is, the terminal device can send this at least one reporting group to the network device. This at least one reporting group can be used by the network device to determine the CSI. To improve the performance and accuracy of the CSI determined by the network device, this at least one reporting group can be determined based on multiple CSI-related benefits corresponding to the multiple reporting groups. For example, under limited resources, this at least one reporting group is one or more reporting groups with the highest CSI-related benefit among the multiple reporting groups.

[0181] As an example, the reporting group with the lowest CSI-related benefit among at least one reporting group is the first reporting group. The CSI-related benefit corresponding to the first reporting group is greater than the CSI-related benefit corresponding to any other reporting group among the multiple reporting groups besides the at least one reporting group.

[0182] As one implementation, when the first structure indicates that the latent vector is divided into K feature blocks, a portion of the latent vector can be at least one of the K feature blocks. That is, the terminal device can send this at least one feature block to the network device.

[0183] As another implementation, when the first structure indicates that the latent vector is divided into K feature blocks, the terminal device can generate multiple reporting groups to be sent based on the K feature blocks. The latent vector portion can be at least one of the multiple reporting groups, which will not be elaborated further.

[0184] In some implementations, once the terminal device determines part or all of the latent vector to be transmitted, quantization and / or payload encapsulation can be performed. The quantized / encapsulated code stream segment can adopt a separate source-channel coding path, i.e., quantization followed by entropy coding / channel coding; or it can adopt a joint source-channel coding and modulation path, where the encoder directly outputs a complex symbol stream, only needing to meet power constraints and maintain group-level mapping relationships.

[0185] To ensure the proper functioning of the model, the terminal device or network device can also monitor the first model and / or the second model. When the first model and the second model are paired bilateral models, the terminal device or network device can monitor both models simultaneously.

[0186] In some implementations, the terminal device monitors the first model and / or the second model based on a portion or all of the latent vectors output by the first model. The portion or all of the latent vectors output by the first model can refer to a portion or all of the latent vectors sent by the terminal device. In other words, the terminal device can perform model performance monitoring based on a portion or all of the latent vectors sent to the network device.

[0187] Based on the monitoring results, the terminal device can report the monitoring results of the first model and / or the second model. These monitoring results are used to indicate or determine whether the first model and / or the second model are functioning correctly. The monitoring data and results can be parameters that directly indicate whether the model is functioning correctly, or monitoring values ​​used by the network device to determine whether the model is functioning correctly.

[0188] As an example, a monitoring result of 0 indicates normal operation, while a result of 1 indicates an anomaly. The reverse is also true.

[0189] As an example, monitoring results can use SGCS or MMSE as metrics to determine whether the model is functioning correctly by network devices.

[0190] When a terminal device sends part or all of the latent vector to a network device, it also needs to send that part or all of the reporting information so that the network device can decode the CSI feedback. This reporting information can be carried in either the first part or the second part of the CSI feedback. Therefore, the reporting information can be encapsulated at different locations. Considering different standard evolution paths, implementation constraints, and different carrying methods, the multi-location encapsulation option for the reporting information can improve compatibility and deployment flexibility.

[0191] In some implementations, the reporting information sent by the terminal device may include a content indication control field. This content indication control field helps the network device acting as the receiving end to reliably determine the parsing assumptions of the current UCI payload, such as the actual group number sent, the number of blocks within the group, or the length information. For example, the content indication control field represents the groupNum of the highest group number actually sent.

[0192] As an implementation approach, placing content indication class control fields, such as groupNum, in CSI Part 1 can prioritize ensuring that network devices infer the length and structure of subsequent group payloads.

[0193] As another implementation, the content indication control field can be placed in CSI Part 2, suitable for configurations where Part 1 already contains sufficient information and Part 2 is used to enhance parsing. In a further implementation, the content indication can also be placed in the Group 1 payload, delivered simultaneously with the base group, thus ensuring parsing determinism without additional space occupied by Part 1 fields.

[0194] The aforementioned multi-location encapsulation provides optional landing paths for different protocol bearer methods. When the highest resolvability is required, CSI Part 1 bearer is preferred; when it's necessary to reduce header field expansion or align with existing frameworks, Part 2 or Group 1 bearer can be selected. This flexibility helps lower the barriers to standardization and engineering integration, while maintaining the core objectives of resolvability determinism and progressive reporting.

[0195] The above text combined Figures 6 to 8 This paper introduces a CSI feedback method where the terminal device determines a first structure related to a first model based on configuration or training, and generates or sends partial or complete CSI feedback with latent vectors containing the first structure. This method allows compressed output features to be constructed into payloads with clear grouping boundaries and hierarchical relationships without relying on the internal structural details of a specific model, aligning with the existing NR (New Network Response) principle of prioritizing basic information and allowing optional augmentation information.

[0196] Furthermore, the text above introduced the definition and constraint methods for feature importance, which improves interpretability and verifiability while avoiding binding to specific AI / ML model forms. In other words, it establishes an importance expression and constraint method that is independent of the black-box nature of the model, so that "why certain features are more important and should be reported first" has clear semantics and verifiable conditions.

[0197] For network devices, a first structure also needs to be determined in order to decode part or all of the received latent vectors. The following section combines... Figure 9 This section provides a detailed explanation of the methods used for wireless communication on the network device side. For brevity, Figure 6 Terms already explained or content already described will not be repeated.

[0198] See Figure 9 , Figure 9 The process shown includes steps S910 and S920, which are described below. It should be noted that... Figure 9 The method shown may also include other steps, which are not limited in the embodiments of this application.

[0199] In step S910, the network device determines a first structure based on the first configuration information and / or the capabilities of the terminal device. The first structure is related to a first model of the terminal device; it can be understood that latent vectors with the first structure are generated by the first model. In other words, the first model is used by the terminal device to determine the CSI-compressed latent vectors with the first structure.

[0200] When the first configuration information is determined based on the information reported by the terminal device, the network device can receive the reported information from the terminal device and determine the first structure based on the reported information. After determining the first structure, the network device can issue the first configuration information.

[0201] When the first configuration information is determined based on the configuration information issued by the network device, the first configuration information is also the issued configuration information.

[0202] When the first configuration information is determined according to the protocol predefined information, the network device determines the first structure in the same way as the terminal device.

[0203] When the first structure is determined based on the capabilities of the terminal device, the network device can receive the capability information reported by the terminal device and determine the first structure.

[0204] In step S920, the network device determines the CSI based on the second model and the first structure. As mentioned above, the second model is a decoder. In some scenarios, the second model is a network-side model paired with the first model.

[0205] The network device can receive part or all of the latent vectors sent by the terminal device, i.e., receive CSI feedback. The network device can receive part or all of the latent vectors based on a first structure. After receiving part or all of the latent vectors, they can be used as input to a second model for CSI reconstruction. Since part or all of the latent vectors are partitioned based on the first structure, the parameter information of the first structure can also be used as input information for the second model, so that the network device can decode more accurately.

[0206] For example, a network device receives at least one reporting group sent by a terminal device and determines a CSI based on the at least one reporting group.

[0207] For example, a network device receives at least one feature block sent by a terminal device and determines the CSI based on the at least one feature block.

[0208] As one implementation, when a terminal device sends reporting information that is partially or fully related to a latent vector, this reporting information can also be used by the network device to decode part or all of the received latent vector. This reporting information can also be used as input to the second model.

[0209] To ensure the proper functioning of the models, network devices can also monitor the first model and / or the second model. When the first and second models are paired bilateral models, the network device can monitor both models simultaneously.

[0210] In some implementations, the network device can monitor the first model and / or the second model based on some or all of the received latent vectors. In other words, the network device can monitor model performance based on some or all of the latent vectors sent by the terminal device.

[0211] In some other implementations, the network device can receive monitoring results of the first model and / or the second model sent by the terminal device.

[0212] The above text combined Figures 6 to 9 This paper introduces various implementation methods for terminal devices to provide CSI feedback based on latent vectors with a first structure, and for network devices to determine CSI based on this CSI feedback. For ease of understanding, the following example uses multiple reporting groups of feature block mappings, combined with... Figure 10 The determination of the first structure and the specific calculation method are illustrated by example. Figure 10 In this context, UE stands for Terminal Equipment, and gNB stands for Network Equipment. The encoder on the UE side is the first model, and the decoder on the gNB side is the second model.

[0213] See Figure 10 On the UE side, the encoder input is an explicit CSI matrix or a precoding vector H. The encoder output is a series of feature blocks, i.e., a feature block set. .like Figure 10 As shown, a series of feature blocks can be represented as f 1. f 2. f 3、…、 f K .

[0214] In the K feature blocks, the first k The importance of each feature block is defined as its marginal contribution to the task reward: ; in, The return function described above; This represents the CSI representation obtained by the network device based on the decoding / reconstruction of received feature blocks. As mentioned earlier, in addition to marginal contribution, importance can also be necessity.

[0215] See also Figure 10 K feature blocks are mapped to multiple reporting groups, namely group 1, group 2 to group V. Group 1 is the core base group, group 2 is the first-level enhancement group, and group V is the detail supplement group. Group 3 or other groups not shown can be multi-level enhancement groups or other reporting groups.

[0216] As discussed earlier, group-level importance can be defined as the marginal benefit of the group's features as a whole to the task. For the... v The additional incremental benefit for the group, assuming the previous group has already reached its goal, is: ;in, ; ; This is the CSI representation obtained solely based on the feature blocks contained in the preceding group.

[0217] The grouping configuration corresponding to the above V reporting groups not only divides the feature blocks into several groups, but also satisfies the group-level asymptotic availability and predictability constraints based on multiple benefit thresholds, so that the system can still operate stably when only a portion of the groups are received, and monotonically increases when more groups are received.

[0218] The three constraints satisfied by the group configuration can be expressed as follows.

[0219] Constraint 1: Each group must meet its respective revenue threshold constraint; that is, different prefix groups must satisfy revenue thresholds at each level. For example, group 1 must reach the minimum revenue threshold. Higher prefix groups satisfy the revenue thresholds at each level: , .

[0220] Constraint 2: Monotonically increasing returns. For example, when receiving both Group 1 and Group 2, enhancing details based on Group 1 does not decrease overall performance. And so on: .

[0221] Constraint 3: Diminishing marginal returns for each group can be expressed as: .

[0222] In addition, the block set within the reporting group must be recognizable by both the terminal device and the network device, and the feature block ID must be known to both parties. When there are multiple ordered partitions that satisfy the threshold and asymptotic constraints, the implementation can choose any one of the solutions, or select according to predefined rules.

[0223] Specifically, the process for determining group configurations based on importance is as follows: (1) Given a payoff threshold sequence ,in, Network devices are configurable or business-defined, and the minimum revenue threshold required for the task must be met. ; (2) Using sets This indicates which feature blocks each group contains, and assumes the encoder outputs a set of feature blocks. Feature blocks ; (3) Define the block index complete set ; (4) Define a set of block indices for V groups: The requirement is that they form an ordered partition, meaning that each feature block belongs to exactly one group: ; (5) Define before receipt v The set of cumulative information (feature index) following the group, i.e., the group prefix is ​​continuous: ; (6) Define the receiver in the case of only obtaining The CSI representation obtained from decoding / reconstruction of the corresponding block: ; (7) Define the corresponding task rewards: ; (8) Based on the above definition, the three constraints are expressed as follows: Constraint 1: Satisfy the task reward threshold for each group: , ; Constraint 2: The overall return does not decrease when more groups are received. ; Constraint 3: Diminishing marginal returns: ,Right now: 。

[0224] It should be noted that the solution method allows for various offline construction strategies. That is, it can either find the optimal or near-optimal partition that satisfies the three constraints through optimization / iteration, or it can use a more easily implemented greedy algorithm (e.g., first ensuring group 1 reaches the threshold, then adding blocks level by level until the next threshold is met). Regardless of the algorithm used, the final output is fixed as the grouping mapping and related rules in the grouping configuration, enabling both the sender and receiver to consistently identify which blocks each group contains. Furthermore, it allows for block indices within a group to be non-contiguous in order of importance, thus improving flexibility.

[0225] See also Figure 10 The UE can progressively send multiple reporting groups based on group-level importance. After receiving at least one reporting group, the decoder on the gNB side can output the reconstructed explicit CSI matrix or precoding vector. Among them, Group 1, as the core foundation group, can be used to decode primary information, while Group 1, Group 2, and other enhancement groups can be used to decode fine information. It should be understood that the legend on the gNB side is for illustration only; the primary information and fine information are the input content of the UE before compression after being recovered by the gNB (such as channel matrix, precoding vector, etc.).

[0226] Depend on Figure 10 As can be seen, gNB can also determine whether to append incremental feature blocks or reporting groups. The appending of incremental feature blocks or reporting groups can be implemented based on the context ID. The context ID can be associated with the first identifier.

[0227] Figure 10 As an example of an AI / ML-enhanced CSI compression scenario, a method for compressed feature grouping and progressive incremental reporting is demonstrated. When reporting packets based on the first structure, the encapsulation of packet payloads in UCI, the resource-based progressive reporting method, and the discarding mechanism are also issues that need to be considered. The terminal device can perform encapsulation, reporting, and / or discarding based on the network device configuration.

[0228] In some implementations, when a terminal device performs CSI feedback, it can determine, based on the network device's configuration, some or all of the latent vectors to be sent, the relevant encapsulation rules, the reporting or appending strategy, and the discarding mechanism. The network device's configuration can be determined based on the first configuration information or other configuration information; this is not limited here.

[0229] As one implementation approach, during the inference phase, network devices can use a strategy based on semi-static RRC configuration and dynamic downlink control information (DCI) control to provide instructions. For example, for the structure configuration corresponding to the first structure, the network device can indicate the bearer and reporting rules during the inference phase, enabling terminal devices to progressively report based on the first structure and predictably discard data.

[0230] For example, network devices can perform semi-static configuration via RRC signaling, sending or specifying the first identifier (e.g., groupProfileID) and its rule set to terminal devices, or sending packet parameters for CSI reports. The relevant parameters must include at least one of the following: Structural parameters: V; The relevant parameters for group mapping are: NvMapping and groupPriorityOrder. Add the following parameters to the relevant strategy: contextMode and contextReuseRule; The parameter for the minimum guaranteed reporting group when resources are scarce: minGuaranteedGroup (usually group 1 when the first structure consists of multiple reporting groups).

[0231] Building upon this, network devices can also perform successive dynamic control via DCI to achieve dynamic resource constraints and fine-grained scheduling. The relevant parameters of DCI include at least one of the following: Indicates the highest group number that is allowed to be sent to this time: maxGroupID; The terminal device is requested to append parameters for a specified group: requestAdditionalGroups; The parameter `requestAdditionalBlocks` requests the terminal device to append a specified feature block.

[0232] In the example above, the first configuration information can be carried in RRC signaling and / or DCI.

[0233] As one implementation approach, the terminal device can determine the payload encapsulation and discarding strategy based on the different parts of the latent vector divided by the first structure. When the terminal device determines to send part or all of the latent vector, the payload encapsulation strategy needs to take into account the relevant reporting information.

[0234] Taking multiple reporting groups generated by feature block mapping as an example, the terminal device generates feature blocks in each CSI feedback opportunity. The payload is organized into V groups based on the NvMapping determined by the first identifier or first configuration information. To ensure receiver parsing and context alignment, a reporting content indicator control field, i.e., reporting information, is set in the UCI payload. The reporting information may include: First identifier: Used to identify the grouping configuration used, ensuring that both the terminal side and the network side use consistent grouping configuration and encapsulation rules; groupNum: Used to indicate the highest group number actually sent this time, enabling network devices to determine the parsing length assumption for subsequent group payloads; contextID: Used for subsequent incremental appends and cache consistency.

[0235] In some implementations, at least one parameter of the aforementioned reported information is included in the CSI report configuration.

[0236] For each reporting group v Group payload is defined as the concatenation of feature block code segments within a group: ;in, For feature blocks The quantized / encapsulated bitstream segment. The total payload reported in one go can be expressed as: payload = payload1 ... payload groupNum .

[0237] The previous section introduced the configuration methods for payload encapsulation and discarding strategies. When the terminal device sends a portion of the latent vector, the network side can also trigger an on-demand incremental appending mechanism to improve CSI accuracy. Under conditions of controlling resource fluctuations or UCI conflicts, it is necessary to establish clear rules for feature selection, discarding, and incremental retransmission. For example, the system can construct an operable rule system, specifying the transmission and discard sets of terminal devices under different feedback budgets, and defining when and how incremental retransmission is allowed for the same CSI context, making the reporting behavior predictable, achievable, and accurately described in a standardized manner.

[0238] The network device can send a first signaling message to the terminal device. The first signaling message can instruct the terminal device to perform CSI feedback appending. After receiving the first signaling message, the terminal device can send a specified part of the latent vector according to the appending strategy corresponding to the first signaling message or the first structure.

[0239] As one implementation, the first signaling can be carried in the DCI. The first signaling can be indicated by the parameters requestAdditionalGroups or requestAdditionalBlocks.

[0240] As an implementation approach, the first signaling can be triggered based on scenarios such as beam management or precoding mode switching. For example, switching from coarse-grained beam selection to fine-grained beamforming or a sudden increase in service requirements for reliability / throughput. Another example is situations where more detailed information is needed.

[0241] As one implementation approach, after receiving the first signaling, the terminal device can use a context ID. As mentioned earlier, the context ID can be associated with a first identifier. For example, in some cases, the context ID can be the ID of the first model or the pairing ID corresponding to the first model, or it can be a CSI report configuration ID.

[0242] In some implementations, when a specified portion is determined according to a first signaling, the first signaling can indicate the index corresponding to the specified portion. For example, when a terminal device sends a portion of a latent vector based on at least one reporting group, the first signaling can indicate the index of the specified reporting group or an appended reporting group. Similarly, when a terminal device sends a portion of a latent vector based on at least one reporting group or feature block, the first signaling can indicate the index of the specified feature block or an appended feature block. Furthermore, when a terminal device sends a portion of a latent vector based on multiple layers, the first signaling can indicate the index of the appended layer.

[0243] As an example, when a terminal device sends at least one of a plurality of reporting groups, the additional reporting group can be one or more reporting groups other than the at least one reporting group.

[0244] As an example, when a terminal device sends at least one of a plurality of reporting groups, the additional feature block can be one or more feature blocks other than those contained in the at least one reporting group.

[0245] As an example, when a terminal device sends at least one of a plurality of reporting groups, the additional layer can be one or more layers other than the layer corresponding to the at least one reporting group.

[0246] In some implementations, when the specified part is determined according to the append strategy corresponding to the first structure, the first signaling can be used to trigger CSI feedback appending or to indicate the performance requirements of CSI feedback appending.

[0247] As one implementation, the terminal device can determine the portion to be appended based on at least one of the following: a portion of the latent vectors already sent, an appending strategy, and performance requirements for CSI feedback appending. These performance requirements can be CSI accuracy requirements or corresponding compressed features. For example, the terminal device can determine one or more reporting groups to be appended based on at least one reported group already sent and the appending strategy. Similarly, the terminal device can determine one or more feature blocks to be appended based on at least one reported group already sent and performance requirements.

[0248] As an implementation method, the CSI feedback appending performed by the terminal device can also be called on-demand incremental appending of the same context CSI. Based on this appending mechanism, a balance between resource efficiency and performance can be achieved, making it particularly suitable for scenarios where channel changes are slow or only basic accuracy is required initially. Specifically, the core of this appending mechanism is to split the packet payload of the same context CSI into multiple transmissions in time using contextID, and allow network devices to trigger the terminal device to retransmit only the missing enhancement group or specified feature block as needed, such as... Figure 11 As shown.

[0249] Figure 11 This is a schematic diagram of a possible implementation of CSI feedback supplementation based on the reporting group. Figure 11 The process of triggering the append mechanism is shown through three time points, T1 to T3. Figure 11 The additional procedure shown is executed by the UE and gNB.

[0250] At time T1: The UE initially reports the basic group. Before time T1, the gNB has configured contextMode via RRC signaling, or configured CSI reports and related policies. The UE generates and reports the UCI payload for the first time. CSI part 1 contains contextID=X and the first identifier, and at least group 1 is included in CSI part 2. After receiving the data, the gNB can complete basic precoding / scheduling decisions based on the information reported by the UE (content indication control fields) and cache the status of the received groups / blocks locally.

[0251] At time T2: The gNB triggers an add request. When an event occurs requiring higher-precision CSI, the gNB initiates an add request via the DCI, i.e., the first signaling. For example, the first signaling can request additional groups to add from group 1 to group m. Alternatively, the first signaling can request additional blocks to add a specified number of feature blocks. The triggering condition for the first signaling can be from beam management or precoding mode switching. In some cases, the DCI may include a field indicating the CSI group number to be added.

[0252] At time T3: The UE performs incremental reporting, only filling in missing parts. In subsequent CSI feedback opportunities, the UE carries the same contextID=X and the same first identifier, or under the same CS report configuration, and only sends the code stream segments corresponding to the enhancement group or specified group / feature block that were not sent before. After receiving the data, the gNB merges the received payloads multiple times based on the CSI report configuration ID or contextID to obtain a more complete feature set and drive the decoder to output higher-precision CSI recovery or better downstream decisions.

[0253] based on Figure 11 The illustrated appending mechanism allows the network to supplement the accuracy of the same CSI report configuration on demand without changing the model attributes, thereby avoiding the waste of resources caused by continuous high-overhead reporting and quickly obtaining enhanced information when needed.

[0254] As described above, after determining the first structure and generating a latent vector with the first structure, the terminal device can first send a portion of the latent vector for CSI feedback. Upon receiving the first signaling, the terminal device can also send the remaining portion of the latent vector to append more CSI feedback. The following section combines... Figure 12 and Figure 13 Taking the first structure indicating multiple reporting groups as an example, the signaling flow of the initial / no incremental reporting and the triggered request to add incremental reporting are illustrated separately. Figure 12 and Figure 13 The introduction is all from the perspective of the interaction between UE and gNB. UE is the terminal device, and gNB is the network device.

[0255] See Figure 12 In step S1210, the UE and gNB initially negotiate a set of group profiles. In some cases, the initial negotiation may occur during the period when the UE submits an suitability report.

[0256] In step S1220, the gNB sends RRC signaling to the UE. The RRC signaling can specify the group configuration and appending policy used by the UE through groupProfileID. groupProfileID can be carried in the first configuration information.

[0257] In step S1230, the gNB sends DCI signaling to the UE and dynamically controls resource scheduling through maxGroupID and CSI-RS.

[0258] In step S1240, the encoder on the UE side outputs feature blocks and groups and discards the feature blocks according to the groupProfile corresponding to groupProfileID and the allocated resources. In some cases, GroupProfile is the configuration of CSI reported by the UE.

[0259] In step S1250, the UE sends UCI signaling to the gNB, including reporting information and the assembled group payload. The reporting information is the content indication control field, which may contain contextID=X.

[0260] See Figure 13 In step S1310, gNB triggers a request to add an enhancement group / specified group / feature block increment.

[0261] In step S1320, the gNB sends DCI signaling (first signaling) to the UE. The DCI signaling includes the parameters requestAdditionalGroups or requestAdditionalBlocks.

[0262] In step S1330, the UE transmits the code stream segments corresponding to the enhancement group / designated group / feature block that were not previously transmitted.

[0263] In step S1340, the UE sends UCI signaling to the gNB, including reporting information and additional group payload. The reporting information is the content indication control field, which may contain contextID=X.

[0264] In step S1350, the gNB merges the received payloads from multiple times and drives the decoder to output a higher-precision CSI reconstruction.

[0265] In summary, this application can construct the features (e.g., semantic features) of the CSI compressed output into a payload structure with clear group boundaries and importance levels. This allows for selective transmission and discarding based on priority / importance when the feedback budget changes, and supports progressive incremental reporting of the same CSI context when higher accuracy is required. Based on this method, predictable behavior consistent with the existing CSI segmentation concept in 5G NR, namely, Part 1 is fixed and decorable, while Part 2 can be enhanced and can be grouped and discarded, can be achieved.

[0266] Through the aforementioned feature grouping and progressive incremental reporting mechanism, this application can prioritize the reliable delivery of high-importance features when uplink control resources are limited or UCI conflicts occur, and discard low-importance features in a controllable order, making performance degradation smoother and reducing the overall failure risk. For example, when resources permit or business needs require improved precoding accuracy, enhanced features can be added as needed to monotonically improve reconstruction quality or precoding decision quality, avoiding repeated transmission of the entire packet payload and improving feedback resource utilization efficiency. At the same time, the grouping structure and drop / incremental rules provide clear protocol-based expressions for what the terminal device sends, what it drops, and when it replenishes it, thereby improving the predictability, feasibility, and interoperability verification feasibility of system behavior, and enhancing the interpretability and standardization of CSI compression features at the engineering level.

[0267] 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.

[0268] Figure 14 This is a schematic block diagram of another 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 and a second processing unit 1420.

[0269] The first processing unit 1410 can be used to determine the first structure based on the first configuration information and / or the capabilities of the terminal device.

[0270] The second processing unit 1420 can be used to perform CSI compression based on the first model to obtain a latent vector with the first structure; wherein the first configuration information is determined based on at least one of the reported information of the terminal device, the configuration information issued by the network device, and the protocol predefined information.

[0271] Optionally, the first model is trained based on auxiliary information, and the device 1400 further includes a first receiving unit, which can be used to receive the auxiliary information; wherein, the auxiliary information includes parameter information of the first structure.

[0272] Optionally, the reported information includes the capability information of the terminal device, which is used to indicate whether the terminal device supports latent vectors with the first structure.

[0273] Optionally, the device 1400 further includes: a third processing unit, configured to monitor the first model and / or the second model based on part or all of the latent vectors output by the first model; and a first sending unit, configured to report the monitoring results of the first model and / or the second model; wherein the monitoring results are used to indicate or determine whether the first model and / or the second model are working properly.

[0274] Optionally, the device 1400 further includes a second transmitting unit, which can be used to transmit part or all of the latent vector based on the first structure.

[0275] Optionally, the device 1400 further includes a third transmitting unit, which can be used to transmit part or all of the reporting information of the latent vector to the network device; wherein part or all of the latent vector is used for CSI feedback, and the reporting information is carried in a first part corresponding to the CSI feedback, or the reporting information is carried in a second part corresponding to the CSI feedback.

[0276] Optionally, the first structure includes the latent vector being divided into multiple reporting groups; the second sending unit is further configured to send at least one of the multiple reporting groups to the network device; wherein the at least one reporting group is used by the network device to determine CSI, and the at least one reporting group is determined based on multiple CSI-related benefits corresponding to the multiple reporting groups.

[0277] Optionally, the reporting group with the smallest CSI-related benefit among the at least one reporting group is the first reporting group, and the CSI-related benefit corresponding to the first reporting group is greater than the CSI-related benefit corresponding to any reporting group other than the at least one reporting group among the plurality of reporting groups.

[0278] Optionally, the first structure includes the latent vector being divided into K feature blocks, where K is a positive integer; the second sending unit is further configured to send at least one feature block among the K feature blocks based on feature importance; or, the second processing unit 1420 is further configured to generate multiple reporting groups to be sent based on the feature importance of the K feature blocks; wherein the feature importance is associated with CSI-related benefits.

[0279] Optionally, the K feature blocks are arranged in descending order according to the importance of the features, and any one of the multiple reporting groups includes at least one feature block. The multiple reporting groups are generated based on multiple revenue thresholds and the K feature blocks arranged in descending order.

[0280] Optionally, the K feature blocks satisfy one of the following: the K feature blocks have the same quantization level; at least two feature blocks among the K feature blocks have different quantization levels; all feature blocks in the same reporting group have the same quantization level, and feature blocks in different reporting groups have different quantization levels; at least two feature blocks in the same reporting group have different quantization levels.

[0281] Optionally, the first configuration information includes at least one of the following: the number of reporting groups, the number of feature blocks, or the number of layers corresponding to the first structure; the mapping relationship between multiple feature blocks and multiple reporting groups corresponding to the first structure; multiple benefit thresholds corresponding to the first structure; the discarding rule corresponding to the first structure; the encapsulation rule of at least one reporting group or at least one feature block sent by the terminal device; and the appending strategy corresponding to the first structure.

[0282] Optionally, the first structure includes multiple layers corresponding to the latent vector, and the appending strategy includes appending based on the layer index.

[0283] Optionally, the first configuration information corresponds to a first identifier, and the first identifier is at least one of the following: the identifier of the first configuration information, the identifier of the CSI report configuration corresponding to the first configuration information, the pairing identifier corresponding to the first model, and the identifier corresponding to the first structure.

[0284] Optionally, the apparatus 1400 further includes: a second receiving unit, configured to receive a first signaling, the first signaling being used to instruct the terminal device to perform CSI feedback appending; and a fourth sending unit, configured to send a specified portion of the latent vector according to the first signaling or the appending strategy corresponding to the first structure; wherein, when the specified portion is determined according to the first signaling, the first signaling is also used to indicate the index corresponding to the specified portion.

[0285] Optionally, the first model is an AI or ML-based encoder.

[0286] Optionally, the first processing unit 1410 and the second processing unit 1420 in the device 1400 may be a processor 1610. The device 1400 may also include a memory 1620 and a transceiver 1630, see details below. Figure 16 .

[0287] Figure 15This 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 15 The apparatus 1500 shown includes a first processing unit 1510 and a second processing unit 1520.

[0288] The first processing unit 1510 can be used to determine the first structure based on the first configuration information and / or the capabilities of the terminal device.

[0289] The second processing unit 1520 can be used to determine CSI based on the second model and the first structure; wherein, the first structure is related to the first model of the terminal device, the first model is used by the terminal device to determine the latent vector with the first structure after CSI compression, and the first configuration information is determined according to at least one of the reported information of the terminal device, the configuration information issued by the network device, and the protocol predefined information.

[0290] Optionally, the first model is trained based on auxiliary information, and the device 1500 further includes a first transmitting unit, which can be used to transmit the auxiliary information to the terminal device; wherein, the auxiliary information includes parameter information of the first structure.

[0291] Optionally, the reported information includes the capability information of the terminal device, which is used to indicate whether the terminal device supports latent vectors with the first structure.

[0292] Optionally, the device 1500 further includes a third processing unit, which can be used to monitor the first model and / or the second model based on part or all of the latent vectors sent by the terminal device; or, the device 1500 further includes a first receiving unit, which can be used to receive the monitoring results of the first model and / or the second model sent by the terminal device; wherein the monitoring results of the first model and / or the second model are used to indicate or determine whether the first model and / or the second model are working properly.

[0293] Optionally, the device 1500 further includes a second receiving unit, which can be used to receive part or all of the latent vector based on the first structure.

[0294] Optionally, the device 1500 further includes a third receiving unit, which can be used to receive part or all of the reporting information of the latent vector sent by the terminal device; wherein part or all of the latent vector is used for CSI feedback, and the reporting information is carried in a first part corresponding to the CSI feedback, or the reporting information is carried in a second part corresponding to the CSI feedback.

[0295] Optionally, the first structure includes the latent vector being divided into multiple reporting groups; the second receiving unit is further configured to receive at least one reporting group sent by the terminal device; the second processing unit 1520 is further configured to determine CSI based on the at least one reporting group; wherein the at least one reporting group belongs to the multiple reporting groups, and the at least one reporting group is determined based on multiple CSI-related benefits corresponding to the multiple reporting groups.

[0296] Optionally, the reporting group with the smallest CSI-related benefit among the at least one reporting group is the first reporting group, and the CSI-related benefit corresponding to the first reporting group is greater than the CSI-related benefit corresponding to any reporting group other than the at least one reporting group among the plurality of reporting groups.

[0297] Optionally, the first structure includes the latent vector being divided into K feature blocks, where K is a positive integer, and the K feature blocks are sent or multiple reporting groups are generated based on feature importance, wherein the feature importance is associated with CSI-related benefits.

[0298] Optionally, the K feature blocks are arranged in descending order according to the importance of the features, and any one of the multiple reporting groups includes at least one feature block. The multiple reporting groups are generated based on multiple revenue thresholds and the K feature blocks arranged in descending order.

[0299] Optionally, the K feature blocks satisfy one of the following: the K feature blocks have the same quantization level; at least two feature blocks among the K feature blocks have different quantization levels; all feature blocks in the same reporting group have the same quantization level, and feature blocks in different reporting groups have different quantization levels; at least two feature blocks in the same reporting group have different quantization levels.

[0300] Optionally, the first configuration information includes at least one of the following: the number of reporting groups, the number of feature blocks, or the number of layers corresponding to the first structure; the mapping relationship between the multiple feature blocks corresponding to the first structure and the multiple reporting groups; the multiple benefit thresholds corresponding to the first structure; the discarding rules corresponding to the first structure; the encapsulation rules for at least one reporting group or at least one feature block sent by the terminal device; and the appending strategy corresponding to the first structure.

[0301] Optionally, the first structure includes multiple layers corresponding to the latent vector, and the appending strategy includes appending based on the layer index.

[0302] Optionally, the first configuration information corresponds to a first identifier, and the first identifier is at least one of the following: the identifier of the first configuration information, the identifier of the CSI report configuration corresponding to the first configuration information, the pairing identifier corresponding to the first model, and the identifier corresponding to the first structure.

[0303] Optionally, the apparatus 1500 further includes: a second transmitting unit, configured to transmit a first signaling, the first signaling being used to instruct the terminal device to perform CSI feedback appending; and a fourth receiving unit, configured to receive a specified portion of the latent vector transmitted by the terminal device; wherein the specified portion is determined according to the appending strategy corresponding to the first signaling or the first structure; when the specified portion is determined according to the first signaling, the first signaling is also used to indicate the index corresponding to the specified portion.

[0304] Optionally, the first model is an AI or ML-based encoder, and the second model is an AI or ML-based decoder.

[0305] Optionally, the first processing unit 1510 and the second processing unit 1520 in the device 1500 may be a processor 1610. The device 1500 may also include a memory 1620 and a transceiver 1630, as detailed below. Figure 16 .

[0306] 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.

[0307] 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.

[0308] 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.

[0309] 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.

[0310] 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.

[0311] 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)).

[0312] 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.

[0313] 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.

[0314] This application also provides a computer program. This computer program can be applied to the 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.

[0315] 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. Moreover, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0316] 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.

[0317] 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.

[0318] 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.

[0319] 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.

[0320] 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.

[0321] 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.

[0322] 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.

[0323] 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.

[0324] 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.

[0325] 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.

[0326] 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 determines the first structure based on the first configuration information and / or the capabilities of the terminal device; The terminal device performs Channel State Information (CSI) compression based on the first model to obtain a latent vector with the first structure; The first configuration information is determined based on at least one of the information reported by the terminal device, the configuration information issued by the network device, and the protocol predefined information.

2. The method according to claim 1, characterized in that, The first model is trained based on auxiliary information, and the method further includes: The terminal device receives the auxiliary information; The auxiliary information includes the parameter information of the first structure.

3. The method according to claim 1 or 2, characterized in that, The reported information includes the capability information of the terminal device, which is used to indicate whether the terminal device supports latent vectors with the first structure.

4. The method according to any one of claims 1-3, characterized in that, The method further includes: The terminal device monitors the first model and / or the second model based on part or all of the latent vectors output by the first model; The terminal device reports the monitoring results of the first model and / or the second model; The monitoring results are used to indicate or determine whether the first model and / or the second model are working properly.

5. The method according to any one of claims 1-4, characterized in that, The method further includes: The terminal device transmits part or all of the latent vector based on the first structure.

6. The method according to claim 5, characterized in that, The method further includes: The terminal device sends part or all of the latent vector reporting information to the network device; Wherein, part or all of the latent vector is used for CSI feedback, and the reported information is carried in the first part corresponding to the CSI feedback, or the reported information is carried in the second part corresponding to the CSI feedback.

7. The method according to claim 5 or 6, characterized in that, The first structure includes dividing the latent vector into multiple reporting groups; the terminal device sends part or all of the latent vector based on the first structure, including: The terminal device sends at least one of the plurality of reporting groups to the network device; The at least one reporting group is used by the network device to determine CSI, and the at least one reporting group determines the CSI-related benefits based on the multiple reporting groups corresponding to the multiple CSI.

8. The method according to claim 7, characterized in that, The reporting group with the smallest CSI-related benefit among the at least one reporting group is the first reporting group, and the CSI-related benefit corresponding to the first reporting group is greater than the CSI-related benefit corresponding to any reporting group other than the at least one reporting group among the plurality of reporting groups.

9. The method according to any one of claims 5-8, characterized in that, The first structure includes dividing the latent vector into K feature blocks, where K is a positive integer; the terminal device sends part or all of the latent vector based on the first structure, including: The terminal device sends at least one feature block from the K feature blocks based on feature importance; or... The terminal device generates multiple reporting groups to be sent based on the feature importance of the K feature blocks; The importance of the aforementioned feature is related to CSI-related benefits.

10. The method according to claim 9, characterized in that, The K feature blocks are arranged in descending order according to the importance of the features. Each of the plurality of reporting groups includes at least one feature block. The plurality of reporting groups are generated based on a plurality of revenue thresholds and the K feature blocks arranged in descending order.

11. The method according to claim 9 or 10, characterized in that, The K feature blocks satisfy one of the following: The K feature blocks have the same quantization level; At least two of the K feature blocks have different quantization levels; All feature blocks in the same reporting group have the same quantization level, while feature blocks in different reporting groups have different quantization levels. At least two feature blocks in the same reporting group have different quantization levels.

12. The method according to any one of claims 1-11, characterized in that, The first configuration information includes at least one of the following: The number of reporting groups, the number of feature blocks, or the number of layers corresponding to the first structure; The mapping relationship between multiple feature blocks corresponding to the first structure and multiple reporting groups; Multiple revenue thresholds corresponding to the first structure; The discarding rule corresponding to the first structure; The encapsulation rules for at least one reporting group or at least one feature block sent by the terminal device; The appending strategy corresponding to the first structure.

13. The method according to claim 12, characterized in that, The first structure includes multiple layers corresponding to the latent vector, and the appending strategy includes appending based on the layer index.

14. The method according to any one of claims 1-13, characterized in that, The first configuration information corresponds to a first identifier, which is at least one of the following: the identifier of the first configuration information, the identifier of the CSI report configuration corresponding to the first configuration information, the pairing identifier corresponding to the first model, and the identifier corresponding to the first structure.

15. The method according to any one of claims 1-14, characterized in that, The method further includes: The terminal device receives a first signaling message, which instructs the terminal device to add CSI feedback. The terminal device sends a specified portion of the latent vector according to the appending strategy corresponding to the first signaling or the first structure. When the designated portion is determined according to the first signaling, the first signaling is also used to indicate the index corresponding to the designated portion.

16. The method according to any one of claims 1-15, characterized in that, The first model is an encoder based on artificial intelligence or machine learning.

17. A method for wireless communication, characterized in that, include: The network device determines the first structure based on the first configuration information and / or the capabilities of the terminal device; Network devices determine Channel State Information (CSI) based on the second model and the first structure; Wherein, the first structure is related to the first model of the terminal device, the first model is used by the terminal device to determine the latent vector with the first structure after CSI compression, and the first configuration information is determined based on at least one of the reported information of the terminal device, the configuration information issued by the network device, and the protocol predefined information.

18. The method according to claim 17, characterized in that, The first model is trained based on auxiliary information, and the method further includes: The network device sends the auxiliary information to the terminal device; The auxiliary information includes the parameter information of the first structure.

19. The method according to claim 17 or 18, characterized in that, The reported information includes the capability information of the terminal device, which is used to indicate whether the terminal device supports latent vectors with the first structure.

20. The method according to any one of claims 17-19, characterized in that, The method further includes: The network device monitors the first model and / or the second model based on part or all of the latent vectors sent by the terminal device; or... The network device receives the monitoring results of the first model and / or the second model sent by the terminal device; The monitoring results of the first model and / or the second model are used to indicate or determine whether the first model and / or the second model are working properly.

21. The method according to any one of claims 17-20, characterized in that, The method further includes: The network device receives part or all of the latent vector based on the first structure.

22. The method according to claim 21, characterized in that, The method further includes: The network device receives part or all of the reported information of the latent vector sent by the terminal device; Wherein, part or all of the latent vector is used for CSI feedback, and the reported information is carried in the first part corresponding to the CSI feedback, or the reported information is carried in the second part corresponding to the CSI feedback.

23. The method according to claim 21 or 22, characterized in that, The first structure includes dividing the latent vectors into multiple reporting groups; the network device receives part or all of the latent vectors based on the first structure, including: The network device receives at least one reporting group sent by the terminal device; The network device determines the CSI based on the at least one reporting group; Wherein, the at least one reporting group belongs to the plurality of reporting groups, and the at least one reporting group is determined based on the plurality of CSI-related benefits corresponding to the plurality of reporting groups.

24. The method according to claim 23, characterized in that, The reporting group with the smallest CSI-related benefit among the at least one reporting group is the first reporting group, and the CSI-related benefit corresponding to the first reporting group is greater than the CSI-related benefit corresponding to any reporting group other than the at least one reporting group among the plurality of reporting groups.

25. The method according to any one of claims 21-24, characterized in that, The first structure includes the latent vector being divided into K feature blocks, where K is a positive integer. The K feature blocks are sent or multiple reporting groups are generated based on feature importance, and the feature importance is associated with CSI-related benefits.

26. The method according to claim 25, characterized in that, The K feature blocks are arranged in descending order according to the importance of the features. Each of the plurality of reporting groups includes at least one feature block. The plurality of reporting groups are generated based on a plurality of revenue thresholds and the K feature blocks arranged in descending order.

27. The method according to claim 25 or 26, characterized in that, The K feature blocks satisfy one of the following: The K feature blocks have the same quantization level; At least two of the K feature blocks have different quantization levels; All feature blocks in the same reporting group have the same quantization level, while feature blocks in different reporting groups have different quantization levels. At least two feature blocks in the same reporting group have different quantization levels.

28. The method according to any one of claims 17-27, characterized in that, The first configuration information includes at least one of the following: The number of reporting groups, the number of feature blocks, or the number of layers corresponding to the first structure; The mapping relationship between multiple feature blocks corresponding to the first structure and multiple reporting groups; Multiple revenue thresholds corresponding to the first structure; The discarding rule corresponding to the first structure; The encapsulation rules for at least one reporting group or at least one feature block sent by the terminal device; The appending strategy corresponding to the first structure.

29. The method according to claim 28, characterized in that, The first structure includes multiple layers corresponding to the latent vector, and the appending strategy includes appending based on the layer index.

30. The method according to any one of claims 17-29, characterized in that, The first configuration information corresponds to a first identifier, which is at least one of the following: the identifier of the first configuration information, the identifier of the CSI report configuration corresponding to the first configuration information, the pairing identifier corresponding to the first model, and the identifier corresponding to the first structure.

31. The method according to any one of claims 17-30, characterized in that, The method further includes: The network device sends a first signaling message, which instructs the terminal device to add CSI feedback. The network device receives a specified portion of the latent vector sent by the terminal device; The designated portion is determined according to the appending strategy corresponding to the first signaling or the first structure; when the designated portion is determined according to the first signaling, the first signaling is also used to indicate the index corresponding to the designated portion.

32. The method according to any one of claims 17-31, characterized in that, The first model is an encoder based on artificial intelligence or machine learning, and the second model is a decoder based on artificial intelligence or machine learning.

33. 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 1-16.

34. A device 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 17-32.

35. A communication device, characterized in that, Includes units or modules for performing the method as described in any one of claims 1-16 or 17-32.

36. 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 1-16 or 17-32.

37. 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 1-16 or 17-32.

38. 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 1-16 or 17-32.

Images