Training method of csi information processing model, csi information processing method and storage medium

Abstract

Embodiments of the present application provide a CSI information processing model training method, a CSI information processing method and a storage medium. The method comprises: a user equipment receiving data collection information sent by a network equipment; the user equipment obtaining CSI information according to the data collection information; then, the user equipment providing the CSI information to the network equipment, and the network equipment side training a CSI information processing model. After the network equipment side trains the CSI information processing model, the network equipment shares parameters of the CSI information processing model to the user equipment, and the user equipment obtains a CSI information processing model on the user equipment side according to the parameters shared by the network equipment. Thus, according to the same CSI information, corresponding CSI information processing models are obtained on the network equipment side and the user equipment side respectively. When processing the same or similar CSI information subsequently, the corresponding CSI information processing model can be called according to the characteristics of the network equipment and the CSI information for processing. The embodiments of the present application can improve the compression efficiency of the CSI information and reduce the information restoration error.

Description

Technical Field

[0001] This application relates to the field of wireless communication, and more particularly to a training method for a CSI information processing model, a CSI information processing method, and a computer storage medium. Background Technology

[0002] Channel state information (CSI) reporting is one of the key technologies for achieving efficient communication. CSI describes the channel attributes of a communication link, helping communication systems adapt to current channel conditions, especially in multi-antenna systems, ensuring high reliability and high-speed communication. Today, with the evolution of communication technologies, the amount of CSI information has increased dramatically. The storage and transmission of this CSI information has become a critical issue that communication systems urgently need to address, posing unprecedented challenges to traditional communication architectures and processing methods. Summary of the Invention

[0003] This application provides a training method for a CSI information processing model, a CSI information processing method, and a computer storage medium, which helps to efficiently transmit CSI information.

[0004] In a first aspect, embodiments of this application provide a training method for a CSI information processing model, comprising: receiving data collection information sent by a network device; obtaining CSI information based on the data collection information; and obtaining a first CSI information processing model based on the CSI information.

[0005] Using the above method, corresponding CSI information can be obtained based on the data collection information sent by the network device. Furthermore, a CSI information processing model can be trained or generated based on the obtained CSI information. Thus, the resulting CSI information processing model corresponds to the data collection information. The network device can classify the data collection information and distinguish the CSI information processing models corresponding to different CSI information. When processing CSI information, it can select the model corresponding to the CSI information. This helps the user device to efficiently process CSI information and the network device to accurately reconstruct CSI information during transmission between the user device and the network device.

[0006] In one implementation, the data collection information includes first data collection information and / or second data collection information.

[0007] The above method allows us to obtain different CSI information through different data collection methods, and thus generate different CSI information processing models using different CSI information.

[0008] In one implementation, when the collected information includes the first information of data collection, obtaining a first CSI information processing model based on the CSI information includes: sending the CSI information to the network device; receiving parameters of a second CSI information processing model sent by the network device; and obtaining the first CSI information processing model based on the parameters.

[0009] Using the above method, network devices can perform model training on the network side based on CSI information to obtain a second CSI information processing model on the network device side. Then, based on the parameters of the second CSI information processing model sent by the network device side, a first CSI information processing model can be obtained, thereby maintaining the corresponding CSI information processing model parameter adaptation between network devices and user devices.

[0010] In one implementation, where the data collection information includes the first data collection information, the CSI information is used by the network device to train a second CSI information processing model.

[0011] User equipment can collect CSI information in a targeted manner based on the characteristics of the CSI information indicated in the first information of data collection. As a result, the second CSI information processing model trained by the network equipment using this CSI information can be adapted to the first information of data collection and also to the associated ID corresponding to the first information of data collection.

[0012] In one embodiment, the second CSI information processing model includes a second encoder model and a second decoder model.

[0013] Network devices can obtain matching encoder and decoder models by training encoder and decoder model pairs. As a result, user equipment and network devices can use the matching encoder and decoder models respectively to encode and decode the same CSI information, thereby improving encoding efficiency and reducing the error between the reconstructed CSI information and the original CSI information.

[0014] In one implementation, the network device sends parameters of a second encoder model to the user equipment.

[0015] The network device sends the parameters of the second encoder model to the user equipment, and the user equipment can set the encoder model on the user equipment side according to the parameters of the second encoder model.

[0016] In one implementation, the data collection first information is used to indicate: characteristics of CSI information measured by the user equipment and / or CSI information measured by the user equipment.

[0017] By collecting the first information, the network device can instruct the user equipment to collect CSI information for training the second CSI information processing model, so that the user equipment can obtain CSI information in a targeted manner to train the second CSI information processing model and the first CSI information processing model.

[0018] In one implementation, the data collection first information includes at least one of the following: reported data sample type, reported data sample format, CSI-RS configuration information, transmission bandwidth information, maximum number of CSI data samples collected, data label consistency check mechanism, CSI dataset size, CSI data collection time interval or data collection sampling rate, CSI data encryption technology indication information, CSI data quality threshold information, and the association ID of the first CSI information processing model.

[0019] By collecting first information through data collection, various characteristics of CSI information can be defined. Network devices can send first information to user equipment to enable user equipment to collect the required CSI information.

[0020] In one implementation, when the collected information includes the second information of data collection, obtaining the first CSI information processing model based on the CSI information includes: training a reference CSI information processing model based on the CSI information to obtain the first CSI information processing model.

[0021] By acquiring CSI information based on the second information collected from the data collection, and then using the CSI information to train a reference CSI information processing model, a first CSI information processing model is obtained. This makes the first CSI information processing model compatible with the second information collected from the data collection, and also further ensures that the first CSI information processing model can be used normally on the user equipment.

[0022] In one implementation, before training a reference CSI information processing model based on the CSI information to obtain the first CSI information processing model, the method further includes: sending the CSI information to the network device; receiving parameters of a second CSI information processing model sent by the network device; obtaining a reference CSI information processing model based on the parameters; and obtaining the first CSI information processing model based on the CSI information and the reference CSI information processing model.

[0023] User equipment can send CSI information to network equipment, and then set and train the first CSI information processing model on the user equipment side according to the parameters of the second CSI information processing model returned by the network equipment, so that the first CSI information processing model and the second CSI information processing model are adapted to each other.

[0024] In one implementation, the reference CSI information processing model includes: a reference CSI information encoding model.

[0025] The reference CSI information encoding model is a model used to encode CSI information. By configuring the reference CSI information encoding model, a CSI information encoding model corresponding to the associated ID can be generated on the user equipment side, enabling efficient encoding operations for CSI information.

[0026] In one embodiment, the first CSI information processing model includes an actual encoder model. Obtaining the first CSI information processing model based on the CSI information and the reference CSI information processing model includes: training a nominal decoder model based on the reference CSI information encoding model; and training the nominal decoder model based on the CSI information to obtain the actual encoder model.

[0027] By training the nominal decoder model, we can eventually obtain the actual encoder model corresponding to the user device.

[0028] In one implementation, the second information for data collection includes at least one of the following: data sample type, reported data sample format, transmission bandwidth information, data tag consistency check mechanism for data collection, maximum number of CSI data collection samples, dataset size, time interval or sampling rate of CSI data collection, CSI-RS configuration information, threshold information for CSI data quality, association ID of the first CSI information processing model, and threshold information for model performance indicators.

[0029] By sending the second data collection information to the user equipment, the network device can define the characteristics of the CSI information that needs to be acquired. After training the first CSI information processing model using the CSI information, the first CSI information processing model can correspond to the second data collection information, the first data collection information, and the associated ID.

[0030] In one embodiment, when the data collection information includes the CSI-RS configuration information, the CSI-RS configuration information includes: the start time of reference signal transmission, the number of continuous transmissions of the reference signal or the period of reference signal transmission, and the end time of reference signal transmission.

[0031] With the CSI-RS configuration information, user equipment can collect CSI information more accurately according to the instructions of network equipment.

[0032] In one embodiment, if the collected information includes the second data collection information, before receiving the data collection information sent by the network device, the method further includes: sending data collection request information to the network device.

[0033] By sending a data collection request to the network device, the user equipment can inform the network device to send corresponding second data collection information to the user equipment, so that the user equipment can collect CSI information that is compatible with the first data collection information and train the user equipment's first CSI information processing model.

[0034] In one implementation, the data collection request information is used to request the network device to instruct the user equipment to collect data.

[0035] By collecting data and requesting information, user equipment can proactively initiate the training of the first CSI information processing model.

[0036] In one embodiment, the data collection request information includes: the frequency of the reference signal, the subcarrier spacing of the reference signal, the bandwidth of the reference signal, the number of ports of the reference signal, the start time of the reference signal transmission, the number of continuous transmissions of the reference signal or the period of the reference signal transmission, and the termination time or condition of the reference signal transmission.

[0037] By collecting data request information, user equipment can notify network equipment to send relevant parameters of the CSI-RS reference signal, which helps network equipment to specifically instruct user equipment to collect CSI information according to the needs of user equipment.

[0038] In one implementation, the first CSI information processing model is configured with a network model ID.

[0039] By using the network model ID, user equipment and network equipment can quickly determine the CSI information processing model required to process CSI information. When training the CSI information processing model, they can also selectively acquire CSI information to train the model.

[0040] In one implementation, the network model ID includes an association ID and / or a dataset ID.

[0041] The association ID corresponds to the features on the network device side where the CSI information processing model resides, while the dataset ID corresponds to the features of the CSI information used to train the CSI information processing model. Through the network model ID, it is possible to effectively distinguish the network devices corresponding to different CSI information processing models, as well as the CSI information used to train those models.

[0042] In one implementation, where the first CSI information processing model includes an actual encoder model, the actual encoder model is configured with an actual encoder model ID.

[0043] By configuring the actual encoder model ID, the actual encoder can be accurately identified on the user equipment side.

[0044] In one implementation, the actual encoder model ID includes an association ID and / or a dataset ID.

[0045] By configuring the association ID and dataset ID, it is possible to distinguish between the network devices used to train the CSI information processing model and the CSI information used to train the CSI information processing model.

[0046] In one implementation, the association ID is used to represent the network device corresponding to the first CSI information processing model, and the dataset ID is used to represent the CSI information used to train the first CSI information processing model.

[0047] By using the association ID and dataset ID, the CSI information processing model can be distinguished, so that when user equipment reports different CSI information to network equipment, the appropriate CSI information processing model can be selected.

[0048] In one implementation, different dataset IDs correspond to different CSI datasets, and the same dataset ID corresponds to similar CSI datasets.

[0049] The same dataset ID corresponds to similar and identical CSI datasets, enabling the same or similar CSI information to be processed using the same CSI information processing model.

[0050] In one implementation, the associated ID includes a static portion and a dynamic portion; the static portion represents basic network identification information, and the dynamic portion represents additional conditions of the network device.

[0051] By associating the static and dynamic parts of the ID, it is possible to distinguish between different network devices, as well as different states of the same network device.

[0052] Secondly, embodiments of this application provide a CSI information processing method, including: receiving a CSI reference signal sent by a network device; and sending CSI information to the network device, wherein the CSI information is processed using a first CSI information processing model provided in any embodiment of this application.

[0053] By using the CSI information processing model trained in the embodiments of this application, CSI information can be processed efficiently when the user equipment reports CSI information to the network device, while ensuring that the error after the CSI information is restored is small.

[0054] Thirdly, embodiments of this application provide a training method for a CSI information processing model, comprising: sending data collection information to a user equipment; receiving CSI information obtained by the user equipment based on the data collection information; and obtaining a second CSI information processing model based on the CSI information.

[0055] In one implementation, the data collection information includes first data collection information and / or second data collection information.

[0056] In one embodiment, when the collected information includes the first information of data collection, after receiving the CSI information obtained by the user equipment based on the data collection information, the method further includes: sending parameters of a second CSI information processing model to the user equipment, the parameters being used by the user equipment to obtain the first CSI information processing model.

[0057] In one implementation, where the collected information includes the first information for data collection, the CSI information is used by the network device to train a second CSI information processing model.

[0058] In one embodiment, the second CSI information processing model includes a second encoder model and a second decoder model.

[0059] In one implementation, the network device sends parameters of a second encoder model to the user equipment.

[0060] In one implementation, the data collection first information is used to indicate: information required for the user equipment to measure CSI and / or the user equipment to measure CSI.

[0061] In one implementation, the data collection first information includes at least one of the following: reported data sample type, reported data sample format, CSI-RS configuration information, transmission bandwidth information, maximum number of CSI data samples collected, data label consistency check mechanism, CSI dataset size, CSI data collection time interval or data collection sampling rate, CSI data encryption technology indication information, CSI data quality threshold information, and the association ID of the second CSI information processing model.

[0062] In one implementation, where the collected information includes the second information for data collection, the user equipment uses the CSI information to train an untrained first CSI information processing model.

[0063] In one embodiment, the training method for the CSI information processing model further includes: receiving the CSI information sent by the user equipment; training an encoder and decoder model pair based on the CSI information; and obtaining the second CSI information processing model based on the encoder and decoder model pair.

[0064] In one implementation, the reference CSI information processing model includes: a reference CSI information encoding model.

[0065] In one embodiment, the second CSI information processing model includes an encoder and decoder model pair; the second CSI information processing model is used by the user equipment to obtain the actual encoder model, and the user equipment trains a nominal decoder model based on the reference CSI information encoding model; the user equipment is also used to train the nominal decoder model based on the CSI information to obtain the actual encoder model.

[0066] In one implementation, the second information for data collection includes at least one of the following: data sample type, reported data sample format, transmission bandwidth information, data tag consistency check mechanism for data collection, maximum number of CSI data collection samples, dataset size, time interval or sampling rate of CSI data collection, CSI-RS configuration information, threshold information for CSI data quality, association ID of the second CSI information processing model, and threshold information for model performance indicators.

[0067] In one embodiment, when the data collection information includes the CSI-RS configuration information, the CSI-RS configuration information includes: the start time of reference signal transmission, the number of continuous transmissions of the reference signal or the period of reference signal transmission, and the end time of reference signal transmission.

[0068] In one embodiment, if the collected information includes the second data collection information, before sending the data collection information to the user equipment, the method further includes: receiving data collection request information sent by the user equipment.

[0069] In one implementation, the data collection request information is used to request the network side to instruct the terminal to collect data.

[0070] In one embodiment, the data collection request information includes: the frequency of the reference signal, the subcarrier spacing of the reference signal, the bandwidth of the reference signal, the number of ports of the reference signal, the start time of the reference signal transmission, the number of continuous transmissions of the reference signal or the period of the reference signal transmission, and the termination time or condition of the reference signal transmission.

[0071] In one implementation, the second CSI information processing model is configured with a network model ID.

[0072] In one implementation, the network model ID includes an association ID and / or a dataset ID.

[0073] In one implementation, when the parameters of the second CSI information processing model are used by the user equipment to obtain the first CSI information processing model, the first CSI information processing model includes an actual encoder model, which is configured with an actual encoder model ID.

[0074] In one implementation, the actual encoder model ID includes an association ID and / or a dataset ID.

[0075] In one implementation, the association ID is used to represent the network device corresponding to the second CSI information processing model, and the dataset ID is used to represent the CSI information used to train the second CSI information processing model.

[0076] In one implementation, different dataset IDs correspond to different CSI datasets, and the same dataset ID corresponds to similar CSI datasets.

[0077] In one implementation, the associated ID includes a static portion and a dynamic portion; the static portion represents basic network identification information, and the dynamic portion represents additional conditions of the network device.

[0078] Fourthly, embodiments of this application provide a CSI information processing method, including: sending a CSI reference signal to a user equipment; receiving CSI information sent by the user equipment; and processing the CSI information using a second CSI information processing model; wherein the second CSI information processing model is obtained using the method provided in any embodiment of this application.

[0079] Fifthly, embodiments of this application provide a communication device that has the functions of implementing the first or second aspect described above. For example, the communication device includes modules or units that perform the operations involved in the first, second, third, and fourth aspects described above. The modules or units can be implemented by software, or by hardware, or by hardware executing corresponding software.

[0080] In one possible design, the communication device includes a processing unit and a communication unit, wherein the communication unit can be used to transmit and receive signals to enable communication between the communication device and other devices; the processing unit can be used to perform some internal operations of the communication device. The functions performed by the processing unit and the communication unit can correspond to the operations involved in the first, second, third, and fourth aspects mentioned above.

[0081] In one possible design, the communication device includes a processor that can be coupled to a memory. The memory can store necessary computer programs or instructions for implementing the functions described in the first, second, third, and fourth aspects above. The processor can execute the computer programs or instructions stored in the memory, causing the communication device to implement the methods in any possible design or implementation of the first, second, third, and fourth aspects above, when the computer programs or instructions are executed.

[0082] In one possible design, the communication device includes a processor and a memory, the memory of which may store necessary computer programs or instructions for implementing the functions described in the first, second, third, and fourth aspects above. The processor may execute the computer programs or instructions stored in the memory, and when the computer programs or instructions are executed, cause the communication device to implement the methods in any possible design or implementation of the first, second, third, and fourth aspects above.

[0083] In one possible design, the communication device includes a processor and an interface circuit, wherein the processor is configured to communicate with other devices via the interface circuit and execute the methods in any possible design or implementation of the first, second, third, and fourth aspects described above.

[0084] Understandably, in the fifth aspect above, the processor can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, integrated circuit, etc.; when implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. Furthermore, there can be one or more processors, and one or more memories. The memory can be integrated with the processor, or the memory and processor can be separate. In specific implementations, the memory can be integrated with the processor on the same chip, or it can be set on different chips. This application does not limit the type of memory or the arrangement of the memory and processor.

[0085] Sixthly, embodiments of this application provide a non-terrestrial network communication system, including a transmitting end device and a receiving end device. The transmitting end device is used to implement the method for network devices provided in any embodiment of this application, and the receiving end device is used to implement the method for user equipment provided in any embodiment of this application.

[0086] In a seventh aspect, an embodiment of this application provides a communication device including a module for performing the methods provided in any embodiment of this application.

[0087] Eighthly, embodiments of this application provide a communication device, including one or more processors configured to perform the methods provided in any embodiment of this application.

[0088] Ninthly, embodiments of this application provide a chip system, including: a memory for storing a computer program; a processor; and when the processor retrieves and runs the computer program from the memory, a communication device equipped with the chip system executes the method provided in any embodiment of this application.

[0089] In a tenth aspect, embodiments of this application also provide a computer program product, the computer program product including instructions that, when executed on a processor, cause the processor to perform the method provided in any embodiment of this application.

[0090] Eleventhly, embodiments of this application provide a terminal device, including: a memory for storing computer programs; a processor; when the processor calls and runs the computer program from the memory, the terminal device executes the method provided in any embodiment of this application.

[0091] In a twelfth aspect, embodiments of this application also provide a computer-readable storage medium storing a computer program or instructions that, when executed by a communication device, implement the method provided in any embodiment of this application.

[0092] The technical effects brought about by the second to twelfth aspects above can be found in the description of the beneficial effects of the corresponding solutions in the first aspect above, and will not be repeated here. Attached Figure Description

[0093] Figure 1 A schematic diagram of the architecture of the communication system used in the embodiments of this application;

[0094] Figure 2 This is a schematic diagram of a method according to an embodiment of this application;

[0095] Figure 3 This is a schematic diagram of the CSI information processing flow according to an embodiment of this application;

[0096] Figure 4 This is a schematic diagram of a method according to another embodiment of this application;

[0097] Figure 5 Another schematic diagram of the method provided in the embodiments of this application;

[0098] Figure 6 This is another schematic diagram of the method provided in the embodiments of this application. Detailed Implementation

[0099] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings. This application will focus on various aspects, embodiments, or features of a system that may include multiple devices, components, modules, etc. It should be understood and appreciated that each system may include additional devices, components, modules, etc., and / or may not include all the devices, components, modules, etc. discussed in conjunction with the accompanying drawings. Furthermore, combinations of these solutions may also be used.

[0100] Furthermore, in the embodiments of this application, words such as "in one possible implementation," "exemplarily," "for example," "e.g.," "as," and "again" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as an "example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of the term "example" is intended to present concepts in a concrete manner. In the embodiments of this application, "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably, and it should be noted that their intended meanings are consistent unless their distinction is emphasized.

[0101] The technical solutions in this application embodiment can be applied to various communication systems, such as Universal Mobile Telecommunications System (UMTS), Wireless Local Area Network (WLAN), Wireless Fidelity (Wi-Fi) system, 4th generation (4G) communication system, such as Long Term Evolution (LTE) system, 5G communication system, such as New Radio (NR) system, and future evolution communication systems, such as 6th generation (6G) mobile communication system, etc.

[0102] In the embodiments of this application, "sending information to...(user equipment or module)" and "sending information to...(user equipment or module)" can be understood as the destination of the information being the user equipment (terminal) or module. This can include sending information directly or indirectly to the user equipment. "Receiving information from...(user equipment or module)" and "receiving information from...(user equipment or module)" can be understood as the source of the information being the user equipment, and can include receiving information directly or indirectly from the user equipment. Information may undergo necessary processing between the source and destination, such as format changes, but the destination can understand the valid information from the source. Similar expressions in this application can be understood in a similar way, and will not be elaborated further here.

[0103] The application scenarios of the embodiments of this application will be described below first.

[0104] Figure 1 This is a schematic diagram of the architecture of the communication system used in the embodiments of this application. Figure 1 As shown, the communication system includes network devices (such as...) Figure 1 110a and 110b, collectively referred to as 110, may also include at least one terminal (such as...). Figure 1 In this application embodiment, 120a-120j are collectively referred to as 120). Figure 1 In the communication system shown, network device 110a has a module capable of implementing radio access network (RAN) functions, and network device 110b can be combined with network device 110a to achieve access to a wireless network, the Internet, or a core network. Network device 100 may also include other devices, such as wireless relay devices and / or wireless backhaul devices. Figure 1 (Not shown in the diagram), wireless relay devices and / or wireless backhaul devices can also be integrated into network device 110. Terminal 120 is connected to network device 110 wirelessly, and network devices 110a and 110b can be connected wirelessly. Different terminals can be interconnected via wired or wireless means.

[0105] In one specific embodiment of this application, network device 110a is a network device that moves relative to the Earth's surface, and network device 110b is a network device that is stationary relative to the Earth's surface.

[0106] At least one of the network devices 110 can also connect to or transmit and receive information with evolved universal terrestrial radio access (E-UTRA), new radio (NR), and future radio access systems or WiFi systems as defined in the 3rd Generation Partnership Project (3GPP). Network device 110 can also connect to devices from two or more of the aforementioned different radio access systems. Network device 110 can also connect to an open radio access network (O-RAN).

[0107] Network device 110 can be used to help terminals access the communication system wirelessly.

[0108] Network device 110a may be configured with a module for implementing base station functions. This module can perform the functions of: a base station, an evolved NodeB (eNodeB or eNB), a transmission reception point (TRP), a next-generation NodeB (gNB) in a 5th generation (5G) mobile communication system, a next-generation base station in a 6th generation (6G) mobile communication system, a base station in a future mobile communication system, or an access node in a WiFi system. The aforementioned base station may include a macro base station, a micro base station, or an indoor station, and may also be a relay node or a donor node. Network device b can cooperate with network device a or independently connect user equipment to the wireless network.

[0109] In another application scenario, multiple wireless access modules can work together to help a terminal achieve wireless access. Different wireless access modules can each implement some functions of the network device 110. For example, a wireless access module can be a central unit (CU), a distributed unit (DU), or a radio unit (RU). The CU can perform the functions of the base station's radio resource control protocol and packet data convergence protocol (PDCP), as well as the service data adaptation protocol (SDAP). The DU performs the functions of the base station's radio link control layer and medium access control (MAC) layer, and can also perform some or all of the physical layer functions. For specific descriptions of these protocol layers, refer to the relevant 3GPP technical specifications. The RU can perform radio frequency signal transmission and reception functions. The CU and DU can be implemented using two independent wireless access modules, or they can be integrated into the same RAN node, such as within a baseband unit (BBU). The RU can be located in radio frequency equipment, such as in a remote radio unit (RRU) or an active antenna unit (AAU). The CU can be further divided into two types: CU-control plane and CU-user plane.

[0110] Terminal 120 can be a device with wireless transceiver capabilities, capable of sending signals to network device 110a, network device 110b, or other devices with signal transceiver capabilities, or receiving signals from network device 110a or network device 110b. In this embodiment, terminal 120 can also be referred to as user equipment (UE), mobile station, mobile terminal, etc. Terminal 120 can be widely used in various scenarios, such as near field communications (NFC) device-to-device (D2D), vehicle-to-everything (V2X) communication, machine-type communication (MTC), Internet of Things (IoT), virtual reality, augmented reality, industrial control, autonomous driving, telemedicine, smart grid, smart furniture, smart office, smart wearables, smart transportation, smart cities, etc. Terminal 120 can be a mobile phone, tablet computer, computer with wireless transceiver function, wearable device, vehicle, airplane, ship, robot, robotic arm, smart home device, air network equipment, ground node, high-altitude base station, etc. The embodiments of this application do not limit the specific technology or device form used in the terminal.

[0111] Communication between network devices and terminals, between network devices 110a and 110b, between terminals and network devices, and between terminals can be conducted using licensed spectrum, unlicensed spectrum, or both simultaneously. Communication can be conducted using spectrum below 6 GHz, spectrum above 6 GHz, or both simultaneously. The embodiments of this application do not limit the spectrum resources used for wireless communication.

[0112] The 5G core network (5G core / new generation core, 5GC / NGC) includes multiple functional units such as access and mobility management function (AMF) network elements, session management function (SMF) network elements, user plane function (UPF) network elements, session function (SF) network elements, authentication server function (AUSF) network elements, policy control function (PCF) network elements, application function (AF) network elements, unified data management (UDM) network elements, and network slice selection function (NSSF) network elements.

[0113] The sensing function (SF) is primarily responsible for sensing control and sensing computation. For example, it can select sensing devices and methods, control sensing services, and process sensing measurement data; it can process sensing measurement data independently. It can also work with the network data analytics function (NWDAF) to achieve intelligent analysis and prediction. The network repository function (NRF) stores the context information of the sensing function, allowing other network elements to discover and select suitable sensing functions through queries.

[0114] Sensing signals are signals used for sensing measurements. Sensing nodes receive sensing signals and perform sensing functions through these measurements. The sensing signals mentioned in this paper include, but are not limited to: positioning reference signal (PRS), sounding reference signal (SRS), channel state information reference signal (CSI-RS), demodulation reference signal (DMRS), phase tracking reference signal (PT-RS), primary synchronization signal (PSS), secondary synchronization signal (SSS), correctly demodulated communication data signals, or dedicated sensing signals, etc., without limitation.

[0115] In the embodiments of this application, the base station function implemented by the network device 110a can also be executed by a module (such as a chip) in the base station, or by a control subsystem containing base station functions. This control subsystem containing base station functions can be a control center in the aforementioned application scenarios such as smart grids, industrial control, intelligent transportation, and smart cities. The terminal function can also be executed by a module (such as a chip or modem) in the terminal, or by a device containing terminal functions.

[0116] In the embodiments of this application, network device 110a can send downlink signals or downlink information to terminal 120 or network device 110b, with the downlink information carried on the downlink channel; terminal 120 can send uplink signals or uplink information to network device 110a or network device 110b, with the uplink information carried on the uplink channel. To communicate with network device 110a, terminal 120 needs to establish a wireless connection in the cell covered by the signal of network device 110a. The cell with which terminal 120 has established a wireless connection can be called the serving cell of the terminal. When terminal 120 communicates with the serving cell, it will also receive signals from neighboring cells.

[0117] The widespread adoption of smartphones and other user devices enables users to access communication systems anytime, anywhere, and use these systems to send and receive data. The extensive use of applications such as social media, video streaming, online games, and e-commerce has significantly increased the demand for data traffic. The widespread adoption of technologies such as high-definition video, streaming media, and live broadcasting has placed higher demands on network bandwidth. High-frequency communication technology and massive MIMO technology provide technical support for the high-speed transmission of large amounts of data. With the development of high-frequency communication and massive MIMO technologies, the scale and dimensions of channel state information feedback have expanded dramatically.

[0118] Meanwhile, artificial intelligence (AI) technology is also one of the fastest-growing technologies in the fields of computer technology and communications. Applying AI technology to CSI transmission, deep learning models can extract effective features from high-dimensional data. AI models can adaptively optimize compression strategies based on channel characteristics. Furthermore, through training, AI models can maintain low CSI reconstruction errors even at high compression rates.

[0119] However, AI-based CSI compression technology also faces several challenges hindering its performance. Firstly, the consistency between training and inference is a significant issue. Due to differences in the working environments and processing flows between the network and user equipment sides, the data characteristics and distribution relied upon during training are difficult to perfectly match the actual inference stage. This results in models that perform well during training failing to accurately process CSI data during real-world inference applications, leading to performance degradation. Secondly, the mismatch in data distribution between the two models severely impacts compression effectiveness. Additional conditions on the network-side dataset, such as TxRU mapping (Transmit and Receive Units), antenna height, and antenna downtilt angle, differ significantly from those on the user equipment side, such as phase normalization and RF characteristics. This inconsistency in data distribution makes it difficult for the decoder model trained on the network side and the encoder model trained on the user equipment side to work collaboratively, leading to data distortion and reduced compression efficiency during CSI compression and recovery, severely limiting the application effectiveness of AI-based CSI compression technology.

[0120] Therefore, embodiments of this application provide a training method for a CSI information processing model, used to train a CSI information processing model to obtain a CSI information processing model capable of more efficiently compressing and / or decompressing CSI information. The training method for the CSI information processing model provided in embodiments of this application includes, as follows: Figure 2 The steps are shown.

[0121] Step S21: The user equipment receives the first information of data collection sent by the network device.

[0122] Correspondingly, the network device sends the first information for data collection to the user equipment.

[0123] The first information for data collection can be one type of information included in the data collection information. The first information for data collection instructs the user equipment to collect first CSI information and send it to the network equipment. The first information for data collection includes at least one of the following: reported data sample type, reported data sample format, CSI reference signal (CSI-RS) configuration information, transmission bandwidth usage information, maximum number of CSI data samples to be collected, data tag consistency check mechanism, CSI dataset size, CSI data collection time interval or data collection sampling rate, CSI data encryption technology indication information, CSI data quality threshold information, and the associated ID of the first CSI information processing model.

[0124] The reported data sample type refers to the type of channel state information or other related data fed back by the user equipment to the network equipment. Reported data sample types can include precoding matrices, channel matrices, etc. The reported data sample format refers to the format of the channel state information or other related data fed back by the user equipment to the network equipment. Reported data sample formats can include scalar quantization or codebook-based quantization, such as the R16 eType-II codebook and the Rel-18 Doppler codebook. The R16 eType-II codebook is the second codebook in Protocol 16. The Rel-18 Doppler codebook was introduced in Protocol 18. A codebook is essentially a predefined set containing a set of possible precoding matrices or beamforming vectors used to optimize signal transmission and reception. CSI-RS configuration information can include multiple parameters that determine the CSI-RS transmission method, resource allocation, and measurement behavior. Transmission occupied bandwidth refers to the actual bandwidth range occupied by the signal in the frequency domain, and can be used to describe the spectral characteristics of the signal. Transmission occupied bandwidth can include: the bandwidth required to adaptively adjust the transmitted information based on the current bandwidth occupancy. The maximum number of CSI data samples collected refers to the maximum number of CSI data samples that can be collected within a specific time period. The data label consistency check mechanism ensures the consistency and accuracy of the reported CSI data and its associated labels (such as timestamps, device IDs, location information, etc.). When a user equipment sends CSI information to a network device, the data label consistency check mechanism ensures that the collected CSI information conforms to the associated ID or additional conditions on the network side. The CSI dataset size refers to the total amount of CSI data stored or processed, typically measured by the number of data points, storage space, or time duration. The CSI data collection interval refers to the time difference between two consecutive CSI data collections, and the CSI data collection sampling rate can refer to the number of times CSI data is collected per unit time. For example, the CSI data collection interval or sampling rate can be used to instruct the terminal to adjust the data interval or sampling rate based on information such as data quality. CSI encryption technology indication information refers to relevant information used to identify or manage the CSI data encryption method. User equipment can use CSI encryption technology indication information to indicate the encryption technology used during data transmission, based on security levels and real-time requirements. CSI data quality threshold information can refer to a set of predefined standards or limit values ​​used to assess and determine whether CSI data meets quality requirements.

[0125] The associated ID is assigned by the network side based on additional conditions. The aforementioned CSI-RS configuration information includes the start time of reference signal transmission, the number of continuous transmissions or the period of reference signal transmission, and the termination time of reference signal transmission. CSI-RS configuration information may include the start time of reference signal transmission, the number of continuous transmissions or the period of reference signal transmission, and the termination time of reference signal transmission, etc.

[0126] Step S22: The user equipment collects first information based on the data to obtain first CSI information.

[0127] In this embodiment, CSI information may include first CSI information and second CSI information. For distinction, the CSI information obtained by the user equipment based on the first data collection information can be considered the first CSI information. The CSI information obtained by the user equipment based on the second data collection information can be considered the second CSI information. CSI information can also be referred to as CSI data. When CSI information is divided into a dataset, it can also be referred to as a CSI dataset. The user equipment can obtain the first CSI information according to the various instructions in the first data collection information regarding how to collect the first CSI information. After receiving the first data collection information, the user equipment receives a reference signal on the specified time and frequency resources according to the base station's resource configuration information. Then, the user equipment processes the received reference signal, calculates channel state information, and after quantizing the calculated channel state information, the user equipment obtains the first CSI information.

[0128] After receiving the initial data collection information, the terminal measures the actual CSI information according to the instructions. The user monitors the quality of the obtained CSI information based on data quality threshold information, comparing the quality indicators of the collected actual CSI information with the corresponding quality thresholds during the data acquisition process.

[0129] The data quality may include at least one of the following: signal strength, signal-to-interference-plus-noise ratio (SINR), positioning information, channel quality indicator (CQI), temporal consistency and spatial correlation, reference signal received power (RSRP), data missing rate, data delay, positioning accuracy, etc.

[0130] The signal strength mentioned above refers to the power level of CSI information received by a user equipment in a wireless communication system. Signal-to-interference-plus-noise ratio (SINR) refers to the ratio of the strength of the useful signal to the total strength of the interfering signal and noise. Location information refers to location-related information within the CSI information. Location accuracy refers to the accuracy of the location position when the location-related information within the CSI information is used for positioning. Channel quality indicator represents the channel quality received by the user equipment. Time consistency refers to the stability and continuity of CSI information over time. Spatial correlation refers to the similarity and correlation of CSI information over space. Reference signal received power is a key indicator used in wireless communication systems to measure the strength of the received signal. Data loss rate refers to the proportion of CSI information lost or not acquired during CSI information acquisition or transmission. Data latency refers to the delay between the acquisition time and the generation time of CSI information during CSI information acquisition.

[0131] By monitoring the data quality of CSI information, we can check in real time whether the data quality of CSI information meets the threshold. Once it is found that the data quality does not meet the requirements, we can immediately take corrective measures or mark it as invalid data to prevent low-quality data from entering the subsequent processing flow, so as to ensure the quality and reliability of the data.

[0132] The aforementioned corrective measures may include re-collecting the corresponding real CSI information, cleaning and correcting the already collected CSI information, or using data augmentation methods to resynthesize CSI information that meets the requirements. Data cleaning, correction, and synthesis of CSI information can be performed on the server at the user's device. The CSI information data collection strategy can be adjusted promptly based on real-time monitoring results. For example, if the data quality of CSI information in a certain area is found to be poor, the sampling rate of data collection can be increased.

[0133] Step S23: The user equipment sends the first CSI information to the network equipment.

[0134] Correspondingly, the network device receives the first CSI information sent by the user equipment.

[0135] The first CSI information may include at least one of the following: channel quality indicator, precoding matrix indicator, and rank indicator. The channel quality indicator reflects the quality of the downlink channel and is used by network equipment to select appropriate modulation and coding schemes. The precoding matrix indicator can be used to indicate the optimal precoding matrix for beamforming to enhance signal strength and reduce interference. The rank indicator indicates the number of spatial layers supported by the channel and is used for multi-user multiple-input multiple-output (MIMO) scheduling.

[0136] The first CSI information is sent to the network device in the form of a dataset. The dataset contains a quantized and encrypted real CSI dataset, authentication information, dataset ID, association ID, and selectively encrypted information depending on whether encryption technology was used during transmission. All user devices report the dataset information to the network device. After receiving the real CSI datasets reported by all user devices, the network device further processes and optimizes the data, and finally blends all the real CSI datasets to obtain the target CSI dataset.

[0137] During the transmission of CSI information datasets, encryption technology can be used to encrypt the CSI information during transmission and to authenticate the sender and receiver. Encryption technology can ensure the security and confidentiality of data during transmission, preventing data theft, tampering, or forgery. Encryption technologies can include at least one of the following: Advanced Encryption Standard (AES), Elliptic Curve Cryptography (ECC), Data Encryption Standard (DES), Triple Data Encryption Standard (3DES), or homomorphic encryption, etc.

[0138] By encrypting the transmission, both user equipment and network equipment can verify the CSI information, ensuring that the source and destination of the CSI information are reliable.

[0139] Step S24: The network device obtains the second CSI information processing model based on the first CSI information.

[0140] Network devices can train an untrained second CSI information processing model based on the first CSI information to obtain the second CSI information processing model.

[0141] In one possible implementation, the user equipment can process the CSI information before sending it to the network device, and a second CSI information processing model can be used to process the CSI information after the network device receives it from the user equipment.

[0142] When the first CSI information is sent as a dataset, the user equipment quantizes and encrypts the actual CSI information. The network device can then dequantize and decrypt the received first CSI information dataset. Simultaneously, the network device can optimize the dequantized and decrypted CSI information to obtain optimized data. Optimization of CSI information can involve merging or removing duplicate or redundant CSI data, further improving data quality and reducing the computational load for subsequent model training. Furthermore, the weights of different datasets can be dynamically adjusted to assess the importance of data under different conditions or scenarios, thereby improving model performance under certain important conditions or scenarios.

[0143] Step S25: The network device sends the encoder model parameters to the user equipment according to the second CSI information processing model.

[0144] The network device can send the encoder model parameters to the user equipment as parameters of the encoder model, using at least a portion of the parameters of the second CSI information processing model.

[0145] The same network device can send encoder model parameters to multiple user devices connected to the network device.

[0146] In one possible implementation, different CSI information processing models can be deployed on the network device side and the user equipment side, with the CSI information processing model deployed on the network device side matching the CSI information processing model deployed on the user equipment side.

[0147] In another possible implementation, the CSI information processing model deployed on the network device side can be deployed locally on the network device or on the server corresponding to the network device. When the network device processes CSI information using the CSI information processing model, it can invoke the CSI information processing model from the server.

[0148] In another possible implementation, the information processing model deployed on the user equipment side can be deployed locally on the user equipment or on the server corresponding to the user equipment. When the user equipment processes CSI information using the CSI information processing model, it can invoke the CSI information processing model from the server.

[0149] Step S26: The user equipment trains the first CSI information processing model on the user equipment side according to the parameters of the encoder model.

[0150] The user equipment (UE) can configure an untrained first CSI information processing model on the UE side based on the parameters of the encoder model, ensuring that the parameters of the configured first CSI information processing model are consistent with those of the second CSI information processing model. The configured first CSI information processing model is then further trained based on the UE side parameters to obtain the UE-side first CSI information processing model.

[0151] The method provided in this application embodiment enables a user equipment to send CSI information to a network device based on the first data collection information. This information is used by the network device to train a second CSI information processing model on the network device side. Since the first data collection information corresponds to the associated ID, the network device can inform the user equipment what kind of CSI information to collect. After the user equipment obtains the corresponding CSI information, the network device uses the corresponding CSI information to train the corresponding second CSI information processing model. Thus, when processing CSI information corresponding to the associated ID and the first data collection information, the corresponding second CSI information processing model and the first CSI information processing model can be used for processing. This effectively compresses, decompresses, and transmits the CSI information while reducing the reconstruction error of the CSI information.

[0152] In a possible implementation, the network device sends different data collection information to the user equipment, which instructs the user equipment to collect different CSI information. This data collection information includes first data collection information and / or second data collection information.

[0153] Network devices send different data collection information to user devices, instructing them to collect CSI information with different characteristics. The network devices then train a CSI information processing model using the CSI information with different characteristics. Subsequently, after the user devices collect CSI information with different characteristics, they process it using the corresponding CSI information processing model and send it to the network devices. The network devices then restore the information using the corresponding CSI information processing model, thus avoiding distortion of the CSI information during the processing and restoration process.

[0154] In one possible implementation, where the collected information includes the first information of data collection, the user equipment can obtain a first CSI information processing model based on the CSI information. Specifically, this may include: the user equipment sending the CSI information to the network device; the user equipment receiving parameters of a second CSI information processing model sent by the network device; and the user equipment obtaining the first CSI information processing model based on the parameters of the second CSI information processing model. The parameters of the second CSI information processing model may include parameters of the encoder model.

[0155] In one possible implementation, the user equipment sends CSI information to the network device, enabling the network device to first train a second CSI information processing model on its side based on the CSI information. Then, the network device shares the parameters of the trained second CSI information processing model with the user equipment that has a network connection to the network device. The user equipment can then set up an untrained first CSI information processing model on its side based on the parameters shared by the network device, and train the first CSI information processing model based on the parameter conditions on its side.

[0156] In one possible implementation, when the data collection information includes the first data collection information, the user equipment uses the CSI information obtained from the first data collection information to train a second CSI information processing model for the network device. When the data collection information includes second data collection information, the user equipment uses the CSI information obtained from the second data collection information to train an untrained first CSI information processing model.

[0157] By collecting first information and second information, the user equipment can collect the CSI information required by the network device to train the second CSI information processing model, as well as the CSI information required by the user equipment to train the first CSI information processing model. Thus, by obtaining CSI information in a targeted manner multiple times, the training of the CSI information processing model can be realized on both the network device side and the user equipment side.

[0158] In one possible implementation, the second CSI information processing model includes a second encoder model and a second decoder model. The second encoder model and the second decoder model constitute an encoder-decoder model pair, in which the encoder model can convert the input data into an intermediate representation (the intermediate representation may be, for example, a vector), and then the decoder model converts the intermediate representation into the target output.

[0159] In one possible implementation, where the second CSI information processing model includes a second encoder model and a second decoder model, the network device sends parameters of the second encoder model to the user equipment. The user equipment configures its own first CSI information processing model based on the parameters of the second encoder model. Further, the first CSI information processing model may include a first encoder model and a first decoder model. After receiving the parameters of the second encoder model, the user equipment can configure the first encoder model based on those parameters, and then train the first decoder model based on the first encoder model.

[0160] In one possible implementation, the first data collection information is used to indicate: the information required by the user equipment (UE) to measure CSI and / or the UE's CSI measurement. The network device, through the first data collection information, can instruct the UE to measure CSI information and indicate to the UE a subset of features of the CSI information that needs to be measured. Thus, a second CSI information processing model trained using the measured CSI information corresponds to the associated ID in the first data collection information.

[0161] In one possible implementation, the association ID is associated with network-side additional conditions of the network device. These network-side additional conditions refer to a set of conditions used to describe and distinguish different network-side characteristics and environmental factors, such as transmit-receive unit (TxRU) mapping, antenna height, antenna downtilt angle, etc.

[0162] The associated ID is used to represent and record network-side additional conditions. The associated ID can be divided into a static part and a dynamic part. The static part contains basic network identification information (such as base station number, frequency band, etc.), while the dynamic part is updated in real time according to changes in network-side additional conditions. The static part of the associated ID is processed using asymmetric encryption; the static part is encrypted with a public key, and the corresponding private key is stored only in the original base station.

[0163] The static portion of the associated ID can use a specific encoding format, such as a string of numbers and letters to represent the base station number, plus a specific frequency band identifier. For example, if the base station number is 001 and the frequency band is 2.4GHz, its static portion can be represented as "001_2.4G". The dynamic portion of the associated ID can be dynamically encoded based on changes in key parameters of additional network conditions. For example, changes in the Transmitter-Receiver Unit (TxRU) mapping generate corresponding dynamic code segments based on different mapping combinations; changes in antenna height and antenna downtilt angle can be converted into corresponding codes through certain numerical conversion rules. Assuming the antenna height changes from 10 meters to 12 meters, the dynamic portion will update the encoding of this change information accordingly based on a preset height encoding rule. These dynamic codes can be combined with the static portion in a certain order to form a complete associated ID.

[0164] In one possible implementation, where the collected information includes the second information of data collection, obtaining the first CSI information processing model based on the CSI information includes: training a reference CSI information processing model based on the CSI information to obtain the first CSI information processing model.

[0165] In one possible implementation, before training a reference CSI information processing model based on the CSI information to obtain the first CSI information processing model, the method further includes: sending the CSI information to the network device; receiving parameters of the CSI information processing model sent by the network device; obtaining a reference CSI information processing model based on the parameters; and obtaining the first CSI information processing model based on the CSI information and the reference CSI information processing model.

[0166] On the network device side, after the encoder and decoder models are trained, the network device can process the encoder parameters using parameter encryption technology and set up access control mechanisms for the processed encoder model parameters. When security vulnerabilities or threats are discovered, encryption strength can be strengthened and access permissions adjusted in a timely manner to ensure that the risk of information leakage remains within a controllable range and adapts to the ever-changing security environment.

[0167] The aforementioned parameter encryption technology can be homomorphic encryption, attribute encryption, or other similar techniques. By encrypting the transmission of encoder model parameters, the security and privacy of the parameters are ensured. The access control mechanism described above can assign different permissions to different user devices, ensuring that only authorized and trusted user device manufacturers or partners can access the shared encoder parameter information, preventing unauthorized access and information leakage.

[0168] After the network device transmits the encoder model parameters to the user equipment, the user equipment receives the encoder parameter information, performs access permission verification and corresponding decryption, configures the reference CSI information encoder model on the UE side according to the encoder model parameters, and stores the reference CSI information encoder model parameters and their corresponding network model ID in the local model library on the UE side.

[0169] The encoder model parameters include encrypted encoder parameters, the corresponding network model ID, model performance threshold information, target CSI subsets, access control information, and encryption-related information. The aforementioned target CSI subsets can refer to the target CSI dataset being divided into multiple subsets, each containing a portion of the target CSI data samples. These subsets are transmitted to multiple terminals. After receiving the encoder parameter information, the terminals upload the target CSI subsets and the network model ID to the server on the user equipment side. The UE-side server combines target CSI subsets with the same network model ID into a target CSI dataset. The user equipment can configure the encoder model using the encoder parameter information, and the structures of the encoder model on the network side and the encoder model on the user equipment side can be consistent or standardized.

[0170] The user equipment can locally configure a model library, which can store reference CSI information encoding model parameters and nominal decoder model parameters for different network model IDs, as well as actual encoder model parameters for different actual encoder model IDs. The actual encoder model ID can refer to the ID of the actual encoder model, and the actual encoder model can also be called the actual CSI information encoder model.

[0171] In one possible implementation, the reference CSI information processing model includes: a reference CSI information encoding model.

[0172] In one possible implementation, the reference CSI information encoding model can have the same structure as the second encoder model on the network device side. In this case, the network device can send the parameters of the second encoder model to the user equipment, and the user equipment can set the encoder model with the same structure according to the received parameters to obtain the reference CSI information encoding model.

[0173] In one possible implementation, the first CSI information processing model includes an actual encoder model. Obtaining the first CSI information processing model based on the CSI information and the reference CSI information processing model includes: training a nominal decoder model based on the reference CSI information encoding model; and training the nominal decoder model based on the CSI information to obtain the actual encoder model.

[0174] In one possible implementation, the second information for data collection includes at least one of the following: data sample type, reported data sample format, transmission bandwidth information, data tag consistency check mechanism for data collection, maximum number of CSI data collection samples, dataset size, time interval or sampling rate of CSI data collection, CSI-RS configuration information, threshold information for CSI data quality, association ID of the first CSI information processing model, and threshold information for model performance indicators.

[0175] To store CSI information datasets, a data management server can be established on the network side to store the target CSI dataset along with its corresponding association IDs and dataset IDs. This server can also handle the further processing and optimization of the real CSI datasets mentioned above. The data management server can process all real CSI datasets reported to the network side and store the target CSI dataset. Data processing includes data decryption and dequantization, merging or removing duplicate or redundant parts in mixed data, data stratification, and data sampling. Data storage involves storing the target CSI dataset along with its corresponding association IDs and dataset IDs, and creating indexes for the association IDs and dataset IDs to quickly query and match the corresponding target CSI dataset.

[0176] The network device trains corresponding encoder AI / machine learning (ML) models and decoder AI / ML models based on the target CSI dataset. It then generates network model IDs based on the associated IDs and dataset IDs, and stores the corresponding network model IDs along with their corresponding encoder and decoder parameter models in a local model library. In a possible implementation, the encoder model is the generative part of the CSI compression model, and the decoder model is the reconstructed part of the CSI compression model.

[0177] The aforementioned network model ID is used to distinguish encoder-decoder model pairs from different network sides. Its form is Network Model ID = {Association ID; Dataset ID-1,..., Dataset ID-M}, where M is an integer representing the number of different terminal-side additional conditions. The local model library on the network device side is used to store encoder and decoder model parameters for different network model IDs.

[0178] In one possible implementation, when the first information for data collection includes the CSI-RS configuration information, the CSI-RS configuration information includes: the start time of reference signal transmission, the number of continuous transmissions of the reference signal or the period of reference signal transmission, and the end time of reference signal transmission.

[0179] In one possible implementation, if the collected information includes the second data collection information, before receiving the data collection information sent by the network device, the method further includes: sending data collection request information to the network device.

[0180] After configuring the reference CSI information coding model, the user equipment can send a data collection request to the network device to request the network to send data collection-related information. After receiving the data collection request, the network sends the second data collection information to the UE.

[0181] The data collection request information is used to request the network side to instruct the terminal to collect data, including the frequency of the reference signal, the subcarrier spacing of the reference signal, the bandwidth of the reference signal, the number of ports of the reference signal, the start time of the reference signal transmission, the number of continuous transmissions of the reference signal or the period of the reference signal transmission, and the timing or conditions for the termination of the reference signal transmission.

[0182] After receiving the second data collection information, the user equipment collects data according to the instructions in the information and performs quality checks on the corresponding data. If the dataset after data checks does not meet the set size of the CSI dataset, the relevant information for data collection can be adjusted according to the actual situation.

[0183] After obtaining CSI information, the user equipment can perform data quality checks on the CSI information. Data quality checks can be used to inspect the quality of CSI information, that is, to check in real time whether the data quality meets the thresholds set by the network side. Once data quality is found to be unsatisfactory, corrective measures are immediately taken or the data is marked as invalid to prevent low-quality data from entering subsequent processing, thereby ensuring data quality and reliability.

[0184] After data collection by the user equipment is completed, a dataset ID is assigned according to additional conditions on the user equipment side. The network model ID corresponding to the reference CSI information encoding model in the UE's local model library is then matched based on the association ID and dataset ID. Matching the association ID and dataset ID with the network model ID means determining whether the network model ID contains the association ID and dataset ID. If they are contained, the match is considered successful; otherwise, the match is considered unsuccessful.

[0185] If a match is successful, the UE activates the corresponding UE-side reference CSI information encoding model. After the terminal-side reference CSI information encoding model is successfully activated, the UE-side nominal decoder model is trained based on the reference CSI information encoding model and the target CSI dataset stored on the UE-side server. Then, based on the nominal decoder model and the collected real CSI dataset, an actual encoder model that can better match the additional conditions on the user equipment side is trained.

[0186] The nominal decoder model can be viewed as the network-side decoder model encoded as a reference CSI information or proxy model within the UE. The success of the UE-side actual encoder model's training is determined by calculating the performance metrics of the UE-side actual encoder model and the UE-side nominal decoder model.

[0187] For example, the model's performance metrics include at least one or more of the following: squared generalized cosine similarity (SGSC) between the real CSI and the recovered CSI, generalized cosine similarity (GSC) between the real CSI and the recovered CSI, normalized mean squared error (NMSE) between the real CSI and the recovered CSI, mean squared error (MSE) between the real CSI and the recovered CSI, structural similarity index (SSIM) between the real CSI and the recovered CSI, number of floating-point operations (FLOPs), model runtime, and compression ratio. When the set model performance metrics are met, the actual encoder model on the UE side is considered to have completed training. When the set model performance metrics are not met, the actual encoder model on the UE side is considered to need incremental training, which involves collecting more real CSI data and retraining to update the actual encoder model and improve its performance.

[0188] The incremental training described above refers to gradually incorporating new data into the training process as it continuously flows in. Each time new data is added, the model is trained, and the training results are merged with the previous model parameters to update the model. This method not only ensures that model performance does not degrade but also helps improve training efficiency and model adaptability. It enables the model to better adapt to new data distributions and features, enhancing its generalization ability, while also saving computational resources, especially when dealing with large amounts of data.

[0189] Once the actual encoder model on the UE side has been trained, the actual encoder model ID (composed of the association ID and the dataset ID), the corresponding actual encoder parameters, the network model ID, and the corresponding nominal decoder parameters are stored in the local model library. The actual encoder model ID = {association ID; dataset ID - k}, where k is a dataset ID related to additional conditions on the terminal side.

[0190] In the event of a match failure, the statistical characteristics of the collected real CSI information dataset are calculated. The statistical characteristics of this dataset are then compared with the statistical characteristics of various real CSI information datasets stored in the UE-side server, and corresponding indicators are calculated. These indicators include similarity metrics such as mean difference, variance ratio, and the goodness of fit of the probability distribution; these metrics quantify the degree of similarity between the two datasets.

[0191] When the calculated index reaches a pre-set threshold, the dataset ID of the collected real CSI information dataset can be associated with the dataset ID of similar datasets. From the perspective of data characteristics, they are similar enough to be regarded as having equal status in subsequent processing, which can reduce the number of model updates or retraining.

[0192] Associating two different dataset IDs means that when the similarity index of different dataset ID-1 and associated ID-2 meets a threshold, if dataset ID-1 matches a specific model ID, then dataset ID-2 can also be considered to match that model ID. Even if dataset ID-2 is not explicitly listed in the initially defined model IDs, this association based on data feature similarity allows dataset ID-2 to be included in the category of datasets that match the model ID, ensuring consistency in their identification hierarchy across the entire system.

[0193] The nominal decoder model and the actual encoder model of the corresponding network model ID can be activated using the dataset ID and association ID of similar datasets. The performance index between the actual encoder model and the nominal decoder model on the UE side is calculated to determine whether the actual encoder model of the similar dataset ID is suitable. If the set model performance index is met, the actual encoder model on the UE side is considered to meet the requirements and does not need further training. If the set model performance index is not met, the actual encoder model on the UE side is considered to need incremental training. This is done by inputting the current real CSI data and retraining to update the actual encoder model to adapt to the current real CSI data, improve model performance, and update the corresponding actual encoder model parameters stored in the local model library.

[0194] If the calculated metrics do not meet the set threshold, this new dataset information is reported to the network side. The network side updates the target CSI dataset and uses the updated or newly added data from the target CSI dataset for incremental training of the network-side encoder and decoder pair, thereby updating the network-side model and the network model ID. Then, the updated reference CSI information encoding model parameters and the network model ID are sent to the UE. Upon receiving this, the UE updates the network model ID and the corresponding reference CSI information encoding model parameters, storing the updated network model ID and reference CSI information encoding model parameters in its local model library. Simultaneously, the corresponding reference CSI information encoding model is activated, and the UE-side nominal decoder model is trained based on the reference CSI information encoding model and the target CSI dataset stored on the UE-side server. Then, based on the nominal decoder model and the collected real CSI dataset, an actual encoder model that better matches the additional conditions on the user equipment side is trained.

[0195] The success of the UE-side actual encoder model training is determined by calculating the performance metric between the real CSI input to the UE-side actual encoder model and the recovered CSI information output by the UE-side nominal decoder model. If the set model performance metric is met, the UE-side actual encoder model is considered to have completed training. If the set model performance metric is not met, the UE-side actual encoder model is considered to require incremental training. This involves collecting more real CSI data and retraining to update the actual encoder model and improve its performance. After the UE-side actual encoder model training is complete, the actual encoder model ID (composed of the association ID and dataset ID), the corresponding actual encoder parameters, the network model ID, and the corresponding nominal decoder parameters are stored in the local model library.

[0196] In one possible implementation, the data collection request information is used to request the network device to instruct the user equipment to collect data.

[0197] In one possible implementation, the data collection request information includes: the frequency of the reference signal, the subcarrier spacing of the reference signal, the bandwidth of the reference signal, the number of ports of the reference signal, the start time of the reference signal transmission, the number of continuous transmissions of the reference signal or the period of the reference signal transmission, and the termination time or condition of the reference signal transmission.

[0198] In one possible implementation, the first CSI information processing model is configured with a network model ID.

[0199] In one possible implementation, the network model ID includes an association ID and / or a dataset ID.

[0200] For example, the dataset ID is associated with the statistical characteristics of CSI information. In one possible implementation, all real CSI information that meets the threshold requirements constitutes a dataset of real CSI information, and the statistical characteristics of this dataset are calculated to understand its distribution characteristics. These statistical characteristics include the dataset's mean, variance, standard deviation, probability distribution, and the distribution pattern of outliers.

[0201] Each user equipment (UE) assigns a dataset ID to the real CSI dataset based on its own terminal-side additional conditions. The terminal can assign a dataset ID according to its own conditions and store the dataset ID, association ID, and the statistical characteristics of the corresponding real CSI dataset in the server on the UE side. The terminal or the server calculates the similarity index of the statistical characteristics between real CSI datasets with different dataset IDs under the same association ID. If the similarity index of the statistical characteristics of the real CSI datasets corresponding to two different dataset IDs meets the threshold, it can be determined that the two different datasets are similar in features. This allows the different dataset IDs of two similar datasets to be associated with each other and can be replaced by each other.

[0202] The additional conditions on the terminal side refer to the conditions that describe and distinguish different terminal characteristics and environmental factors, such as the SVD phase normalization method, UE antenna virtualization, antenna imbalance, antenna spacing / layout, etc. The dataset ID is used to represent and record the additional conditions on the terminal side when the terminal collects the dataset. The dataset ID allocation method for each terminal can be a unified or standardized mechanism. The terminal-side server is mainly used for storing the statistical characteristics of the real CSI dataset, storing the target CSI dataset, storing ID information, and offline training of the UE-side model. Similarity indicators can be indicators such as mean difference, variance ratio, and the goodness of fit of the probability distribution. These indicators can quantitatively reflect the degree of similarity between two datasets.

[0203] Associating two different dataset IDs means that when the similarity index of different dataset ID1 and associated ID2 meets the threshold, although they may be different at the system's identification level at first, they can be regarded as the same dataset from the perspective of data characteristics. By associating them, they are made the same at the system's identification level. Therefore, at this time, dataset ID-1 and dataset ID-2 represent the same dataset ID.

[0204] In one possible implementation, where the first CSI information processing model includes an actual encoder model, the actual encoder model is configured with an actual encoder model ID.

[0205] In one possible implementation, the actual encoder model ID includes an association ID and / or a dataset ID.

[0206] In one possible implementation, the association ID is used to represent the network device corresponding to the first CSI information processing model, and the dataset ID is used to represent the CSI information used to train the first CSI information processing model.

[0207] In one possible implementation, different dataset IDs correspond to different CSI datasets, and the same dataset ID corresponds to similar CSI datasets.

[0208] In one possible implementation, the associated ID includes a static part and a dynamic part; the static part represents basic network identification information, and the dynamic part represents additional conditions of the network device.

[0209] For example, once the model on the UE side achieves good performance indicators, the CSI compression function can be enabled. The NW sends compression information to instruct the UE to collect the required actual CSI information. The UE assigns a dataset ID and collects the corresponding CSI information according to the instructions in the compression information. The UE performs quality checks on the corresponding CSI information and processes the CSI information that does not meet the quality threshold to improve the data quality.

[0210] The compression information includes the CSI data acquisition time point, transmission bandwidth usage, number of CSI data samples, termination conditions for CSI data acquisition, model performance index thresholds, data quality thresholds, and association IDs. Related processing can involve using automated algorithms to correct the relevant data or using techniques such as Generative Adversarial Networks (GANs) or Variational Autoencoders (VAEs) to generate data similar to the original data, thereby improving data quality. The association ID and dataset ID can be paired with the actual encoder model ID to activate the corresponding UE-side actual encoder model. The UE-side actual encoder model is then used to compress the CSI data, forming CSI feedback information, which is reported to the network side. The CSI feedback information includes compressed CSI data, association IDs, dataset IDs, etc. Upon receiving the CSI feedback information, the network side uses the association ID and dataset ID to activate the decoder model in its local model library. The decoder model is then used to obtain the recovered CSI, thus completing the CSI compression process.

[0211] This application also provides a CSI information processing method, such as... Figure 4 As shown, it includes the following steps S41 to S442.

[0212] Step S41: The user equipment receives the CSI reference signal sent by the network device.

[0213] In step S41, when the user equipment receives the CSI reference signal sent by the network device, it also receives the association ID of the network device. The association ID is used to match the first CSI information processing model on the user equipment side.

[0214] Step S42: The user equipment processes the CSI information using the first CSI information processing model.

[0215] When a network device sends a CSI reference signal to a user equipment (UE), it can also send a dataset ID. The UE can then match the association ID and dataset ID with the network model ID on the server side corresponding to the UE to determine the corresponding first CSI information processing model.

[0216] Alternatively, when a network device sends a CSI reference signal to a user equipment, it may not send the dataset ID. After the user equipment obtains the CSI information, it determines the dataset ID based on the actual CSI information obtained. Then, based on the dataset ID and the associated ID, it matches the network model ID to determine the corresponding first CSI information processing model.

[0217] For example, the first CSI information processing model may include a first encoder model. When the user equipment matches the network model ID according to the associated ID and the match is successful, the user equipment activates (calls) the first encoder model.

[0218] Step S43: The user equipment sends the processed CSI information to the network equipment.

[0219] The user equipment (UE) encodes the CSI information using a first CSI information processing model corresponding to the association ID and dataset ID, and then sends the encoded CSI information to the network device. When sending the CSI information, the UE also sends the association ID and dataset ID to the network device, which are used by the network device to match the second CSI information processing model.

[0220] Step S44: The network device processes the CSI information using the second CSI information processing model.

[0221] The network device matches the network model ID with the association ID and dataset ID to determine the corresponding second CSI information processing model. Then, it uses the determined corresponding second CSI information processing model to decode the received CSI information, and then performs post-processing and reconstruction on the CSI information to obtain the recovered CSI information.

[0222] Generally, associated ID designs only consider some network-side factors or lack systematic design. The associated ID in this application is divided into static and dynamic parts. The static part contains basic network identification information (such as base station number, frequency band, etc.) and uses asymmetric encryption. The dynamic part is updated in real time according to additional network-side conditions (such as TxRU mapping, antenna height changes, etc.). This improvement solves the security problem of network-side proprietary information during sharing, preventing information leakage. Furthermore, it enables the UE to more accurately identify changes in additional network-side conditions, avoiding model training and inference biases caused by inaccurate network environment information. This improvement enhances the security and accuracy of data transmission and model training, ensuring the stable operation of the communication system and reducing the probability of communication failures due to inaccurate network environment information. The asymmetric encryption of the static part ensures the security of network identification information, preventing unauthorized acquisition and tampering. The dynamic part reflects changes in additional network-side conditions in real time, allowing the UE to obtain the latest network environment information, enabling the model trained based on this to better adapt to network changes, thereby improving the security and accuracy of data transmission and model training, and ensuring the stability of the communication system.

[0223] In typical model training, the similarity between datasets is rarely used to optimize model training; different datasets are usually processed independently. This application's embodiment calculates similarity indices based on the statistical characteristics of datasets, such as mean difference, variance ratio, and probability distribution fit, to associate similar dataset IDs, making them identical at the system identification level and thereby reducing the number of model updates or retraining sessions. This improvement alleviates the waste of computational resources caused by frequent model updates or retraining, while also improving the model's adaptability to different datasets and avoiding performance fluctuations due to dataset differences. This improvement saves significant computational resources, improves model training efficiency, and maintains stable performance even in complex and changing data environments. Because the similarity index quantifies the degree of similarity between datasets, when different datasets are determined to be similar, they can be treated as the same dataset during model training, reducing redundant training. This allows the model to utilize existing training results more efficiently, avoiding redundant computation on similar data, thereby saving computational resources, improving training efficiency, and ensuring stable model performance across different datasets.

[0224] Generally, data collection information only includes basic measurement information. In this embodiment, the first data collection information not only covers CSI-RS configuration information, data encryption technology indication information, data quality threshold information, and associated IDs, but also adds model performance index threshold information to the second data collection information. Simultaneously, this invention performs quality monitoring and statistical characteristic calculations on the collected data and associates it with similar dataset IDs. This improvement solves the problems of inaccurate and incomplete data collection, and the impact of low-quality data on model training performance. It also utilizes data feature similarity to reduce the number of model updates or retraining attempts, improving data utilization efficiency. This improvement makes terminal CSI measurement data more accurate and efficient, reduces measurement errors and uncertainties, improves data quality and reliability, saves computing resources, and improves model training efficiency. Because the rich data collection information provides more precise guidance for terminal measurements, it helps to obtain more accurate CSI data. The data quality monitoring mechanism checks and processes low-quality data in real time to prevent it from interfering with model training.

[0225] Generally, model training methods are relatively simple, lacking unified management and flexible adjustment mechanisms. In this embodiment, the network side trains an encoder-decoder model pair based on the target CSI dataset and then shares the encoder parameters with the UE. After configuring the reference CSI information encoding model, the UE matches the network model ID with the association ID and dataset ID for further training. There is a corresponding handling mechanism for matching failures, and multiple performance indicators are used to judge the training effectiveness and perform incremental training. This improvement solves the problems of low collaborative training efficiency between the network and UE sides, poor model training results, and difficulty in adapting to different devices and environments. This improvement improves the accuracy and efficiency of model training, enhances the model's adaptability to different devices and environments, and ensures stable system operation in complex scenarios. Because the network and UE sides collaborate on training based on shared encoder parameters, the model can combine the advantages of both sides for optimization. The matching mechanism and incremental training strategy allow the model to adjust its training direction and parameters in a timely manner according to different device and environmental conditions, adapting to new data distributions and features. Multiple performance indicators comprehensively quantify model performance, providing objective and accurate evidence for training, thereby improving training accuracy and efficiency, enhancing model adaptability, and ensuring stable system operation.

[0226] Existing technologies lack real-time monitoring and effective processing methods for data quality, making it difficult to dynamically adjust collection strategies based on data quality. This application's embodiments establish a comprehensive data quality monitoring system on the terminal side. This system not only checks data quality in real time but also takes timely corrective measures such as re-collection, cleaning and error correction, or data augmentation based on the quality status, and dynamically adjusts the data collection sampling rate. This improvement solves the problem of poor model training results due to unstable data quality, preventing low-quality data from entering subsequent processes and affecting model accuracy and reliability. This improvement significantly enhances the data quality input to the model, ensuring the accuracy and stability of model training, thereby improving the accuracy of CSI compression and recovery, and enhancing the overall performance of the communication system. Real-time monitoring can promptly detect low-quality data, and timely corrective measures can remove or repair bad data. Dynamically adjusting the sampling rate optimizes data collection according to actual conditions, making the collected data more representative and reliable. High-quality data provides a solid foundation for model training, enabling the model to learn data features more accurately, thereby improving the accuracy of CSI compression and recovery and enhancing the performance of the communication system.

[0227] Existing technologies lack precise model matching mechanisms, making it difficult to select the optimal model for training and application under different network and user device conditions. This application's embodiments utilize association IDs and dataset IDs for model matching. On the UE side, network model IDs are matched based on association IDs and dataset IDs. If a match is successful, the corresponding model is directly trained. If a match fails, a dataset similarity index is calculated, and the model is activated or a network-side model update is triggered using similar dataset IDs and association IDs, achieving precise model selection and efficient training. This improvement solves the problems of blind model selection and low training efficiency, avoiding ineffective training on mismatched models and improving the accuracy and efficiency of model training. This improvement reduces model training time and computational resource consumption, improves model adaptability to different network and device conditions, and enhances the accuracy of CSI compression and recovery. Because association IDs and dataset IDs contain rich network and device condition information, matching association IDs and dataset IDs can quickly locate the model most suitable for the current conditions. The similarity index calculation and processing mechanism for matching failures fully utilizes the similarity between data, avoiding the resource waste caused by retraining the model, thereby improving training efficiency and model adaptability, and enhancing the accuracy of CSI compression and recovery.

[0228] Figure 5 An example of a method implemented on the network device side is shown, including the following steps S51 to S58.

[0229] Step S51: Assign an associated ID to the network device.

[0230] Network devices can be assigned associated IDs based on their own conditions.

[0231] Step S52: The network device sends the first information for data collection to the user equipment based on the associated ID.

[0232] Step S53: The user equipment measures the real CSI information and associates the real CSI information with the dataset ID.

[0233] User equipment performs data quality monitoring on the measured real CSI information. When associating real CSI information with a dataset ID, it can also calculate the statistical characteristics of the real CSI information and associate the dataset ID based on the calculated statistical characteristics.

[0234] Step S54: The network device receives CSI information sent by the user equipment.

[0235] Step S55: The network device processes and optimizes the received CSI information to obtain the target CSI information.

[0236] Step S56: The network device uses the target CSI information to train the second CSI information processing model and forms the network model ID based on the association ID and the dataset ID.

[0237] The second CSI information processing model may include a second encoder model and a second decoder model.

[0238] Step S57: The network device sends the parameters of the second information processing model to the user equipment.

[0239] Step S58: The user equipment uses the parameters of the second CSI information processing model to obtain the first CSI information processing model and stores the corresponding network model ID.

[0240] The user equipment can activate (set) its encoder model using the parameters of the second encoder model in the second CSI information processing model to obtain a reference CSI information encoding model. Then, using the reference CSI information encoding model, a nominal decoder model is obtained. Finally, the nominal decoder model is trained using CSI information to obtain the actual encoder model.

[0241] Figure 6 This application illustrates, in one example, the process by which a user equipment obtains a first CSI information processing model based on parameters of a second CSI information processing model.

[0242] Step S61: The user equipment sets the encoder model of the user equipment according to the parameters of the second CSI information processing model to obtain the reference encoder model.

[0243] The reference encoder model is equivalent to the reference CSI information encoding model in the aforementioned embodiments.

[0244] Step S62: The user equipment sends a data collection request to the network device.

[0245] Step S63: The network device sends data collection second information to the user equipment.

[0246] Step S64: Obtain CSI information and match the network model ID based on the association ID in the second information of data collection and the dataset ID of the obtained CSI information.

[0247] Step S65: If the match is successful, use the obtained CSI information and the reference encoder model to obtain the first CSI information processing model.

[0248] Step S66: In the event of a failed match, calculate the similarity index of the statistical characteristics of the obtained CSI information.

[0249] Step S67: Match the dataset ID and association ID based on the similarity index of statistical characteristics, and obtain the first CSI information processing model based on the matched reference encoder model.

[0250] Step S68: If the similarity index cannot match the dataset ID and the associated ID, the user equipment sends CSI information to the network equipment.

[0251] Step S69: The user equipment receives the updated parameters of the second CSI information processing model sent by the network device after updating the second CSI information processing model according to the CSI information.

[0252] Step S610: The user equipment sets an updated reference encoder model based on the updated parameters, and obtains a first CSI information processing model based on the updated reference encoder model.

[0253] Meanwhile, the communication device provided in this application embodiment is also used to implement Figures 2 to 6 The methods and their corresponding implementations are described in the text.

[0254] In another embodiment, a communication method is provided, which is applied to a communication system including network devices and user equipment. The communication method may include, for example: Figures 2 to 6 The embodiments and corresponding examples are shown.

[0255] It is understood that, in order to implement the functions in the above embodiments, the base station and user equipment include hardware structures and / or software modules corresponding to perform each function. Those skilled in the art should readily recognize that, based on the units and method steps described in conjunction with the embodiments disclosed in this application, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application scenario and design constraints of the technical solution.

[0256] The communication device provided in this application can be used to implement the functions of the user equipment or terminal in the methods provided in the above-described embodiments of this application, and therefore can also achieve the beneficial effects of the above-described method embodiments. In the embodiments of this application, the communication device can be the final terminal device.

[0257] In one implementation, when the communication device is used to achieve Figure 2 In the relevant embodiments of the method shown, the function of the user equipment is as follows: the transceiver unit is used to receive data collection information sent by the network device.

[0258] The transceiver unit is also used to: collect information based on the data and obtain CSI information.

[0259] The processing unit is used to: obtain a first CSI information processing model based on the CSI information.

[0260] The data collection information includes first data collection information and / or second data collection information.

[0261] Wherein, when the collected information includes the first information of data collection, the processing unit is further configured to: send the CSI information to the network device; receive parameters of the second CSI information processing model sent by the network device; and obtain the first CSI information processing model according to the parameters.

[0262] Wherein, when the data collection information includes the first data collection information, the CSI information is used by the network device to train a second CSI information processing model.

[0263] The second CSI information processing model includes a second encoder model and a second decoder model.

[0264] The network device sends parameters of the second encoder model to the user equipment.

[0265] The data collection first information is used to indicate: the characteristics of the CSI information measured by the user equipment and / or the CSI information measured by the user equipment.

[0266] The data collection first information includes at least one of the following: reported data sample type, reported data sample format, CSI-RS configuration information, transmission bandwidth information, maximum number of CSI data samples collected, data label consistency check mechanism, CSI dataset size, CSI data collection time interval or data collection sampling rate, CSI data encryption technology indication information, CSI data quality threshold information, and the association ID of the first CSI information processing model.

[0267] Wherein, when the collected information includes the second information of data collection, the processing unit is further configured to: train a reference CSI information processing model based on the CSI information to obtain the first CSI information processing model.

[0268] The transceiver unit is further configured to: send the CSI information to the network device; and receive parameters of the CSI information processing model sent by the network device. The processing unit is further configured to: obtain a reference CSI information processing model based on the parameters; and obtain the first CSI information processing model based on the CSI information and the reference CSI information processing model.

[0269] The reference CSI information processing model includes: a reference CSI information encoding model.

[0270] The first CSI information processing model includes an actual encoder model, and the processing unit is further configured to: train a nominal decoder model based on the reference CSI information encoding model; and train the nominal decoder model based on the CSI information to obtain the actual encoder model.

[0271] The second information for data collection includes at least one of the following: data sample type, reported data sample format, transmission bandwidth information, data tag consistency check mechanism for data collection, maximum number of CSI data collection samples, dataset size, time interval or sampling rate of CSI data collection, CSI-RS configuration information, threshold information for CSI data quality, association ID of the first CSI information processing model, and threshold information for model performance indicators.

[0272] Wherein, when the data collection information includes the CSI-RS configuration information, the CSI-RS configuration information includes: the start time of the reference signal transmission, the number of continuous transmissions of the reference signal or the period of the reference signal transmission, and the end time of the reference signal transmission.

[0273] Wherein, if the collected information includes the second information for data collection, the transceiver unit is further configured to: send data collection request information to the network device.

[0274] The data collection request information is used to request the network device to instruct the user equipment to collect data.

[0275] The data collection request information includes: the frequency of the reference signal, the subcarrier spacing of the reference signal, the bandwidth of the reference signal, the number of ports of the reference signal, the start time of the reference signal transmission, the number of continuous transmissions of the reference signal or the period of the reference signal transmission, and the timing or condition for the termination of the reference signal transmission.

[0276] The first CSI information processing model is configured with a network model ID.

[0277] The network model ID includes an association ID and / or a dataset ID.

[0278] Wherein, if the first CSI information processing model includes an actual encoder model, the actual encoder model is configured with an actual encoder model ID.

[0279] The actual encoder model ID includes the association ID and / or dataset ID.

[0280] The association ID is used to represent the network device corresponding to the first CSI information processing model, and the dataset ID is used to represent the CSI information used to train the first CSI information processing model.

[0281] Different dataset IDs correspond to different CSI datasets, and the same dataset ID corresponds to similar CSI datasets.

[0282] The associated ID includes a static part and a dynamic part; the static part represents basic network identification information, and the dynamic part represents additional conditions of the network device.

[0283] This application also provides a communication device, including a transceiver unit and a processing unit. The transceiver unit is configured to: receive a CSI reference signal sent by a network device; and send CSI information to the network device, wherein the CSI information is processed using a first CSI information processing model as provided in any embodiment of this application.

[0284] This application also provides a communication device, including a transceiver unit and a processing unit. The transceiver unit is configured to: send data collection information to a user equipment; and receive CSI information obtained by the user equipment based on the data collection information. The processing unit is configured to: obtain a second CSI information processing model based on the CSI information.

[0285] This application also provides a communication device, including a transceiver unit and a processing unit. The transceiver unit is configured to: transmit a CSI reference signal to a user equipment; and receive CSI information transmitted by the user equipment. The processing unit is configured to: process the CSI information using a second CSI information processing model; the second CSI information processing model is obtained using the method provided in any embodiment of this application.

[0286] For a more detailed description of the aforementioned processing unit and transceiver unit, please refer to [link / reference]. Figure 2 The method embodiments shown and other related embodiments are described.

[0287] In one embodiment, the communication device includes a processor and interface circuitry. The processor and interface circuitry are coupled to each other. It is understood that the interface circuitry can be a transceiver or an input / output interface. Optionally, the communication device may further include a memory for storing instructions executed by the processor, or storing input data required for the processor to execute instructions, or storing data generated after the processor executes instructions.

[0288] When the communication device is used to achieve Figure 2 In the method shown, the processor is used to implement the functions of the above-mentioned processing unit, and the interface circuit is used to implement the functions of the above-mentioned transceiver unit.

[0289] It is understood that the processor in the embodiments of this application may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor may be a microprocessor or any conventional processor.

[0290] The method steps in the embodiments of this application can be implemented in hardware or in software instructions executable by a processor. The software instructions can consist of corresponding software modules, which can be stored in random access memory, flash memory, read-only memory, programmable read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. The storage medium can also be a component of the processor. The processor and storage medium can reside in an ASIC. Alternatively, the ASIC can reside in a base station or user equipment. The processor and storage medium can also exist as discrete components in the base station or user equipment.

[0291] 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 programs or instructions. When the computer program or instructions are loaded and executed on a computer, the processes or functions described in the embodiments of this application are performed entirely or partially. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a network device, a user equipment, or other programmable device. The computer program or instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, the computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; it can also be an optical medium, such as a digital video optical disc; or it can be a semiconductor medium, such as a solid-state drive. The computer-readable storage medium may be a volatile or non-volatile storage medium, or may include both types of storage media.

[0292] In the various embodiments of this application, unless otherwise specified or in case of logical conflict, the terminology and / or descriptions of different embodiments are consistent and can be referenced by each other. The technical features of different embodiments can be combined to form new embodiments according to their inherent logical relationship.

[0293] In this application, "at least one" means one or more, and "more than one" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. In the textual description of this application, the character " / " generally indicates an "or" relationship between the preceding and following related objects; in the formulas of this application, the character " / " indicates a "division" relationship between the preceding and following related objects. "Including at least one of A, B, and C" can mean: including A; including B; including C; including A and B; including A and C; including B and C; including A, B, and C.

[0294] It is understood that the various numerical designations used in the embodiments of this application are merely for descriptive convenience and are not intended to limit the scope of the embodiments of this application. The order of the process numbers described above does not imply the order of execution; the execution order of each process should be determined by its function and internal logic.

Claims

1. A training method for a CSI information processing model, characterized in that, include: Receive data collection information sent by network devices; Based on the data collection information, CSI information is obtained; Based on the CSI information, a first CSI information processing model is obtained.

2. The method according to claim 1, characterized in that, The data collection information includes first data collection information and / or second data collection information.

3. The method according to claim 2, characterized in that, When the collected information includes the first information of data collection, obtaining the first CSI information processing model based on the CSI information includes: Send the CSI information to the network device; Receive parameters of the second CSI information processing model sent by the network device; Based on the parameters, the first CSI information processing model is obtained.

4. The method according to claim 2 or 3, characterized in that, When the data collection information includes the first data collection information, the CSI information is used by the network device to train a second CSI information processing model.

5. The method according to claim 4, characterized in that, The second CSI information processing model includes a second encoder model and a second decoder model.

6. The method according to claim 5, characterized in that, The network device sends the parameters of the second encoder model to the user equipment.

7. The method according to any one of claims 2 to 6, characterized in that, The data collection first information is used to indicate: the characteristics of the CSI information measured by the user equipment and / or the CSI information measured by the user equipment.

8. The method according to any one of claims 2 to 7, characterized in that, The first information for data collection includes at least one of the following: reported data sample type, reported data sample format, CSI-RS configuration information, transmission bandwidth information, maximum number of CSI data samples collected, data label consistency check mechanism, CSI dataset size, CSI data collection time interval or data collection sampling rate, CSI data encryption technology indication information, CSI data quality threshold information, and the association ID of the first CSI information processing model.

9. The method according to claim 2, characterized in that, When the collected information includes the second information of data collection, obtaining the first CSI information processing model based on the CSI information includes: Based on the CSI information, a reference CSI information processing model is trained to obtain the first CSI information processing model.

10. The method according to claim 9, characterized in that, Before training the reference CSI information processing model based on the CSI information to obtain the first CSI information processing model, the method further includes: Send the CSI information to the network device; Receive parameters of the second CSI information processing model sent by the network device; Based on the parameters, a reference CSI information processing model is obtained; The first CSI information processing model is obtained based on the CSI information and the reference CSI information processing model.

11. The method according to any one of claims 9 or 10, characterized in that, The reference CSI information processing model includes: a reference CSI information encoding model.

12. The method according to claim 11, characterized in that, The first CSI information processing model includes an actual encoder model. Obtaining the first CSI information processing model based on the CSI information and the reference CSI information processing model includes: Train a nominal decoder model based on the aforementioned reference CSI information encoding model; Based on the CSI information, the nominal decoder model is trained to obtain the actual encoder model.

13. The method according to claim 9, characterized in that, The data collection of the second information includes at least one of the following: Data sample type, reported data sample format, transmission bandwidth information, data tag consistency check mechanism for data collection, maximum number of CSI data collection samples, dataset size, time interval or sampling rate of CSI data collection, CSI-RS configuration information, threshold information for CSI data quality, association ID of the first CSI information processing model, and threshold information for model performance indicators.

14. The method according to any one of claims 8 or 13, characterized in that, When the data collection information includes the CSI-RS configuration information, the CSI-RS configuration information includes: the start time of the reference signal transmission, the number of continuous transmissions of the reference signal or the period of the reference signal transmission, and the end time of the reference signal transmission.

15. The method according to any one of claims 2 to 14, characterized in that, If the collected information includes the second data collection information, before receiving the data collection information sent by the network device, the method further includes: Send a data collection request to the network device.

16. The method according to claim 15, characterized in that, The data collection request information is used to request the network device to instruct the user equipment to collect data.

17. The method according to claim 16, characterized in that, The data collection request information includes: the frequency of the reference signal, the subcarrier spacing of the reference signal, the bandwidth of the reference signal, the number of ports of the reference signal, the start time of the reference signal transmission, the number of continuous transmissions of the reference signal or the period of the reference signal transmission, and the timing or conditions for the termination of the reference signal transmission.

18. The method according to claim 1, characterized in that, The first CSI information processing model is configured with a network model ID.

19. The method according to claim 18, characterized in that, The network model ID includes the association ID and / or dataset ID.

20. The method according to any one of claims 1-19, characterized in that, In the case where the first CSI information processing model includes an actual encoder model, the actual encoder model is configured with an actual encoder model ID.

21. The method according to claim 20, characterized in that, The actual encoder model ID includes the association ID and / or dataset ID.

22. The method according to claim 21, characterized in that, The association ID is used to represent the network device corresponding to the first CSI information processing model, and the dataset ID is used to represent the CSI information used to train the first CSI information processing model.

23. The method according to claim 21 or 22, characterized in that, Different dataset IDs correspond to different CSI datasets, and the same dataset ID corresponds to similar CSI datasets.

24. The method according to claim 19 or 21, characterized in that, The associated ID includes a static part and a dynamic part; the static part represents basic network identification information, and the dynamic part represents additional conditions of the network device.

25. A CSI information processing method, characterized in that, include: Receive CSI reference signals sent by network devices; Sending CSI information to network devices, wherein the CSI information is processed using the first CSI information processing model as described in any one of claims 1-24.

26. A training method for a CSI information processing model, characterized in that, include: Data collection information sent to user equipment; Receive CSI information obtained by the user equipment based on the data collection information; Based on the CSI information, a second CSI information processing model is obtained.

27. The method according to claim 26, characterized in that, The data collection information includes first data collection information and / or second data collection information.

28. The method according to claim 27, characterized in that, When the collected information includes the first data collection information, after receiving the CSI information obtained by the user equipment based on the data collection information, the method further includes: The parameters of the second CSI information processing model are sent to the user equipment, and the parameters are used by the user equipment to obtain the first CSI information processing model.

29. The method according to claim 27 or 28, characterized in that, When the collected information includes the first information of data collection, the CSI information is used by the network device to train a second CSI information processing model.

30. The method according to claim 29, characterized in that, The second CSI information processing model includes a second encoder model and a second decoder model.

31. The method according to claim 30, characterized in that, The network device sends the parameters of the second encoder model to the user equipment.

32. The method according to any one of claims 27 to 31, characterized in that, The data collection first information is used to indicate: the information required for the user equipment to measure CSI and / or the user equipment to measure CSI.

33. The method according to any one of claims 27 to 32, characterized in that, The first information for data collection includes at least one of the following: reported data sample type, reported data sample format, CSI-RS configuration information, transmission bandwidth information, maximum number of CSI data samples collected, data label consistency check mechanism, CSI dataset size, CSI data collection time interval or data collection sampling rate, CSI data encryption technology indication information, CSI data quality threshold information, and the associated ID of the second CSI information processing model.

34. The method according to claim 27, characterized in that, When the collected information includes the second information of data collection, the user equipment uses the CSI information to train an untrained first CSI information processing model.

35. The method according to claim 34, characterized in that, The method further includes: Receive the CSI information sent by the user equipment; Based on the CSI information, train the encoder and decoder model pair; The second CSI information processing model is obtained based on the encoder and decoder model pair.

36. The method according to claim 34 or 35, characterized in that, The user equipment obtains a reference CSI information encoding model based on the encoder model of the second CSI information processing model, and the reference CSI information encoding model is used to obtain the untrained first CSI information processing model.

37. The method according to claim 36, characterized in that, The second CSI information processing model includes an encoder and decoder model pair; the second CSI information processing model is used by the user equipment to obtain the actual encoder model, and the user equipment trains the nominal decoder model according to the reference CSI information encoding model; the user equipment is also used to train the nominal decoder model according to the CSI information to obtain the actual encoder model.

38. The method according to claim 34, characterized in that, The data collection of the second information includes at least one of the following: Data sample type, reported data sample format, transmission bandwidth information, data tag consistency check mechanism for data collection, maximum number of CSI data collection samples, dataset size, CSI data collection time interval or data collection sampling rate, CSI-RS configuration information, CSI data quality threshold information, association ID of the second CSI information processing model, and model performance index threshold information.

39. The method according to claim 33 or 38, characterized in that, When the data collection information includes the CSI-RS configuration information, the CSI-RS configuration information includes: the start time of the reference signal transmission, the number of continuous transmissions of the reference signal or the period of the reference signal transmission, and the end time of the reference signal transmission.

40. The method according to any one of claims 27 to 39, characterized in that, If the collected information includes the second data collection information, the method further includes, before sending the data collection information to the user equipment: Receive data collection request information sent by the user equipment.

41. The method according to claim 40, characterized in that, The data collection request information is used to request the network side to instruct the terminal to collect data.

42. The method according to claim 41, characterized in that, The data collection request information includes: the frequency of the reference signal, the subcarrier spacing of the reference signal, the bandwidth of the reference signal, the number of ports of the reference signal, the start time of the reference signal transmission, the number of continuous transmissions of the reference signal or the period of the reference signal transmission, and the timing or conditions for the termination of the reference signal transmission.

43. The method according to claim 26, characterized in that, The second CSI information processing model is configured with a network model ID.

44. The method according to claim 43, characterized in that, The network model ID includes the association ID and / or dataset ID.

45. The method according to any one of claims 26-44, characterized in that, When the parameters of the second CSI information processing model are used by the user equipment to obtain the first CSI information processing model, the first CSI information processing model includes an actual encoder model, which is configured with an actual encoder model ID.

46. ​​The method according to claim 45, characterized in that, The actual encoder model ID includes the association ID and / or dataset ID.

47. The method according to claim 46, characterized in that, The association ID is used to represent the network device corresponding to the second CSI information processing model, and the dataset ID is used to represent the CSI information used to train the second CSI information processing model.

48. The method according to claim 46 or 47, characterized in that, Different dataset IDs correspond to different CSI datasets, and the same dataset ID corresponds to similar CSI datasets.

49. The method according to claim 44 or 46, characterized in that, The associated ID includes a static part and a dynamic part; the static part represents basic network identification information, and the dynamic part represents additional conditions of the network device.

50. A CSI information processing method, characterized in that, include: Send CSI reference signal to user equipment; Receive CSI information sent by user equipment; The CSI information is processed using a second CSI information processing model; the second CSI information processing model is obtained using the method described in any one of claims 26-49.

51. A chip system, comprising: Memory, used to store computer programs; and at least one processor; When the at least one processor retrieves and runs a computer program from memory, the communication device equipped with the chip system performs the method of any one of claims 1 to 25.

52. A terminal device, characterized in that, include: Memory, used to store computer programs; and at least one processor; when the at least one processor calls and runs a computer program from memory, the terminal device causes the terminal device to perform the method of any one of claims 1 to 25.

53. A computer-readable storage medium, characterized in that, The storage medium stores a computer program or instructions, which, when executed by a communication device, implement the method as described in any one of claims 1 to 25.

Images