Communication method, apparatus, storage medium, and communication system

Abstract

The application provides a communication method, device, storage medium and system, and belongs to the technical field of communication. The method comprises the following steps: a network device receives first information from a terminal device, wherein the first information is used for indicating the characteristics of a wireless channel between the terminal device and the network device; and the network device sends downlink data to the terminal device according to a target channel tensor, wherein the target channel tensor is determined based on a first channel tensor and a first historical reference channel tensor, the first channel tensor is determined based on the first information, and the channel characteristics of the wireless channel associated with the first historical reference channel tensor are similar to the channel characteristics of the first channel. By using the historical high-precision first historical reference channel tensor to compensate and calibrate the damaged first channel tensor, the accuracy of the target channel tensor obtained by the network device can be improved, and the transmission performance of the downlink data can be improved.

Description

Technical Field

[0001] This application belongs to the field of communication technology, specifically relating to a communication method, communication device, computer-readable storage medium, and communication system. Background Technology

[0002] In current wireless communication systems, in order to obtain downlink channel state information, the terminal device can compress the measured channel tensor and send the compressed channel tensor to the network device. After receiving the compressed channel tensor, the network device decompresses it to recover the complete channel tensor and uses the complete channel tensor for the calculation of the precoding matrix and the transmission of downlink data.

[0003] However, the above method faces the problem of information loss, which makes the accuracy of the channel tensor recovered by the network device through decompression low. This leads to inaccurate operation of the network device in performing precoding processing on downlink data, and thus affects the transmission performance of downlink data. Summary of the Invention

[0004] In view of this, this application provides a communication method, a communication device, a computer-readable storage medium, and a communication system to improve the accuracy of the channel tensor obtained by the network device, thereby improving the transmission performance of downlink data.

[0005] In a first aspect, embodiments of this application provide a communication method. This method can be executed by a network device, or by a component (such as a circuit, chip, or chip system) configured in the network device, or by a logic module or software capable of implementing all or part of the functions of the network device. This application does not limit this. The following description uses a network device as an example.

[0006] The method includes: receiving first information from a terminal device, the first information indicating a first characteristic of a first channel, the first channel being a wireless channel between the terminal device and a network device; and sending downlink data to the terminal device based on a target channel tensor, wherein the target channel tensor is determined based on a first channel tensor and a first historical reference channel tensor, the first channel tensor being determined based on the aforementioned first information, and the channel characteristics of the wireless channel associated with the first historical reference channel tensor being similar to the channel characteristics of the first channel.

[0007] The channel characteristics of the wireless channel associated with the first historical reference channel tensor are similar to those of the first channel, which can be expressed as follows: the wireless channel associated with the first historical reference channel tensor is similar to the first channel; that is, the first historical reference channel tensor is similar to the channel tensor measured by the terminal device for the first channel. By using historical high-precision data (the first historical reference channel tensor) to compensate and calibrate the current compressed and damaged data (the first channel tensor), the accuracy of the target channel tensor obtained by the network device can be improved, thereby improving the accuracy of the precoding matrix determined based on the target channel tensor. This, in turn, improves the transmission performance of downlink data transmitted based on the precoding matrix, such as increasing the received power of the downlink signal carrying downlink data received by the terminal device, reducing inter-device interference, and improving the reliability of the communication link.

[0008] In one possible implementation, the target channel tensor is obtained by inputting a first channel tensor and a first historical reference channel tensor into a fusion network. This fusion network performs nonlinear combination processing on the first channel tensor and the first historical reference channel tensor, and is deployed in a network device. By obtaining the target channel tensor through the fusion network, the network device can adaptively suppress independent noise components in each channel tensor during the fusion of the first channel tensor and the first historical reference channel tensor, thereby obtaining a cleaner channel tensor than a single channel tensor or a channel tensor obtained through linear combination. Furthermore, the fusion network has the ability to model complex interactions between channels, and obtaining the target channel tensor through the fusion network also improves the amount of information in the channel representation.

[0009] In one possible implementation, the target channel tensor is obtained by performing a weighted operation on the first channel tensor and the first historical reference channel tensor. This weighting operation can be a weighted average or a weighted summation. This approach reduces the computational cost of acquiring the target channel tensor, thereby reducing the computational overhead for network devices in acquiring the target channel tensor.

[0010] In one possible implementation, the same shared weights are used to process the antenna positions, subcarrier positions, antenna positions, and subcarrier positions in the first channel tensor, as well as in the first historical reference channel tensor, within the aforementioned fusion network. By employing a weight-sharing strategy in the fusion network, not only can the first historical reference channel tensor be effectively used for compensation and calibration of the first channel tensor, but the parameter size of the fusion network can also be greatly compressed, significantly reducing the risk of overfitting. This allows the fusion rules to generalize to channel tensors of different sizes, improving the generalization ability and robustness of the fusion network under different channel environments.

[0011] In one possible implementation, the first information includes a first sequence, and the aforementioned first channel tensor is determined based on the first information, including: the first channel tensor is obtained by decompressing the reconstructed channel feature sequence, which is determined based on the first sequence and a target first vector; wherein, the target first vector is used to indicate a second feature of the first channel, and the degree of change of the first feature indicated by the first sequence over time is greater than the degree of change of the second feature over time; the first sequence is determined based on a second channel tensor, and the target first vector is determined based on a second channel tensor or a third channel tensor, wherein the second channel tensor and the third channel tensor are determined based on the first channel under different time periods. The network device, based on the first sequence and the target first vector, obtains the reconstructed channel feature sequence and performs decompression processing on the reconstructed channel feature sequence, enabling it to promptly obtain the first channel tensor after receiving the first information from the terminal device, thereby facilitating subsequent operations based on the channel tensor, such as obtaining the target channel tensor. Furthermore, the target first vector may be the first vector currently sent by the terminal device (i.e., the vector obtained together with the first sequence), or it may be the third vector sent by the terminal device before sending the first information. When the target first vector is the third vector sent by the terminal device before sending the first information, the terminal device may not send the aforementioned first vector, which can reduce the communication overhead of the terminal device.

[0012] In one possible implementation, the reconstructed channel feature sequence is obtained by adding the reconstructed residual sequence and the reconstructed stable feature sequence. The reconstructed residual sequence is obtained by decoding the first sequence after performing a dimensionality increase operation, or by decoding the first sequence. The reconstructed stable feature sequence is determined based on the target first vector, and its length is the same as the length of the reconstructed residual sequence. That is, the network device can obtain the reconstructed residual sequence based on the first sequence; obtain the reconstructed stable feature sequence based on the target first vector; and add the reconstructed residual sequence and the reconstructed stable feature sequence to obtain the reconstructed channel feature sequence. In this way, the network device can obtain the reconstructed channel feature sequence based on the first sequence and the target first vector.

[0013] In one possible implementation, the aforementioned first historical reference channel tensor is obtained based on the reconstructed channel feature sequence, the first historical channel feature sequence, and the second historical reference channel tensor. The first historical reference channel tensor corresponds to a first location, which is the location closest to the terminal device among multiple locations. Each of these multiple locations corresponds to at least one historical reference channel tensor, and these multiple locations are within the beam coverage area of ​​the network device. Each channel tensor in the second historical reference channel tensor corresponds one-to-one with a historical channel feature sequence in the first historical channel feature sequence. The first historical reference channel tensor includes N historical reference channel tensors associated with historical channel feature sequences. These N historical channel feature sequences are the top N most similar channel feature sequences obtained by performing similarity calculations between the reconstructed channel feature sequence and each channel feature sequence in the first historical channel feature sequence, where N is a positive integer. In this way, the similarity between the first historical reference channel tensor obtained by the network device and the second channel tensor (i.e., the channel tensor obtained by the terminal device from the measurement of the first channel) can be guaranteed. That is, the first historical reference channel tensor is more similar to or closer to the second channel tensor, so that when the network device performs compensation and calibration on the first channel tensor based on the first historical reference channel tensor, the accuracy of the target channel tensor obtained by the network device can be improved.

[0014] In one possible implementation, before sending downlink data to the terminal device based on the target channel tensor, the method of the first aspect further includes: receiving second information from the terminal device, the second information including a first vector used to indicate the aforementioned second feature, wherein the target first vector is the first vector. The terminal device may send the second information to the network device when the channel aging degree is greater than or equal to a preset threshold. The channel aging degree being greater than or equal to the preset threshold may indicate that the first channel has changed significantly, i.e., the first vector differs significantly from a third vector previously sent by the terminal device, which is the vector most recently sent by the terminal device to the network device before sending the first information, used to indicate the second feature of the first channel. Thus, when the environment of the terminal device changes significantly, the network device can use the most recently acquired first vector from the terminal device to obtain the first channel tensor, thereby ensuring the accuracy of the first channel tensor obtained by the network device, and consequently ensuring the accuracy of the target channel tensor determined based on the first channel tensor.

[0015] In one possible implementation, after receiving the first information from the terminal device, the method of the first aspect further includes: sending third information to the terminal device based on channel reconstruction error information; wherein the channel reconstruction error information is used to indicate the difference between the channel represented by the target channel tensor and the current first channel, and the third information is used to adjust fourth information, which indicates the degree of vector dimension truncation processing performed on the first full sequence, the first full sequence being determined based on the second channel tensor, and the first sequence being the sequence obtained after performing vector dimension truncation processing on the first full sequence. The channel represented by the target channel tensor can be understood as the channel reconstructed by the network device for sending downlink data to the terminal device. The network device can instruct the third information to adjust the fourth information based on the similarity between its reconstructed terminal device channel and the first channel actually measured by the terminal device. For example, when the difference between the channel reconstructed by the network device and the first channel actually measured by the terminal device is large, the network device can make the third information adjust the fourth information to a greater extent, so that the terminal device can report a sequence with more vector dimensions in the next report, thereby making the channel reconstructed by the network device in the future closer to the actual channel. In this way, the vector dimension of the sequence reported by the terminal device can be flexibly adjusted according to the actual situation, thereby improving the flexibility of network scheduling.

[0016] In one possible implementation, the channel reconstruction error information is determined based on the target channel tensor and the channel quality indicator (CQI). That is, the network device can determine the channel characteristics of its reconstructed channel (denoted as channel #11) based on the target channel tensor, and determine the channel characteristics of the first channel (denoted as channel #12) actually measured by the terminal device based on the CQI. Based on the channel characteristics of channel #11 and channel #12, the channel reconstruction error information is determined, thus ensuring that the acquired channel reconstruction error information accurately reflects the differences between channel #11 and channel #12.

[0017] In one possible implementation, the first historical reference channel tensor corresponds to a second location, which is the location closest to the terminal device among a plurality of locations. Each of these locations corresponds to at least one historical reference channel tensor, and all of these locations are within the beam coverage area of ​​the network device. The proximity of the second location to the terminal device ensures that the historical reference channel tensor corresponding to the second location is more similar to the second channel tensor measured by the terminal device compared to the historical reference channel tensors corresponding to other locations (excluding the second location among the plurality of locations). In other words, the channel characteristics of the wireless channel associated with the historical reference channel tensor corresponding to the second location are more similar to the channel characteristics of the first channel. This improves the accuracy of the target channel tensor obtained by the network device through compensation and calibration of the first channel tensor based on the first historical reference channel tensor.

[0018] Secondly, embodiments of this application provide a communication method. This method can be executed by a terminal device, or by a component (such as a circuit, chip, or chip system) configured in the terminal device, or by a logic module or software capable of implementing all or part of the functions of the terminal device. This application does not limit this. The following description uses a terminal device as an example.

[0019] The method includes: sending first information to a network device, the first information indicating a first characteristic of a first channel measured by a terminal device, the first channel being a wireless channel between the terminal device and the network device; receiving downlink data from the network device, wherein the downlink data is obtained based on a target channel tensor, the target channel tensor being determined based on a first channel tensor and a first historical reference channel tensor, the first channel tensor being determined based on the aforementioned first information, and the channel characteristics of the wireless channel associated with the first historical reference channel tensor being similar to the channel characteristics of the aforementioned first channel.

[0020] In one possible implementation, the target channel tensor is obtained by inputting the first channel tensor and the first historical reference channel tensor into a fusion network, which is used to perform nonlinear combination processing on the first channel tensor and the first historical reference channel tensor, and the fusion network is deployed in a network device.

[0021] In one possible implementation, the target channel tensor is obtained by weighting the first channel tensor and the first historical reference channel tensor. This weighting operation can be a weighted average or a weighted summation.

[0022] In one possible implementation, the first information includes a first sequence indicating that a first feature changes over time to a greater extent than a second feature of the first channel changes over time, the second feature being indicated by a first vector; the first sequence and the first vector are determined based on an original channel feature sequence, which is obtained by compressing a second channel tensor obtained from measurements of the first channel.

[0023] In one possible implementation, the original channel feature sequence is obtained by first performing frequency domain compression on the second channel tensor, and then performing spatial domain compression. By first performing frequency domain compression on the complete spatial slice channel tensor at each time step in the second channel tensor, the terminal device can effectively remove redundant information and reduce the amount of compressed data by utilizing the greater sparsity of the frequency domain. Furthermore, by performing spatial domain compression on the frequency-compressed data, the terminal device can provide extremely low-dimensional, clean input data (the original channel feature sequence) for subsequent operations while preserving the core physical structure information of the first channel to the greatest extent.

[0024] In one possible implementation, the first vector is obtained by inputting the original channel feature sequence into the stable split path of the two-stream decoupling autoencoder; the first sequence is obtained by inputting the original residual sequence into the fast-changing split path of the two-stream decoupling autoencoder, wherein the original residual sequence is obtained by subtracting the original stable feature sequence from the original channel feature sequence, and the original stable feature sequence is obtained by copying the first vector according to the length of the original channel feature sequence. By performing feature decomposition on the original channel feature sequence through two-stream decoupling autoencoder, the first sequence and the first vector can be obtained quickly and accurately; and based on the original channel feature sequence, the amount of data (such as the first vector, or the first sequence and the first vector) sent by the terminal device to the network device can be further reduced.

[0025] In one possible implementation, the first vector is obtained by inputting the original channel feature sequence into the stable split path of the dual-stream decoupling autoencoder; the first sequence is obtained by inputting the original residual sequence and the fourth information into the fast-changing split path of the dual-stream decoupling autoencoder. The original residual sequence is obtained by subtracting the original stable feature sequence from the original channel feature sequence. This original stable feature sequence is obtained by copying the first vector according to the length of the original channel feature sequence. The fourth information indicates the degree to which the vector dimension truncation processing is performed on the first full sequence generated during the encoding process of the dual-stream decoupling autoencoder. The first sequence is the sequence obtained after performing vector dimension truncation processing on the first full sequence. The terminal device performs feature decomposition on the original channel feature sequence through dual-stream decoupling autoencoder, enabling it to quickly and accurately obtain the first sequence and the first vector. Furthermore, it can adjust the vector dimension of each feature vector in the first sequence according to the actual situation of the first channel using the fourth information. For example, when the resources of the first channel are limited, the vector dimension of the first sequence can be made smaller, thereby improving the flexibility of feature acquisition.

[0026] In one possible implementation, the above-mentioned two-stream decoupled autoencoder uses a nested strategy during training. This nested strategy is used to sort the physical features of the second full sequence generated during the encoding process of the two-stream decoupled autoencoder according to their importance. The loss function used to train the two-stream decoupled autoencoder includes a first component, a second component, and a third component. The value of the first component is positively correlated with the first difference degree, which is used to characterize the difference between the channel feature sequence recovered based on the second vector and the second sequence and the initial channel feature sequence. The weight of the first component is positively correlated with the degree of truncation of the second sequence. The value of the second component is positively correlated with the correlation difference of the second vector in the time dimension. The value of the third component is positively correlated with the similarity between the second sequence and the second vector. The operation of truncation of the vector dimension of the second full sequence performed by the above-mentioned two-stream decoupled autoencoder is related to the operation of sorting the physical features of the second full sequence according to their importance. By employing a nested strategy and the aforementioned loss function during the training of the two-stream decoupled autoencoder, the vector dimensions of the feature vectors output by the trained two-stream decoupled autoencoder in the fast-changing split path can be arranged according to importance. This allows vector dimensions with smaller or larger index values ​​in the first full sequence to carry more core channel information. By sorting the vector dimensions of the first full sequence according to importance, when truncating the vector dimensions of the first full sequence, vector dimensions with higher importance can be truncated. This ensures that the truncated first sequence retains more core channel information, thereby enabling the first sequence to carry more core channel information while reducing its data volume.

[0027] In one possible implementation, the aforementioned fourth information is determined based on third information from the network device, which is used to adjust the fourth information.

[0028] In one possible implementation, the first vector is obtained by inputting the original channel feature sequence into a stable split path in the two-stream decoupled autoencoder. This includes: when inputting the original channel feature sequence into the stable split path of the two-stream decoupled autoencoder, mapping the original channel feature sequence to different subspaces using independent weight matrices through multiple attention heads in the stable split path, and then fusing the feature vectors from multiple scenarios output by the multiple attention heads. This allows the two-stream decoupled autoencoder to capture common features still existing in the environment of the terminal device even when the scene around the terminal device undergoes abrupt changes, thereby improving the generalization ability of the two-stream decoupled autoencoder.

[0029] In one possible implementation, before receiving downlink data from the network device, the method of the second aspect further includes: sending second information to the network device when the channel aging degree is greater than or equal to a preset threshold, the channel aging degree being used to indicate the degree of change of the current first channel compared to the first channel corresponding to when the terminal device sent the third vector, the third vector being a vector used by the terminal device to indicate a second characteristic of the first channel most recently sent to the network device before sending the first information, the second information including the first vector.

[0030] In one possible implementation, the channel aging degree is related to a first coefficient, a first vector, and a third vector. By determining the channel aging degree through the first and third vectors, the change in the second characteristic of the first channel can be accurately determined, and thus, when the change in the second characteristic is significant, second information is sent to the network device.

[0031] In one possible implementation, the aforementioned channel aging degree satisfies the following relationship with the first coefficient, the first vector, and the third vector: ;in, For channel aging, As the first coefficient, Let be the first vector. This is the third vector. The above relationship can accurately quantify the degree of environmental change around the terminal device, enabling the terminal device to accurately obtain the channel aging degree based on this relationship. Furthermore, through this method, the accuracy requirements of the first vector fed back by the network device to the terminal device can be dynamically adjusted. The value of allows the channel aging degree to be applied to more scenarios with different requirements.

[0032] The second aspect is the implementation on the terminal side, which corresponds to the first aspect. The explanations, supplements, and descriptions of the beneficial effects of the first aspect also apply to the second aspect, and will not be repeated here.

[0033] Thirdly, a communication device is provided. The communication device includes: a module for performing the method in any possible implementation of any of the above aspects, such as a communication module and a processing module. For example, the communication module is used to instruct the transmission and reception functions of the communication device; the processing module is used to perform functions of the communication device other than the transmission and reception functions.

[0034] Optionally, the communication module may include a transmitting module and a receiving module. The transmitting module implements the transmitting function of the communication device described in the third aspect, and the receiving module implements the receiving function of the communication device described in the third aspect.

[0035] Optionally, the communication device described in the third aspect may further include a storage module storing programs or instructions. When the processing module executes the program or instructions, the communication device can perform the methods in any of the possible implementations of any of the above aspects.

[0036] It is understood that the communication device described in the third aspect may be a network device or a terminal device, or a chip (system) or other component or assembly that can be disposed in the network device or the terminal device, or a device that includes the network device or the terminal device. This application does not limit it in this regard.

[0037] Furthermore, the technical effects of the communication device described in the third aspect can be referenced from the technical effects of the methods in any possible implementation of any of the above aspects, and will not be repeated here.

[0038] Fourthly, a communication device is provided, including at least one processor. The at least one processor is coupled to a memory storing programs or instructions. The processor executes the programs or instructions in the memory, causing the communication device to perform a method in any possible implementation of any of the above aspects. Optionally, the communication device further includes a memory. Optionally, the communication device further includes a communication interface, to which the processor is coupled.

[0039] In one implementation, the communication interface may be a transceiver, or an input / output interface.

[0040] In another implementation, the communication device is a chip configured in a terminal device or network device. When the communication device is a chip configured in a terminal device or network device, the communication interface can be an input / output interface.

[0041] Fifthly, a processor is provided, comprising: an input circuit, an output circuit, and a processing circuit. The processing circuit is configured to receive signals through the input circuit and transmit signals through the output circuit, causing the processor to execute a method in any possible implementation of any aspect.

[0042] In specific implementation, the processor can be one or more chips, the input circuit can be input pins, the output circuit can be output pins, and the processing circuit can be transistors, gate circuits, flip-flops, and various logic circuits. The input signal received by the input circuit can be received and input by, for example, but not limited to, a receiver, and the signal output by the output circuit can be output to, for example, but not limited to, a transmitter and transmitted by the transmitter. Furthermore, the input circuit and the output circuit can be the same circuit, which is used as the input circuit and the output circuit at different times. This application does not limit the specific implementation of the processor and various circuits.

[0043] In a sixth aspect, a communication device is provided, including a processor and a memory. The processor is used to read instructions stored in the memory and to receive signals via a receiver and transmit signals via a transmitter to execute the method in any possible implementation of any of the above aspects.

[0044] Optionally, the processor may be one or more, and the memory may be one or more.

[0045] In a seventh aspect, embodiments of this application provide a chip system including one or more processors for calling and executing instructions stored in memory, causing the methods in any of the above aspects or any possible implementations of the above aspects to be executed. The chip system may be composed of chips or may include chips and other discrete devices.

[0046] The chip system may include input circuits or interfaces for transmitting information or data, and output circuits or interfaces for receiving information or data.

[0047] Eighthly, a computer-readable storage medium is provided that stores a computer program or instructions, which, when executed, cause a computer to perform the method in any possible implementation of any of the preceding aspects.

[0048] Ninthly, a computer program product is provided, the computer program product comprising: a computer program or instructions, which, when the computer program is run, causes the method in any possible implementation of any of the preceding aspects to be executed.

[0049] In a tenth aspect, a communication system is provided, including the aforementioned network device and terminal device, wherein the network device is configured to perform the method in any possible implementation of the first aspect, and the terminal device is configured to perform the method in any possible implementation of the second aspect.

[0050] Optionally, the communication system may also include other devices that communicate with terminal devices and / or network devices. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings.

[0052] Figure 1 A schematic diagram of the communication system provided in the embodiments of this application; Figure 2 Flowchart of the communication method provided in the embodiments of this application Figure 1 ; Figure 3 This is a schematic diagram illustrating the compression process performed on the second channel tensor according to an embodiment of this application. Figure 4 A schematic diagram of a dual-stream decoupled self-encoder provided in an embodiment of this application. Figure 1 ; Figure 5 A schematic diagram of a dual-stream decoupled self-encoder provided in an embodiment of this application. Figure 2 ; Figure 6 A schematic diagram of the dual-stream decoupling decoder provided in the embodiments of this application. Figure 1 ; Figure 7 A schematic diagram of the dual-stream decoupling decoder provided in the embodiments of this application. Figure 2 ; Figure 8 Flowchart of the communication method provided in the embodiments of this application Figure 2 ; Figure 9 A schematic diagram of a user equipment (UE) and a base station provided in an embodiment of this application; Figure 10 This is a schematic diagram of the scenario where the UE is located, as provided in the embodiments of this application. Figure 11 A schematic diagram illustrating the model training process provided in this application embodiment; Figure 12 Schematic diagram of the communication device provided in the embodiments of this application Figure 1 ; Figure 13 Schematic diagram of the communication device provided in the embodiments of this application Figure 2 . Detailed Implementation

[0053] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0054] The technical solutions provided in this application can be applied to various communication systems, such as: Global System for Mobile Communications (GSM) systems, General Packet Radio Service (GPRS), Wireless Local Area Network (WLAN), Long Term Evolution (LTE) systems, LTE Frequency Division Duplex (FDD) systems, LTE Time Division Duplex (TDD) systems, sidelink communication systems, Universal Mobile Telecommunication System (UMTS), Worldwide Interoperability for Microwave Access (WiMAX) communication systems, non-terrestrial network (NTN) communication systems, 5th generation (5G) mobile communication systems, or new radio access technology (NR). Among these, 5G mobile communication systems can include non-standalone (NSA) and / or standalone (SA) networking. The technical solutions provided in this application can also be applied to future communication systems. This application does not limit the scope of these applications.

[0055] Figure 1 This is a schematic diagram of a communication system 100 used in an embodiment of this application. The communication system 100 may include network devices, such as... Figure 1 The network device 110 is shown. The communication system 100 may also include terminal devices, such as... Figure 1 The terminal device 120 is shown. The network device 110 and the terminal device 120 can communicate via a wireless link.

[0056] Figure 1 An exemplary network device 110 and a terminal device 120 are shown. Optionally, the communication system 100 may also include multiple network devices and / or multiple terminal devices.

[0057] The network equipment in this application can be network-side equipment such as access network and core network equipment. Access network equipment is sometimes also called access node. Access network equipment has wireless transceiver capabilities and is used to communicate with terminals. Access network equipment includes, but is not limited to, base stations, evolved NodeBs (eNodeBs), transmission reception points (TRPs) in the above-mentioned communication systems, next-generation NodeBs (gNBs) in 5G mobile communication systems, access network equipment or modules of access network equipment in open RAN (ORAN) systems, satellites in NTN communication systems, base stations in future mobile communication systems, or access nodes in WiFi systems. Access network equipment can also be modules or units capable of implementing some of the functions of a base station. Access network equipment can be macro base stations, micro base stations, or indoor stations, relay nodes or donor nodes, or wireless controllers in cloud radioaccess network (CRAN) scenarios. Optionally, access network equipment can also be servers, wearable devices, or vehicle-mounted equipment, etc. For example, the access network equipment in vehicle-to-everything (V2X) technology can be a roadside unit (RSU). Multiple access network devices in a communication system can be base stations of the same type or different types. Base stations can communicate with terminals directly or via relay stations. Terminals can communicate with multiple base stations using different access technologies. The embodiments of this application do not limit the specific technology or device form used in the access network equipment. In this application, the access network equipment is referred to as a network device.

[0058] In this application, the means for implementing the functions of a network device can be a network device itself, or a means capable of supporting the network device in implementing those functions, such as a processor, circuit, chip, or chip system. This means can be installed in or connected to the network device. In the technical solutions provided in this application, the example of a network device being used to implement the functions of a network device is used to describe the technical solutions provided in this application.

[0059] The terminal device in this application can be a wireless terminal device capable of receiving network device scheduling and instruction information. The wireless terminal device can be a device providing voice and / or data connectivity to a user, a handheld device with wireless connectivity, or other processing devices connected to a wireless modem. For example, the terminal device can communicate with one or more core networks or the Internet via a radio access network (RAN). The terminal device can also be referred to as a terminal, user equipment (UE), mobile station, mobile terminal, etc. Terminal devices can be widely used in various scenarios, such as device-to-device (D2D), vehicle-to-everything (V2X) communication, machine-type communication (MTC), Internet of Things (IoT), ultra-reliable low-latency communication (URLLC), virtual reality, augmented reality, industrial control, autonomous driving, telemedicine, smart grids, smart furniture, smart offices, smart wearables, smart transportation, smart cities, or satellite communication, etc. The terminal can be a mobile phone, tablet computer, computer with wireless transceiver capabilities, wearable device, vehicle, aircraft (such as drone, helicopter, airplane), hot air balloon, ship, robot, robotic arm, or smart home device, etc. The embodiments of this application do not limit the form of the terminal device.

[0060] In this application, the apparatus for implementing the functions of a terminal device can be the terminal device itself, or any apparatus capable of supporting the terminal device in implementing those functions, such as a processor, circuit, chip, or chip system. This apparatus can be installed in or connected to the terminal device. In the technical solutions provided in this application, the example of a terminal device being used to implement the functions of a terminal device is used to describe the technical solutions provided in this application.

[0061] Access network equipment and / or terminal equipment can be fixed or mobile. They can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; on water; or in the air on aircraft, balloons, and satellites. This application does not limit the application scenarios of the access network equipment and terminal equipment. They can be deployed in the same or different scenarios; for example, both can be deployed on land simultaneously; or the access network equipment can be deployed on land while the terminal equipment is deployed on water, etc., and so on.

[0062] In practical applications, multiple network devices can collaborate to assist terminals in achieving wireless access, with different network devices each implementing a portion of the base station's functions. For example, network devices can be central units (CUs), distributed units (DUs), CUs (control planes, CPs), CUs (user planes, UPs), or radio units (RUs), etc. CUs and DUs can be set up separately or included in the same network element, such as a baseband unit (BBU). RUs can be included in radio frequency equipment or radio frequency units, such as remote radio units (RRUs), active antenna units (AAUs), or remote radio heads (RRHs).

[0063] In different systems, CU (or CU-CP and CU-UP), DU, or RU may have different names, but those skilled in the art will understand their meaning. For example, in an ORAN system, CU can also be called O-CU (Open CU), DU can also be called O-DU, CU-CP can also be called O-CU-CP, CU-UP can also be called O-CU-UP, and RU can also be called O-RU. Any of the units among CU (or CU-CP, CU-UP), DU, and RU in this application can be implemented through software modules, hardware modules, or a combination of software and hardware modules. CU (or CU-CP and CU-UP), DU, and RU can implement different protocol layer functions.

[0064] To facilitate understanding of the embodiments of this application, the terminology used in this application is first briefly explained. Optionally, the explanation of some terms can also refer to the explanations in the 3rd Generation Partnership Project (3GPP) standard protocol. It should be understood that the technical terms in this application are only examples and not limitations. As technology evolves, technical terms will also change; where the technical meaning is the same, other technical terms should also apply to this application.

[0065] 1. A time step can be understood as a discrete point in time obtained by sampling a continuous time axis, or in other words, a discrete sampling point of a time-varying channel on the time axis. The unit of this time step can be the duration of an orthogonal frequency division multiplexing (OFDM) symbol.

[0066] 2. The Channel State Information Reference Signal (CSI-RS) is a pilot signal transmitted by network devices and received by terminal devices for downlink channel measurement. By measuring the CSI-RS, terminal devices can estimate the characteristics of the downlink channel and obtain information such as the channel tensor.

[0067] 3. The channel tensor is a multidimensional array used to represent the complex baseband response of a wireless channel across multiple physical dimensions, including time, frequency, and space. For example, the time dimension of the channel tensor can be sampled according to the OFDM symbol period, the frequency dimension can be sampled by subcarrier, and the spatial dimension can be determined by the number of transmit and receive antennas. Each element in the channel tensor describes the complex channel gain from each transmit antenna to each receive antenna in a specific time-frequency resource unit.

[0068] Currently, terminal devices can measure the channel tensor of the wireless channel and perform compression processing on the measured channel tensor, and send the compressed channel tensor to the network device. After receiving the compressed channel tensor, the network device decompresses it to recover the complete channel tensor and uses the complete channel tensor for the calculation of the precoding matrix and the transmission of downlink data.

[0069] However, the above methods suffer from information loss. For example, the compression operation of the terminal device will lose some channel information, and the compressed channel tensor may also lose information during air interface transmission. There will be a significant deviation between the channel tensor recovered by the network device based on the received incomplete information and the channel tensor measured by the terminal device. That is, in this case, the accuracy of the channel tensor recovered by the network device through decompression is low. The inaccurate recovered channel tensor will cause the precoding direction to mismatch with the actual channel state, so that the downlink signal carrying downlink data sent by the network device cannot correctly match the real channel conditions of the terminal device, thus affecting the transmission performance of downlink data.

[0070] To address the aforementioned issues, this application proposes a method whereby a network device determines a target channel tensor based on a first channel tensor (i.e., the recovered channel tensor) and a first historical reference channel tensor, and then transmits downlink data based on this target channel tensor. The first channel tensor is determined based on a first feature of the first channel reported by the terminal device, and the channel characteristics of the wireless channel associated with the first historical reference channel tensor are similar to those of the first channel. By using historical high-precision data (the first historical reference channel tensor) to compensate and calibrate the current compressed and damaged data (the recovered channel tensor), the accuracy of the target channel tensor obtained by the network device can be improved, thereby improving the accuracy of the precoding matrix determined based on the target channel tensor, and ultimately improving the transmission performance of downlink data transmitted based on the precoding matrix.

[0071] The solution provided in this application will be described in detail below with reference to the corresponding flowcharts. It is understood that the illustrative flowcharts provided in this application primarily use different devices (such as terminal devices and network devices) as examples of the execution subjects of this interactive illustration to illustrate the method, but this application does not limit the execution subjects of the interactive illustrations. For example, the devices (such as terminal devices and network devices) in the illustrative flowcharts can also be chips, chip systems, or processors that support the implementation of this method on the device, or they can be logic modules or software that can implement all or part of the functions of the device.

[0072] As a general statement, the message or signaling interactions involved in the interaction process of this application embodiment can be standard messages or signaling or newly introduced messages or signaling. This application embodiment does not make specific limitations on this.

[0073] Figure 2 This is a flowchart illustrating a communication method according to an embodiment of this application. This method can be used for interaction between a terminal device and a network device in the aforementioned communication system. The following description uses the terminal device as an example to illustrate this communication method. It can be understood that... Figure 2 The first terminal device in the process can be Figure 1 Any terminal device in the process can also be a component within the terminal device (such as a processor, chip, or chip system). Figure 2 The network devices in the middle can be Figure 1 Any network device, or a component within a network device (such as a processor, chip, or chip system).

[0074] like Figure 2 As shown, the above communication method includes the following steps: S201, the first terminal device sends first information to the network device. Correspondingly, the network device receives the first information from the first terminal device.

[0075] The first information is used to indicate the first characteristic of the first channel.

[0076] The first channel is a wireless channel between the first terminal device and the network device. That is, the first terminal device and the network device can transmit data through the first channel.

[0077] The first feature of the first channel is a feature obtained by the first terminal device from the measurement (such as channel estimation) of the first channel. In other words, the first information can be used to indicate the first feature of the first channel measured by the first terminal device.

[0078] In one possible implementation, the first information may include a compressed second channel tensor. This compressed second channel tensor is used to indicate at least one of the characteristics of the first channel, such as amplitude, phase, frequency offset, path loss, and shadow fading. That is, the first characteristic includes at least one of the characteristics of the first channel, such as amplitude, phase, frequency offset, path loss, and shadow fading. Amplitude can be understood as the ratio of the received signal voltage to the transmitted signal voltage after passing through the first channel, reflecting the attenuation or enhancement of the signal amplitude. Phase can be understood as the angular offset of the received signal relative to the transmitted signal. Frequency offset can be understood as the difference between the carrier frequency of the received signal and the carrier frequency of the transmitted signal. Path loss can be understood as the average power attenuation of the signal in the propagation environment due to distance and propagation mechanisms (diffusion, absorption, etc.). Shadow fading can be understood as slowly varying signal fluctuations caused by large obstacles (such as buildings, hills, trees, etc.). The specific principles of amplitude, phase, frequency offset, path loss, and shadow fading can be found in existing technologies and will not be elaborated here.

[0079] The compressed second channel tensor is determined based on the second channel tensor obtained by the terminal device from the measurement of the first channel. For example, after the first terminal device performs a channel estimation operation on the first channel to obtain the second channel tensor, it can compress the second channel tensor, such as by compressing the second channel tensor through an encoder, to obtain the compressed second channel tensor, and then send the compressed second channel tensor to the network device.

[0080] In another possible implementation, the first information may include a first sequence indicating at least one of the amplitude, phase, and frequency offset of the first channel. That is, the first feature includes at least one of the amplitude, phase, and frequency offset of the first channel. This first feature changes faster over time than a second feature of the first channel (such as path loss or shadowing fading), or in other words, the first feature changes more rapidly over time than the second feature of the first channel. The first sequence is determined based on a second channel tensor obtained by the terminal device from measurements of the first channel. For example, after performing channel estimation on the first channel to obtain a second channel tensor, the first terminal device can first compress the second channel tensor, then perform feature separation on the compressed second channel tensor to obtain the first sequence, and send first information containing the first sequence to the network device. The specific implementation of the first sequence and how the first terminal device obtains the first sequence is described below.

[0081] S202, the network device sends downlink data to the first terminal device based on the target channel tensor. Correspondingly, the first terminal device receives the downlink data from the network device.

[0082] The target channel tensor is determined based on the first channel tensor and the first historical reference channel tensor.

[0083] The first channel tensor can be understood as the channel tensor recovered by the network device based on the first information. There is a difference between the first channel tensor and the second channel tensor measured by the first terminal device. This difference is caused by the compression operation performed on the second channel tensor by the first terminal device.

[0084] The first channel tensor is determined based on the first information. When the first information includes a compressed second channel tensor, the network device can decompress the compressed second channel tensor after receiving the first information to obtain the first channel tensor. For example, the first terminal device compresses the second channel tensor using encoder #1; in this case, after receiving the first information, the network device can decompress the compressed second channel tensor using decoder #1 matched with encoder #1 to obtain the first channel tensor; encoder #1 and decoder #1 are an encoder-decoder pair. When the first information includes a first sequence, the network device can obtain the first channel tensor based on the first sequence and the target first vector received by the network device after receiving the first information. The target first vector, and the specific implementation of the network device obtaining the first channel tensor based on the first sequence and the target first vector, are described below.

[0085] The first historical reference channel tensor includes one or more channel tensors. These one or more channel tensors may be channel tensors actually measured before the first terminal device sends the first information to the network device (i.e., S201), such as channel tensors actually measured by different terminal devices for the wireless channel.

[0086] The channel characteristics of the wireless channels associated with the first historical reference channel tensor are similar to or close to the channel characteristics of the first channel. The wireless channels associated with the first historical reference channel tensor may include the wireless channels associated with each channel tensor in the first historical reference channel tensor. The similarity between the channel characteristics of the wireless channels associated with the first historical reference channel tensor and the channel characteristics of the first channel can mean that the wireless channels associated with the first historical reference channel tensor are similar to the first channel, or that the wireless channels associated with each channel tensor in the first historical reference channel tensor are all similar to the first channel; that is, the first historical reference channel tensor is similar to or close to the second channel tensor actually measured by the terminal device. In this application embodiment, "similar" is a relative concept. For example, compared to other historical reference channel tensors, the channel characteristics of the wireless channels associated with the first historical reference channel tensor are more similar to the channel characteristics of the first channel, or the first historical reference channel tensor is more similar to the second channel tensor, as explained below.

[0087] Network devices can determine a first historical reference channel tensor similar to the second channel tensor in various ways. For example, a network device can determine the first historical reference channel tensor based on the location information of a first terminal device; a network device can also determine the first historical reference channel tensor based on first information, as described below.

[0088] In one possible implementation, after obtaining the first channel tensor and the first historical reference channel tensor, the network device can obtain the target channel tensor through a weighted operation. That is, the target channel tensor is obtained by performing a weighted operation on the first channel tensor and the first historical reference channel tensor. This weighted operation can be a weighted average operation or a weighted summation operation.

[0089] For example, when the first historical reference channel tensor includes a channel tensor, the network device can perform a weighted operation on the first channel tensor and the first historical reference channel tensor based on their respective weights to obtain the target channel tensor. The weights of the first channel tensor and the first historical reference channel tensor can be set by the network device or pre-set in the network device, and the sum of their weights is 1. For example, if the weight of the first channel tensor is 0.8 and the weight of the first historical reference channel tensor is 0.2, then the target channel tensor = 0.8 × first channel tensor + 0.2 × first historical reference channel tensor.

[0090] When the first historical reference channel tensor comprises multiple channel tensors, the network device can perform a weighted operation on these multiple channel tensors and the first channel tensor based on their weights and the weight of the first channel tensor to obtain the target channel tensor. The weights of these multiple channel tensors and the weight of the first channel tensor can be set by the network device or pre-set in the network device, and the sum of the weights of these multiple channel tensors and the weight of the first channel tensor is 1. For example, if the first historical reference channel tensor includes channel tensor #a1 and channel tensor #a2, with the weight of the first channel tensor being 0.6 and the weights of channel tensors #a1 and #a2 both being 0.2, then the target channel tensor = 0.6 × first channel tensor + 0.2 × channel tensor #a1 + 0.2 × channel tensor #a2.

[0091] Alternatively, if the first historical reference channel tensor comprises multiple channel tensors, the network device can perform a weighted operation (i.e., a weighted average or weighted summation) on these multiple channel tensors to obtain a single channel tensor (denoted as channel tensor #A); then, a weighted operation is performed on channel tensor #A and the first channel tensor to obtain the target channel tensor. The weights of these multiple channel tensors can be set by the network device or pre-set in the network device, and the sum of the weights of these multiple channel tensors is 1. For example, the first historical reference channel tensor includes channel tensor #b1 and channel tensor #b2, channel tensor #A = 0.6 × channel tensor #b1 + 0.4 × channel tensor #b2, and the target channel tensor = 0.8 × first channel tensor + 0.2 × channel tensor #A.

[0092] By using the above methods, the computational cost of acquiring the target channel tensor can be reduced, thereby reducing the computational overhead of network devices in acquiring the target channel tensor.

[0093] In another possible implementation, after obtaining the first channel tensor and the first historical reference channel tensor, the network device can input these two tensors into a fusion network to obtain the target channel tensor output by the fusion network. That is, the target channel tensor is obtained by inputting the first channel tensor and the first historical reference channel tensor into the fusion network. The fusion network is deployed within the network device and is used to perform nonlinear combination processing on the first channel tensor and the first historical reference channel tensor.

[0094] The fusion network can be a neural network, such as a multilayer perceptron (MLP), a convolutional neural network (CNN), or a recurrent neural network (RNN). The specific principles of MLP, CNN, and RNN can be found in existing technologies and will not be elaborated upon here.

[0095] The nonlinear mapping model for the fusion process of the first channel tensor and the first historical reference channel tensor can be expressed as: ;in, For the target channel tensor, For nonlinear fusion network functions, For the first channel tensor, For the first historical reference channel tensor One channel tensor, It is a positive integer.

[0096] Network devices obtain the target channel tensor through a fusion network. When fusing the first channel tensor and the first historical reference channel tensor, the network adaptively suppresses independent noise components in each channel tensor, resulting in a cleaner channel tensor than a single channel tensor or a linearly combined channel tensor. Furthermore, the fusion network has the ability to model complex interactions between channels, and obtaining the target channel tensor through the fusion network also enhances the richness of the channel representation. The specific principles of the aforementioned fusion network and nonlinear fusion network function can be found in existing technologies and will not be elaborated upon here.

[0097] Furthermore, in the aforementioned fusion network, the same shared weights are used for processing the antenna positions in the first channel tensor, the subcarrier positions in the first channel tensor, the antenna positions in the first historical reference channel tensor, and the subcarrier positions in the first historical reference channel tensor. These shared weights can be understood as the coding function of the fusion network, or as the coefficients (or weights) of the aforementioned nonlinear fusion network function. That is, by employing a weight-sharing method in the fusion network, the channel gains (or coefficients) corresponding to all antenna positions and subcarrier positions satisfy the same nonlinear fusion relationship.

[0098] For example, for a specific element in the channel tensor with index (m,n,f) (i.e., the transmit antenna, receive antenna, and subcarrier corresponding to index (m,n,f), that is, the channel frequency domain response from the m-th transmit port of the network device to the n-th receive antenna of the terminal device on the f-th subcarrier), the channel gain of this element is calculated as follows: Where m is the transmit antenna index of the network device, n is the receive antenna index of the terminal device, f is the subcarrier index, and other parameters can be understood by referring to the above-mentioned introduction, which will not be repeated here.

[0099] By adopting a weight-sharing strategy in the fusion network, not only can the first historical reference channel tensor be effectively used to compensate and calibrate the first channel tensor, but the parameter size of the fusion network can also be greatly compressed, significantly reducing the risk of overfitting. This allows the fusion rules to be generalized to channel tensors of different sizes, improving the generalization ability and robustness of the fusion network under different channel environments.

[0100] The downlink data is obtained based on the target channel tensor. This downlink data can be understood as data sent from the network device to the first terminal device. This downlink data can be application data, such as video stream data, resource packet data, etc.; it can also be control signaling and paging messages to ensure normal network operation, such as location assistance data, paging messages, etc. This application does not limit the specific type and content of the downlink data.

[0101] After obtaining the target channel tensor, the network device can acquire a precoding matrix based on the target channel tensor. It then uses this precoding matrix to perform precoding processing on the downlink data and sends the precoded downlink signal, which carries the downlink data, to the first terminal device. The specific implementation of the network device acquiring the precoding matrix based on the target channel tensor, and the specific implementation of the network device using the precoding matrix to perform precoding processing on the downlink data, can be found in existing technologies and will not be elaborated here.

[0102] In summary, in the embodiments of this application, the network device can improve the accuracy of the target channel tensor obtained by the network device by using a first historical reference channel tensor that is similar (close) to the second channel tensor obtained by the first terminal device from the first channel measurement to compensate and calibrate the current compressed damaged data (first channel tensor). This can improve the accuracy of the precoding matrix determined based on the target channel tensor, and further improve the transmission performance of downlink data transmitted based on the precoding matrix.

[0103] Optionally, in conjunction with the above embodiments, the first terminal device may first compress the measured second channel tensor to obtain an original channel feature sequence (i.e., the compressed second channel tensor); then perform feature decomposition processing on the original channel feature sequence to obtain a first sequence and a first vector, and send the first sequence (i.e., the aforementioned first information includes the first sequence) to the network device, or send the first sequence and the first vector to the network device. After receiving the first sequence from the first terminal device, the network device may obtain the first channel tensor based on the first sequence and the target first vector (described below); then obtain the target channel tensor based on the first channel tensor and the first historical reference channel tensor; finally, send downlink data to the first terminal device based on the target channel tensor. For ease of understanding, the operations of the first terminal device and the network device will be described in detail below in the above order.

[0104] The second channel tensor is the channel tensor actually measured by the first terminal device for the first channel; in other words, the second channel tensor is the channel tensor of the first channel measured by the first terminal device. The second channel tensor can be referred to the relevant description in S101 above, and will not be repeated here. After measuring the second channel tensor, the first terminal device can perform compression processing on the second channel tensor to obtain the original channel feature sequence. This original channel feature sequence includes T feature vectors corresponding one-to-one with T time steps. These T time steps can be understood as the length of the second channel tensor in the time dimension, where T is a positive integer. The specific method by which the first terminal device performs compression processing on the second channel tensor is described below.

[0105] In one possible implementation, the first terminal device can compress the second channel tensor using a pre-set channel tensor encoder. For example, the first terminal device has a pre-set encoder #2, which can input the second channel tensor into encoder #2 to obtain the original channel feature sequence. The specific principle of performing compression processing on the channel tensor using this channel tensor encoder can be found in existing technologies and will not be elaborated here.

[0106] In another possible implementation, the first terminal device can sequentially perform frequency domain compression and spatial domain compression on the second channel tensor to obtain the original channel feature sequence. That is, the original channel feature sequence is obtained by first performing frequency domain compression on the second channel tensor and then performing spatial domain compression.

[0107] For example, the first terminal device can first obtain the complete spatial slice channel tensor at each of the T time steps corresponding to the second channel tensor, based on the second channel tensor. This complete spatial slice channel tensor can be represented as... , Where C is a complex number, M is the number of transmitting antennas of the network device, and N is the number of receiving antennas of the first terminal device. f This represents the number of subcarriers.

[0108] like Figure 3 As shown, after the first terminal device obtains the complete spatial slice channel tensor corresponding to the t-th time step in T time steps, it can perform frequency domain compression processing on the frequency domain dimension of the complete spatial slice channel tensor while keeping the spatial dimension of the complete spatial slice channel tensor unchanged. For example, the complete spatial slice channel tensor... The input is fed into the frequency domain encoder to obtain the first intermediate tensor after frequency domain compression. , N c This represents the dimension value (i.e., the number of dimensions) after frequency domain compression, where t takes values ​​from 1 to T, and t is a positive integer. Performing compression on this complete spatial slice channel tensor allows the high-dimensional frequency domain vectors within the complete spatial slice channel tensor to be compressed. ( Mapping and compressing to low-dimensional frequency domain feature vectors ( (where R is a real number), meaning the frequency domain dimension of the complete spatial slice channel tensor is N. f It dropped to N c That is, N c Less than N f The first terminal device acquires the first intermediate tensor. Then, a spatial encoder (such as a three-dimensional (3D) convolutional autoencoder) can be used to process this first intermediate tensor. Mapped to a low-dimensional spatial frequency joint eigenvector , , express The vector dimension values ​​(i.e., the number of vector dimensions, or the number of numerical components) are the remaining vector dimension values ​​after frequency and spatial domain compression. Less than After performing the above-mentioned frequency and spatial compression processing on the complete spatial slice channel tensor at each time step, the first terminal device can obtain the low-dimensional spatial frequency joint feature vector at each time step. The low-dimensional spatial frequency joint feature vector at each time step is arranged in chronological order to form the original channel feature sequence. That is, the original channel feature sequence includes T low-dimensional spatial frequency joint feature vectors.

[0109] The formula for the frequency domain encoder described above can be expressed as: ;in, This is the intermediate tensor after frequency domain compression (such as the first intermediate tensor). For the frequency domain encoder network function, For the complete spatial slice channel tensor at time step, Let be the set of trainable parameters for the encoder network. The formula for the frequency domain decoder that matches the frequency domain encoder can be expressed as: ;in, For the reconstructed channel tensor (such as the first channel tensor). For frequency domain decoder network functions, It is a low-dimensional frequency domain intermediate tensor (such as a second intermediate tensor). This represents the set of trainable parameters for the frequency domain decoder network. This frequency domain decoder is deployed in the network device and can be used to perform frequency domain decoding on the second intermediate tensor after spatial domain decoding (described below). The aforementioned frequency domain encoder can use mean-square error (MSE) joint regularization as the training objective function, and its loss function formula is: ;in, For mathematical expectation operators, For the encoding process, For the decoding process, These are the regularization weight coefficients. In response to The regularization penalty function.

[0110] The formula for the spatial encoder described above can be expressed as: ;in, This is the compressed low-dimensional spatial frequency joint feature vector. For spatial encoder network functions, This is the intermediate tensor after frequency domain compression (such as the first intermediate tensor). Let be the set of trainable parameters for the spatial encoder network. The formula for the spatial decoder that matches this spatial encoder can be expressed as: ;in, It is a low-dimensional frequency domain intermediate tensor (such as a second intermediate tensor). For spatial encoder network functions, The low-dimensional spatial frequency joint feature vector (such as the feature vectors in the reconstructed channel feature sequence) is the input to the spatial domain decoder. This represents the set of trainable parameters for the spatial decoder network. The spatial decoder is deployed in the network device and can be used for frequency domain decoding of the first information (described below). The loss function used by the above spatial encoder during training can be: ;in, For mathematical expectation operators, For the encoding process, For the decoding process, These are the regularization weight coefficients. In response to The regularization penalty function is used. It is understood that the specific principles of the frequency domain encoder, frequency domain decoder, spatial domain encoder, and spatial domain decoder can be found in existing technologies, and will not be elaborated upon here.

[0111] For example, for each complete spatial slice channel tensor , ,Right now The frequency domain dimension is 2048, the frequency domain encoder is a 14-layer autoencoder, and the spatial domain encoder is a 3D convolutional 10-layer autoencoder. By performing frequency domain compression on each complete spatial slice channel tensor using this frequency domain encoder, the frequency domain dimension of each complete spatial slice channel tensor can be reduced to 16, yielding the first intermediate tensor. , The first intermediate tensor is processed by a spatial encoder. Perform spatial compression to obtain feature vectors. That is, the spatial compression process can compress the intermediate tensor. The corresponding 256×16×16 dimensions are reduced to 2048 dimensions. It is understandable that the specific principles of the aforementioned 14-layer autoencoder and 3D convolutional 10-layer autoencoder can be found in existing technologies, and will not be elaborated upon here.

[0112] In this embodiment, the first terminal device first performs frequency domain compression on the complete spatial slice channel tensor at each time step in the second channel tensor. This utilizes the greater sparsity of the frequency domain to effectively remove redundant information and reduce the amount of data after frequency domain compression. Furthermore, the first terminal device performs spatial domain compression on the frequency-compressed data, which can provide extremely low-dimensional, clean input data (original channel feature sequence) for subsequent operations while preserving the core physical structure information of the first channel to the greatest extent.

[0113] In addition, during the aforementioned dimensional mapping process, the data (such as the channel tensors of each complete spatial slice) can be processed using complex values, or the data (such as the channel tensors of each complete spatial slice) can be processed by separating the real and imaginary parts into real vectors. For specific principles, please refer to existing technologies, which will not be elaborated here.

[0114] After acquiring the original channel feature sequence, the first terminal device can obtain a first sequence and a first vector based on the original channel feature sequence. In this embodiment, the original channel feature sequence is obtained by compressing the second channel tensor obtained from the first channel measurement, and the first sequence and the first vector are determined based on the original channel feature sequence; the first information includes the first sequence, where the degree of change of the first feature indicated by the first sequence over time is greater than the degree of change of the second feature of the first channel over time, and this second feature is indicated by the first vector. After the first terminal device acquires the first sequence, by sending the first information indicating the first sequence to the network device, the network device can obtain the first channel tensor in a timely manner based on the first information.

[0115] The first terminal device can input the original channel feature sequence into a two-stream decoupled autoencoder to obtain a first sequence and a first vector. This will be explained in detail below.

[0116] A two-stream decoupling autoencoder can be used to decompose the original channel feature sequence into a first sequence and a first vector. This two-stream decoupling autoencoder is deployed in a first terminal device.

[0117] The original channel feature sequence is the sequence obtained by the first terminal device after performing compression processing on the second channel tensor. The second channel tensor can be referred to in the aforementioned related descriptions, and will not be repeated here.

[0118] The first sequence is used to indicate a first characteristic of the first channel. This first characteristic may include at least one of the amplitude, phase, and frequency offset of the first channel. The first characteristic changes over time to a greater extent than the second characteristic (described below). Furthermore, the change in this first characteristic over time is caused by small-scale fading resulting from the rapid movement of the user or scatterer, as well as multipath propagation.

[0119] The first vector is used to indicate a second characteristic of the first channel. This second characteristic may include path loss and / or shadowing fading of the first channel. The second characteristic changes less over time than the first characteristic. Furthermore, this second characteristic is related to fixed objects (such as buildings, trees, etc.) in the scene where the first terminal device is located.

[0120] During the process of acquiring the first sequence and the first vector through a two-stream decoupling autoencoder, the first terminal device can perform truncation (or clipping) processing on the vector dimension of the feature vector to reduce the data size of the first sequence. The following describes different methods for acquiring the first sequence and the first vector based on the original channel feature sequence.

[0121] Case 1.1: The two-stream decoupled autoencoder does not perform truncation on the vector dimension of the feature vector.

[0122] In this scenario, the first terminal device can input the original channel feature sequence into a two-stream decoupling autoencoder to obtain a first sequence and a first vector output by the two-stream decoupling autoencoder. At this point, the first sequence is the sequence that has not undergone vector dimension truncation processing.

[0123] For example, such as Figure 4 As shown, the dual-stream decoupled autoencoder includes a stable splitting path and a fast-changing splitting path. The stable splitting path includes a normalization layer, a fully connected layer, a global average pooling layer, a fully connected mapping layer, and a feature processing module. The fast-changing splitting path includes a gated recurrent unit (GRU) layer and a fully connected mapping layer. It is understood that the above-mentioned layer types can refer to existing technologies, and will not be elaborated further here.

[0124] The first terminal device will generate the original channel feature sequence The input is fed to a two-stream decoupled autoencoder. This is the original channel feature sequence. A feature vector is obtained through a stable splitting path of the two-stream decoupled autoencoder. ( , The vector dimension is L f This refers to the first vector mentioned above. The first vector is output in the stable shunt path of the two-stream decoupled autoencoder. Afterwards, a two-stream decoupled self-encoder (such as...) Figure 4 The feature processing module in the middle can process the first vector In the time dimension, increase the dimension by T time steps, such as the first vector The sequence is copied T times to form the original stable feature sequence; then, the original stable feature sequence is subtracted from the original channel feature sequence to obtain the original residual sequence. ( Finally, the original residual sequence The fast-changing shunt path from the input to the dual-stream decoupled autoencoder (e.g.) Figure 4 The GRU layer and fully connected mapping layer in the process yield T feature vectors output by the fast-variable splitting path, which is the first sequence mentioned above. This first sequence can be represented as... , That is, the vector dimension of the feature vector is L. d L d With L f They can be the same or different.

[0125] In the fast-changing split path of the dual-stream decoupled autoencoder, the GRU layer can learn the temporal variation law of the original residual sequence, so that the first sequence output by the dual-stream decoupled autoencoder can accurately reflect the temporal situation of the first channel.

[0126] Furthermore, in case 1.1, the formula for the two-stream decoupled autoencoder can be expressed as: ;in, Let be the first vector. As the first sequence, This is a two-stream decoupled autoencoder function. Let T be the original channel feature sequence over T consecutive time steps. The formula for the two-stream decoupled decoder matched with the two-stream decoupled autoencoder can be expressed as: ;in, The reconstructed channel feature sequence is recovered based on the first sequence and the target first vector. This is a two-stream decoupling decoder function. The target first vector is used. This dual-stream decoupling decoder is deployed in the network device, and it can be used to process the first sequence and the target first vector received by the network device to obtain the reconstructed channel feature sequence (described below).

[0127] The loss function used during training of the above two-stream decoupled autoencoder can be: ;in, This is the time averaging factor, which is used to average the errors over T time steps. This is the initial channel feature sequence; To decode the reconstructed features, the decoded and reconstructed features are represented using a second vector. and the second sequence The channel feature sequence recovered by the dual-stream decoupling decoder; The background stability weight coefficient, The temporal coherence difference of the second vector is used to calculate the difference between the second vectors extracted at different time steps. This is the second vector currently obtained. The vector (hereinafter referred to as the third vector) most recently sent to the network device before the acquisition of the second vector is used to indicate the second characteristic of the first channel. For decoupling weight coefficients, For the second vector With the second sequence The cosine similarity calculation is performed, and the closer the cosine value obtained from the cosine similarity calculation is to zero, the better the second vector... With the second sequence The more orthogonal.

[0128] The aforementioned initial channel feature sequence can be understood as the sequence obtained after compressing the channel tensor of the wireless channel. This initial channel feature sequence is similar to the original channel feature sequence, but the difference lies in that the initial and original channel feature sequences are input data for the two-stream decoupled autoencoder at two different stages. Specifically, the initial channel feature sequence is the input data used by the two-stream decoupled autoencoder during training, while the original channel feature sequence is the input data used by the two-stream decoupled autoencoder during inference. In other words, the initial and original channel feature sequences are two types of data of the same type. The similarities can be understood by referring to the aforementioned original channel feature sequence, and will not be repeated here.

[0129] The aforementioned second sequence is used to indicate the first feature of the first channel, and the second sequence is obtained based on the aforementioned initial channel feature sequence. This second sequence is similar to the first sequence, except that the second and first sequences are the same type of output data from the two-stream decoupled autoencoder at two different stages. Specifically, the second sequence is the output data of the two-stream decoupled autoencoder during training, and the first sequence is the output data of the two-stream decoupled autoencoder during inference. In other words, the second and first sequences are two types of the same data. The similarities can be understood by referring to the aforementioned first sequence, and will not be repeated here.

[0130] The aforementioned second vector is used to indicate the second feature of the first channel, and the second vector is obtained based on the aforementioned initial channel feature sequence. This second vector is similar to the first vector, except that the second vector and the first vector are the same type of output data from the two-stream decoupled autoencoder at two different stages. That is, the second vector is the output data of the two-stream decoupled autoencoder during training, and the first vector is the output data of the two-stream decoupled autoencoder during inference. In other words, the second vector and the first vector are two types of the same data. The similarities can be understood by referring to the aforementioned first vector, and will not be repeated here.

[0131] In this embodiment, the first vector is obtained by inputting the original channel feature sequence into the stable split path of the dual-stream decoupling autoencoder; the first sequence is obtained by inputting the original residual sequence into the fast-change split path of the dual-stream decoupling autoencoder, the original residual sequence being obtained by subtracting the original stable feature sequence from the original channel feature sequence, and the original stable feature sequence being obtained by copying the first vector according to the length of the original channel feature sequence. The first terminal device performs feature decomposition on the original channel feature sequence through dual-stream decoupling autoencoder, which can quickly and accurately obtain the first sequence and the first vector; and can further reduce the amount of channel feature data (such as the first vector, or the first sequence and the first vector) sent by the first terminal device to the network device based on the original channel feature sequence.

[0132] Furthermore, in the stable splitting path of the aforementioned two-stream decoupled autoencoder, a multi-head attention mechanism can be used. In this case, the first vector obtained by inputting the original channel feature sequence into the stable splitting path of the two-stream decoupled autoencoder can specifically include: when the original channel feature sequence is input into the stable splitting path of the two-stream decoupled autoencoder, the original channel feature sequence is mapped to different subspaces using independent weight matrices by multiple attention heads in the stable splitting path, and then the feature vectors from these multiple attention heads are fused together. The multi-head attention module corresponding to the multi-head attention mechanism can be set... Figure 4 In the fully connected layer, or rather, Figure 4 The fully connected layer in the code can be replaced with a multi-head attention layer. The specific principles of the multi-head attention mechanism can be found in existing technologies and will not be elaborated here.

[0133] In this embodiment, by utilizing multiple parallel attention heads in the stable splitting path of the dual-stream decoupled autoencoder, each attention head maps the original channel features to different subspaces using an independent weight matrix. The local common feature vectors output by all attention heads are then processed through a feature compression layer (such as a fully connected layer). This allows feature vectors from multiple scenarios to be fused into a single global vector (i.e., the first vector), which serves as the anchor point for the current environment of the first terminal device. Thus, even when the scene around the first terminal device undergoes abrupt changes, the dual-stream decoupled autoencoder can still capture the common features remaining in the environment of the first terminal device within T time steps, thereby improving the generalization ability of the dual-stream decoupled autoencoder.

[0134] Case 1.2: The two-stream decoupled autoencoder performs truncation on the vector dimension of the feature vector.

[0135] In this case, the first terminal device can input the original channel feature sequence and the fourth information into the dual-stream decoupling autoencoder to obtain the first sequence and the first vector output by the dual-stream decoupling autoencoder.

[0136] The fourth information indicates the degree to which the vector dimension truncation process is performed on the first full sequence generated during the encoding process of the dual-stream decoupled autoencoder, or it indicates the vector dimension size of the first sequence sent by the first terminal device. This first full sequence can be understood as the sequence obtained after the fast-change splitting path of the dual-stream decoupled autoencoder processes the original residual sequence. This sequence includes T feature vectors, each of which has L... d Each vector dimension. d Each vector dimension can reflect the characteristics of the first channel in the time domain, frequency domain, and spatial domain. d With L f They can be the same or different. Vector dimension truncation can be understood as truncating the vector dimension of the feature vectors, such as truncating the vector dimension of each feature vector in the T feature vectors included in the first full sequence.

[0137] For example, the fourth information is used to indicate the extraction of the first 1 / 2L of the T feature vectors included in the first full sequence. d The vector dimension, that is, retaining the L of each feature vector among the T feature vectors. d The first half of the vector dimensions, arranged by their index numbers. For example, L d The dimension is 100, meaning that the dimension of each feature vector in the T feature vectors included in the first full sequence is 100. The vector dimensions of each feature vector are arranged according to the index number starting from 1, and the vector dimensions with index numbers from 1 to 50 are truncated. That is, the vector dimension of the T feature vectors after truncating the vector dimensions is 50, which means that the vector dimension of the first sequence is 50.

[0138] For example, the fourth information is used to indicate the extraction of the last 1 / 3L of the T feature vectors included in the first full sequence. d The vector dimension, that is, retaining the L of each feature vector among the T feature vectors. d In a vector dimension, the last 1 / 3 of the vector dimensions are arranged according to their index numbers. For example, L d The vector dimension is 240, meaning that the vector dimension of each feature vector in the T feature vectors is 240. The vector dimensions of each feature vector are arranged according to the index number starting from 1, and the vector dimensions with index numbers from 161 to 240 are truncated. That is, the vector dimension of the T feature vectors after truncating the vector dimensions is 80, that is, the vector dimension of the first sequence is 80.

[0139] The fourth information can be related to the state of the first channel. When the channel conditions of the first channel are poor or resources are limited, the fourth information can indicate that the vector dimension in the first full sequence is truncated according to vector dimension #1. When the channel conditions of the first channel are moderate, the fourth information can indicate that the vector dimension in the first full sequence is truncated according to vector dimension #2, where the value of vector dimension #2 is greater than the value of vector dimension #1 and less than the total vector dimensions of all feature vectors in the first full sequence. When the channel conditions of the first channel are good or resources are sufficient, the fourth information can indicate that all vector dimensions in the first full sequence are truncated, i.e., all vector dimensions of each feature vector in the first full sequence are retained. The above channel conditions can be determined based on the signal-to-interference-plus-noise ratio (SINR), the reference signal received power (RSRP), or the channel quality indicator (CQI). For example, a CQI value of 0 to 6 indicates poor channel conditions; a CQI value of 7 to 10 indicates moderate channel conditions; and a CQI value of 11 to 15 indicates good channel conditions. It can be understood that when determining channel conditions using different parameters, the correspondence between the range of parameter values ​​and channel conditions can be determined based on the actual situation; this will not be elaborated further here.

[0140] The fourth information can be determined by the network device based on the state of the first channel and sent to the first terminal device. For example, the network device can determine the state of the first channel based on the CQI reported by the first terminal device; based on the state of the first channel, it can determine the fourth information and send it to the first terminal device, which in turn receives the fourth information from the network device. Alternatively, the fourth information can be determined based on third information from the network device, meaning that the third information can be used to adjust or update the fourth information, as described below.

[0141] The first sequence is the feature obtained after performing vector dimension truncation on the first full sequence. This first sequence includes all feature vectors obtained after performing vector dimension truncation on the first full sequence. The first full sequence can be understood as the sequence obtained after the fast-variable splitting path of the two-stream decoupled autoencoder processes the original residual sequence.

[0142] The first vector is a feature vector output after the original channel feature sequence passes through the stable splitting path of the dual-stream decoupled autoencoder. For details, please refer to the relevant introduction in Case 1.1 above, which will not be repeated here.

[0143] For example, such as Figure 5As shown, the dual-stream decoupled autoencoder includes a stable splitting path and a fast-changing splitting path. The stable splitting path includes a normalization layer, a fully connected layer, a global average pooling layer, a fully connected mapping layer, and a feature processing module. The fast-changing splitting path includes a GRU layer and a fully connected mapping layer. It is understood that the above-mentioned layer types can refer to existing technologies, and will not be elaborated further here.

[0144] The first terminal device will generate the original channel feature sequence and the fourth information The input is fed to a two-stream decoupled autoencoder. This is the original channel feature sequence. A feature vector is obtained through a stable splitting path of the two-stream decoupled autoencoder. ( ), that is, the first vector mentioned above. After the two-stream decoupling autoencoder obtains this first vector, the two-stream decoupling autoencoder (such as...) Figure 5 The feature processing module can obtain the original stable feature sequence based on the first vector, and obtain the original residual sequence based on the original stable feature sequence and the original channel feature sequence. ( ); the original residual sequence The fast-changing shunt path from the input to the dual-stream decoupled autoencoder (e.g.) Figure 5 (GRU layer in the middle), the fast-changing split path is obtained based on the original residual sequence. and the fourth information The first output sequence , L D Less than or equal to L d .

[0145] In the fast-change splitting path of the aforementioned dual-stream decoupled autoencoder, a GRU layer is set up. For details, please refer to the relevant introduction in section 1.1 above; it will not be repeated here. The original residual sequence... and the fourth information After being input into the fast-variable shunt path of the two-stream decoupled autoencoder, the fast-variable shunt path is based on the original residual sequence. Generate the first full sequence ( The first full sequence is obtained in the fast-changing split path. Then, the fast-changing split path (such as...) Figure 5 The fully connected mapping layer in the middle is based on the fourth information. For the first full sequence The vector dimension of the feature vectors in the sequence is truncated to obtain the first sequence. The specific implementation of the first full sequence and the truncation of the first full sequence based on the fourth information can be understood by referring to the relevant introductions above, and will not be repeated here.

[0146] The formula for the above two-stream decoupled autoencoder can be expressed as: ;in, Let be the first vector. For the first sequence, E This is a two-stream decoupled autoencoder function. Given the original channel feature sequence over T consecutive time steps, This is the fourth piece of information. The formula for the two-stream decoupling decoder matched with the two-stream decoupling autoencoder can be expressed as: ;in, The reconstructed channel feature sequence is recovered based on the first sequence and the target first vector. This is a two-stream decoupling decoder function. The target first vector is used. This dual-stream decoupling decoder is deployed in the network device, and it can be used to process the first sequence and the target first vector received by the network device to obtain the reconstructed channel feature sequence (described below).

[0147] In this embodiment, the first vector is obtained by inputting the original channel feature sequence into the stable split path of the dual-stream decoupling autoencoder; the first sequence is obtained by inputting the original residual sequence and the fourth information into the fast-changing split path of the dual-stream decoupling autoencoder. The original residual sequence is obtained by subtracting the original stable feature sequence from the original channel feature sequence. The original stable feature sequence is obtained by copying the first vector according to the length of the original channel feature sequence. The fourth information indicates the degree to which the vector dimension truncation processing is performed on the first full sequence generated during the encoding process of the dual-stream decoupling autoencoder. The first sequence is the sequence obtained after performing vector dimension truncation processing on the first full sequence. The first terminal device performs feature decomposition on the original channel feature sequence through dual-stream decoupling autoencoder, enabling it to quickly and accurately obtain the first sequence and the first vector. Furthermore, it can adjust the vector dimension of the first sequence according to the actual situation of the first channel. For example, when the first channel conditions are poor, the vector dimension of the first sequence is made smaller, thereby saving communication resources when communication resources are scarce.

[0148] Furthermore, in the stable splitting path of the above-mentioned dual-stream decoupled autoencoder, a multi-head attention mechanism can be used. For details, please refer to the relevant introduction in Case 1.1 above, which will not be repeated here.

[0149] Furthermore, in Case 1.2, a nested strategy can be used during the training of the two-stream decoupled autoencoder. This nested strategy is used to sort the physical features of the second full sequence generated during the encoding process of the two-stream decoupled autoencoder according to their importance. The loss function used to train the two-stream decoupled autoencoder may include a first component, a second component, and a third component. The value of the first component is positively correlated with the first dissimilarity, which is used to characterize the difference between the channel feature sequence recovered based on the second vector and the second sequence and the initial channel feature sequence. The weight of the first component is positively correlated with the degree of truncation of the second sequence. The value of the second component is positively correlated with the correlation difference of the second vector in the time dimension. The value of the third component is positively correlated with the similarity between the second sequence and the second vector. The operation of truncation of the vector dimension of the second full sequence by the two-stream decoupled autoencoder is related to the operation of sorting the physical features of the second full sequence according to their importance.

[0150] The nesting strategy can be specifically used to sort the vector dimensions of each feature vector in the T feature vectors included in the second full sequence according to their importance, such as arranging the vector dimensions in descending or ascending order of importance.

[0151] The second full sequence can be understood as the sequence obtained after the initial residual sequence is processed by the fast-variable splitting path of the two-stream decoupled autoencoder during the training process. This second full sequence is similar to the first full sequence mentioned above, except that the second and first full sequences are data generated by the two-stream decoupled autoencoder at two different stages. Specifically, the second full sequence is generated based on the initial residual sequence during the training phase, while the first full sequence is generated based on the original residual sequence during the inference phase. In other words, the second and first full sequences are two types of data of the same type. The similarities can be understood by referring to the first full sequence mentioned above, and will not be repeated here. Furthermore, the initial residual sequence is obtained based on the initial channel feature sequence, which is similar to the original residual sequence. The difference is that the initial residual sequence and the original residual sequence are data generated by the two-stream decoupled autoencoder in two different stages. That is, the initial residual sequence is the data generated by the two-stream decoupled autoencoder based on the initial channel feature sequence during the training stage, and the original residual sequence is the data generated by the two-stream decoupled autoencoder based on the original channel feature sequence during the inference stage. In other words, the initial residual sequence and the original residual sequence are two types of data of the same type. The similarities can be understood by referring to the aforementioned initial residual sequence, which will not be repeated here.

[0152] The loss function of a two-stream decoupled autoencoder can be expressed as: .

[0153] This corresponds to the first component mentioned above. Here is the granular weight function, which represents the loss weights corresponding to different truncation lengths during the training of a two-stream decoupled autoencoder. It is positively correlated with the value of the first component. Corresponding to the first degree of difference mentioned above, This is the time averaging factor, which is used to average the errors over T time steps. This is the initial channel feature sequence; To decode the reconstructed features, the decoded and reconstructed features are represented using a second vector. and the second sequence after being truncated by k vector dimensions The channel feature sequence recovered by the dual-stream decoupling decoder, k can be understood as the second sequence. The vector dimension value. In this embodiment, to ensure basic performance at extremely low bitrates, a higher weight can be assigned to the reconstruction loss of shorter dimensions; that is, for the first sequence with a smaller dimension, a larger weight can be set. .

[0154] This corresponds to the second component mentioned above. This is the background stability weighting coefficient, which can adjust the proportion of the second component in the total loss; The temporal coherence difference of the second vector is used to calculate the difference between the second vectors extracted at different time steps. This is the second vector currently obtained. This is the vector most recently sent to the network device before the acquisition of this second vector, which indicates the second characteristic of the first channel.

[0155] This corresponds to the third component mentioned above. These are the decoupling weight coefficients; For the second vector With the second full sequence The cosine similarity calculation is performed, and the closer the cosine value obtained from the cosine similarity calculation is to zero, the better the second vector... With the second full sequence The more orthogonal.

[0156] It is understood that the above-mentioned initial channel feature sequence, second vector, and second sequence can be referred to the relevant introduction in Case 1.1 above, and will not be repeated here.

[0157] In this embodiment, by using a nested strategy and the aforementioned loss function during the training of the two-stream decoupled autoencoder, the vector dimensions of each feature vector output by the trained two-stream decoupled autoencoder in the fast-changing split path can be arranged according to importance. Furthermore, in this case, the fourth information can also indicate the direction for truncation of the vector dimension of the first full sequence.

[0158] For example, when the vector dimension with a smaller index value in the first full sequence can carry more core channel information, the fourth information can also instruct the second full sequence to be truncated according to the vector dimension index number in ascending order. For example, the fourth information is denoted as... In the first full sequence, each feature vector has L d One dimension; When the value is 0, the first 1 / 3L of the first full sequence is extracted. d One vector dimension; When the value is 1, the first 1 / 2L of the first full sequence is extracted. d One vector dimension; When the value is 2, the entire vector dimension of the first full sequence is extracted. (The above...) Different values ​​of can correspond to different levels of first-sequence compression precision, for example: A value of 0 indicates that the current feedback for the first sequence is in low-precision mode, which corresponds to the highest compression ratio. A value of 1 indicates that the current feedback for the first sequence is in medium precision mode, which corresponds to a medium compression rate. A value of 2 indicates that the current feedback for the first sequence is in high-precision mode, which corresponds to the highest reduction ratio.

[0159] For example, when the vector dimension with a larger index value in the first full sequence can carry more core channel information, the fourth information can also indicate that the first full sequence be truncated according to the vector dimension index number in descending order. Of course, the dual-stream decoupling autoencoder can also be preset with a direction for truncating the first full sequence, that is, in this case, the fourth information may not indicate the direction for truncating the first full sequence.

[0160] By sorting the vector dimensions of the first full sequence according to their importance, when truncating the vector dimensions of the first full sequence, the vector dimensions with higher importance can be truncated. This allows the truncated features (i.e., the first sequence) to retain more core channel information, thereby reducing the amount of data in the first sequence while enabling the first sequence to carry more core channel information.

[0161] It is understandable that the specific implementation of Case 1.2 above is similar to that of Case 1.1 above. The difference is that the input of the fast-change splitting path in Case 1.2 is the original residual sequence and the fourth information, and the first sequence output by the fast-change splitting path is the sequence after performing vector dimension truncation on the first full sequence; the input of the fast-change splitting path in Case 1.1 is the original residual sequence, and the first sequence output by the fast-change splitting path has not undergone vector dimension truncation. That is, the first sequence output by the fast-change splitting path in Case 1.1 is the same as the first full sequence generated in the fast-change splitting path in Case 1.2. The similarities can be referred to for mutual understanding, and will not be elaborated here.

[0162] The above content, through cases 1.1 and 1.2, describes the specific implementation of the first terminal device acquiring the first sequence and the first vector using a dual-stream decoupled autoencoder. After acquiring the first sequence and the first vector, the first terminal device can send first information indicating the first sequence to the network device, and determine whether to send the first vector to the network device based on the channel aging degree. This will be described in detail below.

[0163] Before the network device sends downlink data to the first terminal device, or before the first terminal device receives downlink data from the network device, the method may further include: when the channel aging degree is greater than or equal to a preset threshold, the first terminal device sends second information to the network device, and the network device receives the second information from the first terminal device, wherein the second information includes a first vector, which is used to indicate a second feature of the first channel, and the target first vector (described below) is the first vector.

[0164] Channel aging is used to indicate the degree of change of the current first channel compared to the first channel when the first terminal device sent the third vector. The current first channel can be understood as the channel used by the first terminal device during the acquisition of first information (denoted as channel #1), such as the channel used by the first terminal device to measure the first channel when acquiring the second channel tensor. The third vector is the vector most recently sent to the network device before the first terminal device sent the first information, used to indicate the second characteristic of the first channel. The third vector is determined based on the third channel tensor. The specific implementation of obtaining the third vector based on the third channel tensor is similar to the specific implementation of obtaining the first vector based on the second channel tensor, except that the channel tensor used when obtaining the third vector is different. The similarities can be understood by referring to the aforementioned related introductions, which will not be repeated here. The first channel corresponding to the first terminal device sending the third vector can be understood as the channel used by the first terminal device during the acquisition of the third vector (denoted as channel #2), such as the channel used by the first terminal device to measure the first channel when acquiring the third channel tensor.

[0165] A channel aging degree greater than or equal to a preset threshold indicates that channel #1 has undergone a significant change compared to channel #2. In this case, the environment surrounding the first terminal device may have changed abruptly during the time period from channel #2 to channel #1, meaning that the original environment of the first terminal device has become invalid. This preset threshold can be flexibly set according to the actual situation, and this application does not limit it.

[0166] The channel aging degree mentioned above can be determined in different ways. For example, the channel aging degree can be determined based on the first vector and the third vector transmitted by the first terminal device in the last transmission. Alternatively, the channel aging degree can be determined based on RSRP or reference signal received quality (RSRQ). These will be explained below.

[0167] In one possible implementation, the channel aging degree is related to a first coefficient, a first vector, and a third vector. That is, the channel aging degree can be determined based on the first coefficient, the first vector, and the third vector. Determining the channel aging degree based on the first and third vectors allows for accurate determination of the changes in the second characteristic of the first channel, thus enabling the transmission of second information to the network device when the change in this second characteristic is significant.

[0168] The aforementioned channel aging degree, the first coefficient, the first vector, and the third vector satisfy the following relationship: ;in, For channel aging, As the first coefficient, It is a positive real number. Let be the first vector. It is the third vector. The value of is positively correlated with the feedback accuracy of the first vector; the higher the feedback accuracy of the first vector, the more accurate the first vector obtained by the network device. This relationship can accurately quantify the degree of environmental change around the first terminal device, enabling the first terminal device to accurately obtain the channel aging degree based on this relationship. Furthermore, through this method, the accuracy requirements of the first vector fed back by the network device to the first terminal device can be dynamically adjusted. The value of allows the channel aging degree to be applied to more scenarios with different requirements.

[0169] In another possible implementation, the channel aging degree can be determined based on parameter #1 corresponding to channel #1 and parameter #2 corresponding to channel #2. Parameters #1 and #2 can be RSRP, or they can be RSRQ. For example, if both parameters #1 and #2 are RSRP, the channel aging degree can be the difference between parameters #1 and #2.

[0170] The second information can be sent together with the first information, such as through the same message, or it can be sent separately from the first information, such as through different messages. This application does not limit the method of sending the second information.

[0171] In this embodiment, when the channel aging degree is greater than or equal to a preset threshold, the first terminal device sends second information to the network device, which enables the network device to use the first vector to obtain the first channel tensor, thereby ensuring the accuracy of the first channel tensor obtained by the network device.

[0172] Furthermore, the first terminal device can also refrain from sending the second information if the channel aging degree is less than a preset threshold. A channel aging degree less than the preset threshold indicates that channel #1 has not changed significantly compared to channel #2, meaning the difference between the first vector and the third vector is small. In this case, the network device can obtain the first channel tensor based on the third vector; that is, the target first vector is the third vector (described below). This reduces the communication overhead of the first terminal device and conserves communication resources.

[0173] The above describes the specific operation of the first terminal device. After the first terminal device sends the first information, the network device can first obtain the reconstructed channel feature sequence based on the first sequence and the target first vector; then, it can perform a decompression operation on the reconstructed channel feature sequence to obtain the first channel tensor. That is, the first information includes the first sequence, and the first channel tensor is determined based on the first information and can specifically include: the first channel tensor is obtained by decompressing the reconstructed channel feature sequence, and the reconstructed channel feature sequence is determined based on the first sequence and the target first vector.

[0174] The first sequence is used to indicate a first feature of the first channel, and the first feature indicated by the first sequence changes more than the second feature changes more than the second feature over time. The first sequence is determined based on the second channel tensor, as detailed in the preceding introduction, and will not be repeated here.

[0175] The target first vector is used to indicate the second characteristic of the first channel. The target first vector can be understood as the vector most recently received by the network device from the first terminal device, indicating the second characteristic of the first channel; that is, the target first vector can be either the first vector or the third vector mentioned above. The target first vector is determined based on either the second channel tensor or the third channel tensor. The third channel tensor is obtained by the first terminal device from measuring the first channel before the second channel tensor. In other words, the second and third channel tensors are determined based on the first channel at different time periods. For details, please refer to the aforementioned related introduction, which will not be repeated here. When the target first vector is determined based on the second channel tensor, the target first vector is the first vector mentioned above; when the target first vector is determined based on the third channel tensor, the target first vector is the third vector mentioned above.

[0176] In this embodiment, the network device obtains a reconstructed channel feature sequence based on a first sequence and a target first vector, and decompresses the reconstructed channel feature sequence. After receiving the first information from the first terminal device, it can promptly obtain the first channel tensor, thereby facilitating the execution of subsequent operations based on the channel tensor, such as obtaining the target channel tensor.

[0177] The network device obtains the reconstructed channel feature sequence based on the first sequence and the target first vector, which may specifically include: the network device obtains the reconstructed residual sequence based on the first sequence; obtains the reconstructed stable feature sequence based on the target first vector; and adds the reconstructed residual sequence and the reconstructed stable feature sequence to obtain the reconstructed channel feature sequence.

[0178] For example, a network device can input a first sequence and a target first vector into a dual-stream decoupled decoder to obtain a reconstructed channel feature sequence. The following describes different scenarios.

[0179] Regarding the aforementioned situation 1.1, such as Figure 6 As shown, the dual-stream decoupling decoder deployed in the network device includes a fully connected layer, a stable feature pass-through layer, and an overlay fusion layer. This dual-stream decoupling decoder matches the dual-stream decoupling autoencoder in Case 1.1 above.

[0180] The network device inputs the first sequence and the target first vector into the dual-stream decoupling decoder. The fully connected layer in the dual-stream decoupling decoder performs a decoding operation on the first sequence to obtain the reconstructed residual sequence; that is, the reconstructed residual sequence is obtained by decoding the first sequence. This reconstructed residual sequence includes T feature vectors, where each feature vector has a dimension of L. f The reconstructed residual sequence can be represented as: , .

[0181] The stable feature pass-through layer in the dual-stream decoupled decoder can extend and copy the target first vector in the time dimension to obtain a reconstructed stable feature sequence. For example, the stable feature pass-through layer copies the target first vector into T copies, obtaining T feature vectors, and uses these T feature vectors to form a sequence as the reconstructed stable feature sequence. That is, the reconstructed stable feature sequence is determined based on the target first vector, and the stable feature pass-through layer is used to copy the target first vector to obtain the reconstructed stable feature sequence. The length of the reconstructed stable feature sequence is the same as the length of the reconstructed residual sequence, that is, the length of both the reconstructed stable feature sequence and the reconstructed residual sequence is T. The reconstructed stable feature sequence can be represented as: , , Let the first vector be the target. The vector dimension is L f In the embodiments of this application, the operation of the dual-stream decoupled decoder to acquire the reconstructed stable feature sequence and the operation of the dual-stream decoupled decoder to acquire the reconstructed residual sequence can be performed simultaneously or sequentially in time. This application does not impose any restrictions on this.

[0182] After the dual-stream decoupled decoder acquires the reconstructed residual sequence and the reconstructed stable feature sequence, the superposition and fusion layer in the dual-stream decoupled decoder can perform time-step vector addition of the reconstructed residual sequence and the reconstructed stable feature sequence to obtain the reconstructed channel feature sequence. This reconstructed channel feature sequence can be represented as: , .

[0183] Regarding the aforementioned situation 1.2, such as Figure 7 As shown, the dual-stream decoupling decoder deployed in the network device includes a decoding branch, a stable feature pass-through layer, and an overlay fusion layer. The decoding branch includes a zero-padding layer and a fully connected layer. This dual-stream decoupling decoder matches the dual-stream decoupling autoencoder in Case 1.2 above.

[0184] The network device inputs the first sequence and the target first vector into the dual-stream decoupling decoder. The vector dimension of the feature vector in the first sequence takes the value (L...). D Less than L d At that time, the zero-padding layer in the dual-stream decoupling decoder performs a dimensionality increase operation on the first sequence, such as performing a zero-padding dimensionality increase operation on the vector dimension of the first sequence, so that the vector dimensions of the T feature vectors in the first sequence are all padded to L. d The dimension is calculated to obtain the first sequence after dimensionality increase. The vector dimension of the feature vectors in the first sequence takes values ​​(L...). D ) equals L d In this case, the zero-fill layer may not process the first sequence, but instead transmit the first sequence to the fully connected layer of the dual-stream decoupled decoder for processing.

[0185] The fully connected layer in the dual-stream decoupling decoder can process the first sequence (vector dimension L). d The decoding operation is performed to obtain the reconstructed residual sequence, which includes T feature vectors, each with a vector dimension of L. f The stable feature pass-through layer in the two-stream decoupled decoder can extend and copy the target first vector in the time dimension to obtain the reconstructed stable feature sequence. After the two-stream decoupled decoder obtains the reconstructed residual sequence and the reconstructed stable feature sequence, the superposition and fusion layer in the two-stream decoupled decoder can perform time-step vector addition of the reconstructed residual sequence and the reconstructed stable feature sequence to obtain the reconstructed channel feature sequence.

[0186] It is understandable that the specific implementation of obtaining the reconstructed channel feature sequence in Case 1.2 above is similar to the specific implementation of obtaining the reconstructed channel feature sequence in Case 1.1 above. The difference is that in Case 1.2, the zero-padding layer may perform a dimensionality increase operation on the input first sequence, and the data input to the fully connected layer comes from this zero-padding layer; in Case 1.1, the dual-stream decoupling decoder does not set a zero-padding layer, that is, the dual-stream decoupling decoder does not perform a dimensionality increase operation on the first sequence. The similarities can be understood by referring to each other, and will not be elaborated here.

[0187] In this embodiment, the reconstructed channel feature sequence is obtained by adding the reconstructed residual sequence and the reconstructed stable feature sequence; the reconstructed residual sequence is obtained by decoding the first sequence after performing a dimensionality increase operation, or the reconstructed residual sequence is obtained by decoding the first sequence; the reconstructed stable feature sequence is determined based on the target first vector, and the length of the reconstructed stable feature sequence is the same as the length of the reconstructed residual sequence. Through the above method, the reconstructed channel feature sequence can be accurately obtained based on the first sequence and the target first vector, thereby facilitating the network device to obtain the first channel tensor based on the reconstructed channel feature sequence.

[0188] The way the network device performs decompression on the reconstructed channel feature sequence is related to the way the first terminal device compresses the second channel tensor.

[0189] When the first terminal device compresses the second channel tensor using a pre-set channel tensor encoder, the network device can perform a decompression operation on the reconstructed channel feature sequence using a channel tensor decoder matched with the encoder. For example, when the first terminal device compresses the second channel tensor using encoder #2, the network device can input the reconstructed channel feature sequence into decoder #2 matched with the encoder for decompression, obtaining the first channel tensor output by decoder #2.

[0190] When the first terminal device sequentially performs frequency domain compression and spatial domain compression on the second channel tensor, the network device can first perform spatial domain decompression on the reconstructed channel feature sequence, and then perform frequency domain decompression to obtain the first channel tensor. In other words, the aforementioned first channel tensor obtained by decompressing the reconstructed channel feature sequence can specifically include: the first channel tensor is obtained by first performing spatial domain decompression on the reconstructed channel feature sequence, and then performing frequency domain decompression.

[0191] For example, the network device can input each feature vector in the reconstructed channel feature sequence to the aforementioned spatial domain decoder to obtain a second intermediate tensor output by the spatial domain decoder. After obtaining the second intermediate tensor, the network device can input the second intermediate tensor to the aforementioned frequency domain decoder to obtain multiple complete spatial slice channel tensors; after combining these multiple complete spatial slice channel tensors, a first channel tensor is obtained.

[0192] In this embodiment, the spatial and frequency domain decompression processes sequentially performed by the network device on the reconstructed channel feature sequence can be understood as the reverse process of the first terminal device performing frequency and spatial domain compression on the second channel tensor. The similarities can be understood by referring to each other, and will not be repeated here. Furthermore, the spatial decoder and the frequency decoder can be referred to the aforementioned related descriptions, and will not be repeated here. By first performing spatial decompression processing on the reconstructed channel feature sequence and then performing frequency domain decompression processing, the network device can obtain the first channel tensor.

[0193] After obtaining the first channel tensor, the network device can obtain the target channel tensor based on the first channel tensor and the first historical reference channel tensor. For details, please refer to the relevant description in S202 above, which will not be repeated here. Furthermore, in situation 1.2 above, after obtaining the target channel tensor, the network device can also send third information to the first terminal device based on the target channel tensor, so that the first terminal device can adjust the fourth information in a timely manner based on the third information. This will be explained in detail below.

[0194] After the network device receives the first information from the first terminal device, the above method may further include: the network device sending third information to the first terminal device based on channel reconstruction error information, and correspondingly, the first terminal device receiving the third information from the network device, wherein the third information is used to adjust or update the fourth information, the fourth information being used to indicate the degree of vector dimension truncation processing performed on the first full sequence, the first full sequence being determined based on the second channel tensor, and the first sequence being the sequence obtained after performing vector dimension truncation processing on the first full sequence.

[0195] Channel reconstruction error information is used to indicate the difference between the channel represented by the target channel tensor and the current first channel. The channel represented by the target channel tensor can be understood as the wireless channel reconstructed by the network device based on the target channel tensor. The current first channel can be understood as the first channel actually measured by the first terminal device.

[0196] The aforementioned channel reconstruction error information is determined based on the target channel tensor and CQI.

[0197] For example, the network device can obtain the predicted SINR based on the target channel tensor; and the network device can obtain the actual SINR of the first terminal device based on the latest CQI reported by the first terminal device. After the network device obtains the predicted SINR and the actual SINR, it can calculate the channel reconstruction error information by the ratio of a first value to the actual SINR, where the first value is the difference between the predicted SINR and the actual SINR. At this time, the channel reconstruction error information can be expressed as: ;in, For channel reconstruction error information, To predict SINR, This is the actual SINR.

[0198] After the network device obtains the channel reconstruction error information, it can use the value of this channel reconstruction error information and the minimum channel reconstruction error information as a basis. Maximum channel reconstruction error information Relationship, determine the third information .For example, and , , The relationship can be: In this case, the third information With the fourth information The relationship can be: , This is the updated fourth piece of information. (The above) and The relationship can be pre-set in the first terminal device, sent to the first terminal device by the network device, or pre-defined by the protocol; this application does not limit this. The first terminal device receives... After that, you can follow this and The relationship, on Make adjustments or updates to obtain Furthermore, the first terminal device may, in the next acquisition of the sequence used to indicate the first feature of the first channel, use the sequence in the next cycle when acquiring the first sequence. Perform vector dimension truncation on the first full sequence.

[0199] By determining the channel reconstruction error information using the target channel tensor and CQI, the channel reconstruction error information can accurately reflect the difference between the wireless channel reconstructed by the network device based on the target channel tensor and the first channel actually measured by the first terminal device. This makes it easier for the network device to determine the third information based on the reconstruction error information.

[0200] The third information may be sent by the network device to the first terminal device through higher-layer signaling (such as radio resource control (RRC) signaling, medium access control control element (MAC CE)) or dedicated physical layer control signaling, and this application does not limit this.

[0201] The fourth information can be found in the aforementioned descriptions and will not be repeated here. The fourth information, updated based on the third information, can be used to indicate the degree to which the first terminal device performs vector dimension truncation processing on the first full sequence after it acquires it again.

[0202] In this embodiment, the greater the difference between the channel represented by the target channel tensor and the current first channel, the greater the difference between the channel reconstructed by the network device for sending downlink data to the first terminal device and the first channel actually measured by the first terminal device. In this case, the network device can adjust the fourth information with the third information so that the first terminal device can report a first sequence with more vector dimensions next time. Based on this first sequence, the channel reconstructed by the network device in the future is closer to the actual channel, thereby improving the accuracy of subsequent downlink data transmission.

[0203] Furthermore, the network device may send the third information before sending downlink data to the first terminal device, or it may send the third information after sending downlink data to the first terminal device; this application does not limit this.

[0204] Optionally, in conjunction with the above embodiments, there are various ways for the network device to obtain the first historical reference channel tensor, such as determining the first historical reference channel tensor based on the location information of the first terminal device, or determining the first historical reference channel tensor based on first information. These will be described separately below.

[0205] Method 1: Determine the first historical reference channel tensor based on the location information of the first terminal device.

[0206] In Method 1, the network device pre-configures at least one historical reference channel tensor corresponding to each of multiple locations, all of which are within the beam coverage area of ​​the network device. After receiving the first information sent by the first terminal device, the network device can calculate the spatial distance between the latest location information reported by the first terminal device and the multiple locations, determine the second location closest to the first terminal device among the multiple locations, and determine at least one historical reference channel tensor corresponding to the second location as the first historical reference channel tensor.

[0207] At least one historical reference channel tensor corresponding to each of the above locations can be obtained by measurement by the terminal device.

[0208] At least one historical reference channel tensor associated with each of the aforementioned locations can be obtained by performing clustering processing on the set of historical reference channel tensors according to spatial correlation. The set of historical reference channel tensors includes a historical reference channel tensor corresponding to each location in different locations, and these different locations are within the beam coverage area of ​​the network device.

[0209] For example, before pre-setting at least one historical reference channel tensor corresponding to each location in multiple locations in the network device, the first device can collect channel tensors corresponding to different locations within the beam coverage area of ​​the network device (denoted as historical reference channel tensor set #1). These channel tensors corresponding to different locations can be understood as the channel tensors obtained by estimating the wireless channel between the terminal device and the network device at different locations within the beam coverage area of ​​the network device. After obtaining the historical reference channel tensor set #1, the first device can use a clustering algorithm (such as K-means clustering) to perform clustering operations on the data in the historical reference channel tensor set #1 according to spatial correlation, obtaining at least one historical reference channel tensor corresponding to each location in the aforementioned multiple locations. That is, these multiple locations can be understood as the center positions of each cluster after clustering operations on different locations in the historical reference channel tensor set #1. After obtaining at least one historical reference channel tensor corresponding to each location in multiple locations, the first device can pre-set these at least one historical reference channel tensor corresponding to each location in the multiple locations in the network device. The aforementioned first device can be a network device or other electronic device, and this application does not limit it. Furthermore, at least one historical reference channel tensor corresponding to each of the aforementioned locations can be pre-configured in the network device in the form of a database.

[0210] In this embodiment, the first historical reference channel tensor corresponds to a second location, which is the location closest to the first terminal device among a plurality of locations. Each of these locations corresponds to at least one historical reference channel tensor, and all of these locations are within the beam coverage area of ​​the network device. That is, the first historical reference channel tensor includes at least one historical reference channel tensor corresponding to the second location. The second location is closest to the first terminal device, making the historical reference channel tensor corresponding to the second location more similar to the second channel tensor measured by the first terminal device compared to the historical reference channel tensors corresponding to other locations (other locations among the plurality of locations besides the second location). In other words, the channel characteristics of the wireless channel associated with the historical reference channel tensor corresponding to the second location are more similar to the channel characteristics of the first channel. In this application's embodiments, "similarity" is a relative concept. For example, the historical reference channel tensor corresponding to the second location is more similar to the second channel tensor (i.e., high similarity), while the historical reference channel tensor corresponding to other locations has low similarity to the second channel tensor. Similarly, the channel characteristics of the wireless channel associated with the historical reference channel tensor corresponding to the second location are more similar to the channel characteristics of the first channel (i.e., high similarity), while the channel characteristics of the wireless channel associated with the historical reference channel tensor corresponding to other locations have low similarity to the channel characteristics of the first channel.

[0211] The above method enables network devices to compensate and calibrate the first channel tensor based on the first historical reference channel tensor, thereby improving the accuracy of the target channel tensor obtained by the network devices.

[0212] Method 2: Determine the first historical reference channel tensor based on the location of the first terminal device and the reconstructed channel feature sequence.

[0213] In Method 2, the network device pre-configures at least one historical reference channel tensor corresponding to each of multiple locations. These multiple locations are within the beam coverage area of ​​the network device. Each channel tensor in the at least one historical reference channel tensor corresponds to a historical channel feature sequence, and the historical channel feature sequence corresponding to the historical reference channel tensor can be a sequence obtained by sequentially performing frequency domain compression and spatial domain compression on the historical reference channel tensor. The specific implementation of determining the historical channel feature sequence based on the historical reference channel tensor is similar to the aforementioned specific implementation of determining the original channel feature sequence based on the second channel tensor, the difference being that the two specific implementations use different channel tensors. The similarities can be understood by referring to each other, and will not be elaborated here.

[0214] At least one historical reference channel tensor corresponding to each of the above locations was obtained by the terminal device.

[0215] At least one historical reference channel tensor associated with each of the aforementioned locations can be obtained by performing clustering processing on the set of historical reference channel tensors according to spatial correlation. The set of historical reference channel tensors includes a historical reference channel tensor corresponding to each location in different locations, and a historical channel feature sequence corresponding to the historical reference channel tensor. These different locations are located within the beam coverage area of ​​the network device.

[0216] For example, before pre-setting at least one historical reference channel tensor corresponding to each location in multiple locations in the network device, the first device can collect the channel tensors corresponding to different locations within the beam coverage area of ​​the network device (denoted as historical reference channel tensor set #2) and obtain the channel feature sequence corresponding to the channel tensor. The channel tensors corresponding to different locations can be understood as the channel tensors obtained after performing channel measurement (or estimation) on the wireless channel between the terminal device and the network device at different locations within the beam coverage area of ​​the network device.

[0217] After acquiring the historical reference channel tensor set #2, the first device can use a clustering algorithm (such as K-means clustering) to perform clustering operations on the data in the historical reference channel tensor set #2 according to spatial correlation, thereby obtaining at least one historical reference channel tensor corresponding to each of the aforementioned multiple locations. This clustering process can be represented as: ;in, This represents the historical channel database space that has not undergone clustering (i.e., the historical reference channel tensor set #2). This space contains K samples, each corresponding to a channel tensor at a different location and a corresponding channel feature sequence. Indicates clustered There are 1 historical channel database subspace, and each historical channel database subspace has 1 Each sample is a subspace of the historical channel database. After partitioning the historical channel database subspace as described above, for each sample in the historical channel database space, its cluster ID and the center position of each cluster (denoted as...) can be recorded. The sample contains the channel tensor and the channel feature sequence of the sample. Through the above operations, different samples can be divided into different clusters, and the center of each cluster corresponds to the above-mentioned multiple locations; in other words, through the above operations, at least one historical reference channel tensor corresponding to each of the above-mentioned multiple locations can be obtained, as well as the historical channel feature sequence corresponding to the historical reference channel tensor.

[0218] After receiving the first information sent by the first terminal device, the network device can calculate the spatial distance between the latest location information reported by the first terminal device and multiple locations, determine the first location closest to the first terminal device among the multiple locations, and determine at least one historical reference channel tensor corresponding to the first location as the second historical reference channel tensor. That is, the first historical reference channel tensor corresponds to the first location, and the second historical reference channel tensor corresponds to the first location; the first location is the location closest to the first terminal device among the multiple locations, and each of the multiple locations corresponds to at least one historical reference channel tensor.

[0219] Continuing with the example above, the location coordinates of the first terminal device are: The coordinates of the center location of the m-th cluster (i.e., the multiple locations mentioned above) are: By selecting the cluster center with the smallest spatial distance, the target subspace ID (i.e., cluster ID) to which the first terminal device belongs is determined. This target subspace ID is denoted as... , The positioning process can be represented as: ;in, For the ID index of the target subspace, The physical location coordinates of the current user equipment. The coordinates are the center coordinates of the m-th cluster in the historical database.

[0220] The aforementioned second historical reference channel tensor corresponds to the first historical channel feature sequence. This first historical channel feature sequence includes one or more historical channel feature sequences, and the number of channel feature sequences included in the first historical channel feature sequence is the same as the number of channel tensors included in the second historical reference channel tensor. Each channel tensor in the second historical reference channel tensor corresponds to a specific channel feature sequence in the historical channel feature sequence.

[0221] After determining the second historical reference channel tensor based on the location of the first terminal device, the network device can perform similarity matching (such as cosine similarity matching) between the reconstructed channel feature sequence and the first historical channel feature sequence corresponding to the second historical reference channel tensor. This identifies the top N historical channel feature sequences with the highest similarity among the first historical channel feature sequences, and the historical reference channel tensors corresponding to these N historical channel feature sequences are defined as the first historical reference channel tensor, where N is a positive integer. In other words, the first historical reference channel tensor includes historical reference channel tensors associated with N historical channel feature sequences. These N historical channel feature sequences are the top N most similar historical channel feature sequences obtained by performing similarity calculations between the reconstructed channel feature sequence and each channel feature sequence in the historical channel feature sequence.

[0222] Continuing with the example above, the reconstructed channel feature sequence is denoted as... ID index is The target subspace (i.e., the first historical channel feature sequence) includes the historical channel feature sequence denoted as Calculate the reconstructed channel feature sequences respectively The target subspace includes historical channel feature sequences. The cosine similarity between them. The formula for calculating the cosine similarity can be expressed as: After the calculation is completed, the historical channel feature sequences included in the target subspace are sorted in descending order according to their cosine similarity scores, and the channel tensors corresponding to the top N historical channel feature sequences with the highest similarity scores are obtained as the first historical reference channel tensor.

[0223] In this embodiment, the first historical reference channel tensor is obtained based on the reconstructed channel feature sequence, the second historical reference channel tensor, and the first historical channel feature sequence. The first historical reference channel tensor corresponds to a first location, which is the location closest to the first terminal device among multiple locations. Each location among the multiple locations corresponds to at least one historical reference channel tensor, and the multiple locations are within the beam coverage area of ​​the network device. Each channel tensor in the second historical reference channel tensor corresponds one-to-one with a historical channel feature sequence in the first historical channel feature sequence. The first historical reference channel tensor includes N historical reference channel tensors associated with historical channel feature sequences. The N historical channel feature sequences are the top N most similar channel feature sequences (in the first historical channel feature sequence) obtained by performing similarity calculations between the reconstructed channel feature sequence and each channel feature sequence in the first historical channel feature sequence, where N is a positive integer.

[0224] The first location corresponding to the second historical reference channel tensor is closest to the location of the first terminal device, and the N historical channel feature sequences corresponding to the first historical reference channel tensor are the channel feature sequences most similar to the reconstructed channel feature sequences among the first historical channel feature sequences. This makes the first historical reference channel tensor more similar to the second channel tensor measured by the first terminal device compared to other historical reference channel tensors #1. In other words, the channel features of the wireless channel associated with the first historical reference channel tensor are more similar to the channel features of the first channel. The other historical reference channel tensors #1 can be understood as: the historical reference channel tensors corresponding to other locations besides the first location among multiple locations, and the other historical reference channel tensors besides the first historical tensor in the second historical tensor. In the embodiments of this application, "similarity" is a relative concept. For example, the first historical reference channel tensor is more similar to the second channel tensor (i.e., the similarity is high), while the other historical reference channel tensor #1 has a low similarity to the second channel tensor. Similarly, the channel characteristics of the wireless channel associated with the first historical reference channel tensor are more similar to the channel characteristics of the first channel (i.e., the similarity is high), while the channel characteristics of the wireless channel associated with the other historical reference channel tensor #1 have a low similarity to the channel characteristics of the first channel.

[0225] By employing the above method, the similarity between the first historical reference channel tensor and the second channel tensor obtained by the network device can be guaranteed. That is, the first historical reference channel tensor and the second channel tensor are similar or close. This allows the network device to significantly improve the accuracy of the target channel tensor obtained by the network device when it compensates and calibrates the first channel tensor based on the first historical reference channel.

[0226] Method 3: Determine the first historical reference channel tensor based on the reconstructed channel feature sequence.

[0227] In Method 3, the network device pre-configures a correspondence between multiple historical channel feature sequences and multiple historical reference channel tensors, with each historical channel feature sequence corresponding one-to-one with the other. These historical reference channel tensors can be channel tensors measured by the terminal device at different locations within the beam coverage area of ​​the network device. The specific implementation of determining historical channel feature sequences based on historical reference channel tensors is similar to the aforementioned implementation of determining original channel feature sequences based on the second channel tensor, except that the two implementations use different channel tensors. The similarities can be understood by referring to each other, and will not be elaborated further here.

[0228] After acquiring the reconstructed channel feature sequence, the network device can perform similarity matching (such as cosine similarity matching) with multiple historical channel feature sequences to obtain the Y most similar reconstructed channel feature sequences. The Y historical reference channel tensors corresponding to these Y reconstructed channel feature sequences are then determined as the first historical reference channel tensor. This method improves the efficiency of the network device in acquiring the first historical reference channel tensor.

[0229] It can be understood that the above communication method can be interpreted as the interaction between the first terminal device and the network device within a cycle. The above method can also be executed periodically. For example, after the network device sends downlink data to the first terminal device based on the first information, the first terminal device can reacquire the first information and send the first information to the network device. After receiving the first information, the network device sends downlink data to the first terminal device based on the first information.

[0230] The above, in conjunction with the method embodiments, provides an overall overview of the communication method provided in this application. For ease of understanding, the following will be combined with... Figure 8 and Figure 9 The above method will be introduced through specific scenarios. This method is applicable to... Figure 1 The communication system shown mainly involves the interaction between the UE (the aforementioned first terminal device) and the base station (the aforementioned network device). In this scenario, the UE obtains a rapidly changing feature sequence (i.e., the aforementioned first feature) and a stable feature vector (i.e., the aforementioned second feature) based on a first channel, and sends the rapidly changing feature sequence and the stable feature vector to the base station; the base station obtains the current channel tensor based on the rapidly changing feature sequence and the stable feature vector, and obtains the target channel tensor based on the current channel tensor and the first historical reference channel tensor, and finally sends downlink data to the UE based on the target channel tensor.

[0231] like Figure 8 As shown, the communication method includes the following steps: S801, the base station sends CSI-RS to the UE. Correspondingly, the UE receives the CSI-RS from the base station.

[0232] For details on CSI-RS, please refer to the aforementioned introduction; further explanation is not required here.

[0233] S802, the UE performs channel estimation on the first channel based on CSI-RS to obtain the original channel tensor.

[0234] The original channel tensor corresponds to the second channel tensor in the aforementioned embodiment.

[0235] The specific implementation of S802 described above can refer to existing technologies. The above S801 and S802 correspond to... Figure 9 Operation of the mid-channel estimation module.

[0236] S803, the UE performs frequency domain compression and spatial domain compression on the original channel tensor in sequence to obtain the original channel feature sequence.

[0237] The UE can first obtain the complete spatial slice channel tensor at T time steps based on the original channel tensor; then input these complete spatial slice channel tensors at T time steps into the frequency domain decoder to obtain T first intermediate tensors; next, input these T first intermediate tensors into the spatial domain decoder to obtain T feature vectors. Arrange these T feature vectors in chronological order to obtain the original channel feature sequence.

[0238] The above S803 corresponds to Figure 9 The operation of the feature extraction module of the two-stage autoregressive encoder.

[0239] S804, the UE inputs the original channel feature sequence and code rate indicator into the dual-stream decoupled autoencoder to obtain the fast-changing feature sequence and stable feature vector.

[0240] The rate indicator corresponds to the fourth information in the aforementioned embodiments. The fast-changing feature sequence corresponds to the first sequence in the aforementioned embodiments; and this fast-changing feature sequence is obtained by truncating the vector dimension of the first full sequence generated in the two-stream decoupled autoencoder. The stable feature vector corresponds to the first vector in the aforementioned embodiments.

[0241] The above S804 corresponds to Figure 9 The operation of the stable fast-changing component decoupling module. Furthermore, this stable fast-changing component decoupling module can be the aforementioned dual-stream decoupling self-encoder.

[0242] In step S805, the UE sends a rapidly changing feature sequence and a stable feature vector to the base station. Correspondingly, the base station receives the rapidly changing feature sequence and stable feature vector from the UE.

[0243] The UE can calculate the channel aging degree based on the stable feature vector, and when the channel aging degree is greater than or equal to a threshold, it sends a rapidly changing feature sequence and a stable feature vector to the base station. The rapidly changing feature sequence and the stable feature vector can be sent through the same message.

[0244] In addition, before transmitting the fast-changing feature sequence and stable feature vector, the UE can use scalar quantization or vector quantization algorithms to map the floating-point values ​​corresponding to the fast-changing feature sequence and stable feature vector to a finite discrete ladder, and then encode them into a binary bit stream before transmitting them.

[0245] The above S805 corresponds to Figure 9 Operation of the mid-channel information quantization module.

[0246] S806, the base station performs zero-padding up-dimensionality operation on the fast-changing feature sequence, and inputs the fast-changing feature sequence after vector dimension padding and the stable feature vector into the dual-stream decoupled decoder to obtain the reconstructed residual sequence.

[0247] After receiving the binary bitstream used to characterize the rapidly changing feature sequence and stable feature vector, the base station performs a reverse quantization mapping operation, that is, according to the preset quantization interval and codebook, it restores these discrete bitstreams to continuous floating-point feature vectors. This operation corresponds to... Figure 9 Operation of the mid-channel information inverse quantization module.

[0248] After obtaining the aforementioned continuous floating-point feature vectors, the base station performs zero-padding dimensionality-up operations on the rapidly changing feature sequence until the vector dimension of the rapidly changing feature sequence is restored to the preset complete vector dimension, that is, the vector dimension corresponding to the aforementioned first full sequence.

[0249] S807, the base station performs a copying and expansion operation on the stable feature vector to obtain a reconstructed stable feature sequence.

[0250] The base station copies the stable feature vector T times, resulting in T stable feature vectors. These T stable feature vectors form the reconstructed stable feature sequence.

[0251] The above S806 and S807 can be performed simultaneously or in a specific order, without restriction.

[0252] S808, the base station adds the reconstructed residual sequence and the reconstructed stable feature sequence to obtain the reconstructed channel feature sequence.

[0253] The above S806 to S808 correspond to Figure 9 Operation of the medium-stability fast-changing component coupling module.

[0254] S809, the base station sequentially performs spatial domain decompression and frequency domain decompression on the reconstructed channel feature sequence to obtain the current channel tensor.

[0255] The base station can first input each feature vector in the reconstructed channel feature sequence into the spatial domain decoder to obtain T second intermediate tensors; then input these T second intermediate tensors into the frequency domain decoder to obtain multiple channel tensors; finally, restore these multiple channel tensors to the current channel tensor. This current channel tensor corresponds to the first channel tensor in the aforementioned embodiment.

[0256] The above S809 corresponds to Figure 9 The operation of the two-stage decoding module.

[0257] S8010, the base station obtains the second historical reference channel tensor based on the location information of the first terminal device.

[0258] The base station can obtain the second historical reference channel tensor based on a pre-set historical channel information database. This historical channel information database corresponds to the clustered database in the aforementioned embodiments. A historical channel database subspace.

[0259] S8011, the base station obtains the first historical reference channel tensor from the second historical reference channel tensor based on the reconstructed channel feature sequence.

[0260] The above S8010 to S8011 correspond to Figure 9 Operation of the two-stage historical channel filtering network.

[0261] S8012, the base station inputs the current channel tensor and the first historical reference channel tensor into the fusion network to obtain the target channel tensor.

[0262] The converged network is an MLP.

[0263] The above S8012 corresponds to Figure 9 Operation of the MLP channel information reconstruction module.

[0264] Furthermore, after obtaining the target channel tensor, the base station can obtain a rate adjustment indicator based on the target channel tensor and the latest CQI reported by the UE, and send the rate adjustment indicator to the UE. This rate adjustment indicator corresponds to the third information in the aforementioned embodiments.

[0265] S8013, the base station sends downlink data to the UE based on the target channel tensor. Correspondingly, the UE receives the downlink data from the base station.

[0266] In this embodiment, when the channel aging degree is less than a preset threshold, the UE sends a fast-changing feature sequence to the base station, meaning that the UE does not send a stable feature vector to the base station at this time. After receiving the fast-changing feature sequence from the UE, the base station can perform subsequent operations based on the fast-changing feature sequence and the stable feature vector #a. The stable feature vector #a can be understood as the latest stable feature vector received by the base station from the UE, and this stable feature vector #a corresponds to the third vector in the aforementioned embodiment. The base station's operation based on the fast-changing feature sequence and the stable feature vector #a is similar to the base station's operation based on the fast-changing feature sequence and the stable feature vector, except that the stable feature vector used by the base station is different. The similarities can be understood by referring to the aforementioned related descriptions, such as replacing the stable feature vector in S807 to S8014 with stable feature vector #2, which will not be elaborated here.

[0267] For the operations of obtaining the first historical reference channel tensor and obtaining the target channel through the fusion network, these two operations can be performed by setting up an artificial intelligence (AI) model. For example, the two-stage historical channel filtering network and the MLP channel information reconstruction module can be set as an AI channel reconstruction model, and the first historical reference channel tensor and the target channel tensor can be obtained by the AI ​​channel reconstruction model.

[0268] It is understood that the specific implementations of S801 to S8013 mentioned above can refer to the foregoing. Figure 2 The relevant descriptions in the illustrated embodiments will not be repeated here.

[0269] Furthermore, the above methods are applicable to scenarios involving high UE mobility and sudden changes in the UE's environment. For example, ... Figure 10 As shown, the UE is vehicle #1. When vehicle #1 enters the tunnel from an open road, the environment around vehicle #1 undergoes a drastic change. In existing technologies, satellite positioning is typically used to obtain the location information of vehicle #1, and the beam for sending information to vehicle #1 is determined based on this location information. However, when vehicle #1 enters the tunnel, the satellite positioning signal is instantly lost or experiences severe drift. At this time, the base station cannot obtain the location information of vehicle #1, and the prediction algorithm based on the location information of vehicle #1 is directly paralyzed, unable to preload any information. That is, the base station cannot obtain the precoding matrix and therefore cannot send data to vehicle #1. In the above embodiment, the dual-stream decoupled autoencoder in vehicle #1 can use an attention mechanism to lock onto the new environmental background (i.e., the environment when vehicle #1 enters the tunnel) and extract stable feature vectors, thereby avoiding the loss of environmental features of vehicle #1 and thus avoiding the situation where data cannot be sent to vehicle #1 due to the inability to obtain the precoding matrix. It is understood that the specific interaction process between vehicle #1 and the base station can be understood by referring to the relevant descriptions in the above embodiments, and will not be repeated here.

[0270] Figure 11 This is a schematic diagram of a model training method provided in an embodiment of this application. This model training method can be used to train a fusion network. Furthermore, this model training method can be executed by a first device, which can be a server, a computer, or the like.

[0271] like Figure 11 As shown, the above model training method includes the following steps: S1101, the first device acquires sample data and tag data.

[0272] The sample data includes a first channel tensor and at least one historical reference channel tensor.

[0273] The first channel tensor is the channel tensor obtained by the network device based on the first information sent by the terminal device. The method for obtaining this first channel tensor is the same as described above. Figure 2 The method of obtaining the first channel tensor in the illustrated embodiment is similar, except that the terminal device here may be the same as or different from the aforementioned first terminal device. Other similarities can be understood by referring to the aforementioned related introduction, and will not be repeated here.

[0274] At least one historical reference channel tensor is a channel tensor similar to the second channel tensor (the channel tensor measured by the terminal device), such as at least one historical reference channel tensor corresponding to the location closest to the aforementioned terminal device. The method for determining at least one historical reference channel tensor is the same as described above. Figure 2 The method for determining the first historical reference channel tensor in the illustrated embodiment is similar, except that: here, the location of the terminal device is used to obtain at least one historical reference channel tensor, while the location of the first terminal device is used to obtain at least one historical reference channel tensor. The terminal device here may be the same as or different from the first terminal device mentioned above. Other similarities can be understood by referring to the relevant descriptions above, and will not be repeated here.

[0275] The tag data is a second channel tensor. This second channel tensor is the channel tensor obtained by the terminal device from the measurement of the first channel, where the first channel is the wireless channel between the terminal device and the aforementioned network device. The method for obtaining the second channel tensor is the same as described above. Figure 2 The second channel tensor is obtained in a similar way in the embodiment shown. The difference is that the terminal device here may be the same as or different from the first terminal device mentioned above. Other similarities can be understood by referring to the relevant introduction above, and will not be repeated here.

[0276] S1102, the first device trains the fusion network based on sample data and label data.

[0277] The fusion network can be a neural network, such as an MLP, CNN, or RNN. The specific principles of MLP, CNN, and RNN can be found in existing technologies and will not be elaborated here.

[0278] In this embodiment, by inputting sample data into the fusion network, a target channel tensor output by the fusion network can be obtained. After outputting the target channel tensor, the fusion network can calculate the loss (error) between the target channel tensor and the aforementioned label data, and update its internal parameters according to the gradient of the loss. The target channel tensor can be referenced from the aforementioned... Figure 2 The relevant descriptions in the illustrated embodiments will not be repeated here.

[0279] The above describes the process of one training iteration of a fusion network. This process can be executed multiple times during the training of a fusion network, and in each iteration, different sample and label data can be used to adjust the internal parameters of the fusion network. The fusion network stops training when: the loss function converges to a preset range, meaning the output of the fusion network tends to stabilize; or, the number of training iterations reaches a preset number.

[0280] Furthermore, after the aforementioned fusion network is trained, it can be deployed in network devices, enabling the network devices to perform channel tensor fusion operations using the trained fusion network. The specific implementation of the network devices using the trained fusion network to perform channel tensor fusion operations can be found in the aforementioned... Figure 2 The relevant descriptions in the illustrated embodiments will not be repeated here.

[0281] It should be understood that Figures 2 to 11 The flowcharts or scene diagrams shown are for illustrative purposes only and are not intended to limit the embodiments of this application to the examples illustrated. In fact, those skilled in the art can interpret the embodiments based on... Figures 2 to 11 The examples in the document can be transformed into equivalent ways to obtain more implementations.

[0282] The above text combined Figures 2 to 11 The above-described methods provided in the embodiments of this application are described in detail below. The following will combine... Figures 12 to 13 The device embodiments of this application are described in detail below. It should be understood that the communication device of this application embodiment can execute the various methods of the foregoing embodiments of this application, that is, the specific working processes of the various products below can be referred to the corresponding processes in the foregoing method embodiments.

[0283] In the embodiments described above, the terminal device (such as the first terminal device mentioned above) can execute some or all of the steps in each embodiment; the network device can execute some or all of the steps in each embodiment. These steps or operations are merely examples, and the embodiments of this application can also perform other operations or variations of various operations. Furthermore, the steps can be executed in different orders as presented in the embodiments, and it is not necessary to execute all the operations in the embodiments of this application. Moreover, the sequence number of each step does not imply the order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0284] Figure 12 This is a schematic block diagram of a communication device provided in an embodiment of this application. Figure 12As shown, the communication device 1200 may include a communication module 1220. The communication module 1220 can implement corresponding communication functions, which can be internal communication functions of the communication device 1200 or communication functions between the communication device 1200 and other devices. Optionally, the communication module 1220 may also be referred to as a communication interface or transceiver module. Optionally, the communication device 1200 further includes a processing module 1210. The processing module 1210 can implement corresponding processing functions.

[0285] Optionally, the communication device 1200 further includes a storage module, which can be used to store instructions and / or data; the processing module 1210 can read the instructions and / or data in the storage module so that the communication device 1200 can implement the aforementioned method embodiments.

[0286] In one possible design, the communication device 1200 may correspond to the network device in the above method embodiments, or to a component (such as a circuit, chip, or chip system) configured in the network device. The communication device 1200 can be used to perform the steps or processes performed by the network device in any of the above method embodiments.

[0287] For example, the communication module 1220 is used to receive first information from the terminal device; wherein the first information is used to indicate a first characteristic of a first channel, and the first channel is a wireless channel between the first terminal device and the network device.

[0288] The processing module 1210 is used to instruct the communication module 1220 to send downlink data to the terminal device based on the target channel tensor; wherein, the target channel tensor is obtained by weighting the first channel tensor and the first historical reference channel tensor, or the target channel tensor is obtained by inputting the first channel tensor and the first historical reference channel tensor into a fusion network, which is used to perform nonlinear combination processing on the first channel tensor and the first historical reference channel tensor, and the fusion network is deployed in the network device; the first channel tensor is determined based on first information, and the channel characteristics of the wireless channel associated with the first historical reference channel tensor are similar to the channel characteristics of the first channel.

[0289] The above are merely examples; for detailed steps or procedures, please refer to the descriptions in the foregoing embodiments.

[0290] In one possible design, the communication device 1200 may correspond to the terminal device in the above method embodiments, or a component (such as a circuit, chip, or chip system) configured in the terminal device. The communication device 1200 may be used to execute the steps or processes performed by the terminal device in any of the above method embodiments.

[0291] For example, the processing module 1210 is used to instruct the communication module 1220 to send first information to the network device; wherein the first information is used to indicate a first characteristic of a first channel measured by the first terminal device, the first channel being a wireless channel between the first terminal device and the network device.

[0292] The communication module 1220 is used to receive downlink data from the network device; wherein the downlink data is obtained based on a target channel tensor; the target channel tensor is obtained by weighting a first channel tensor and a first historical reference channel tensor, or by inputting the first channel tensor and the first historical reference channel tensor into a fusion network, the fusion network being used to perform nonlinear combination processing on the first channel tensor and the first historical reference channel tensor, and the fusion network being deployed in the network device; the first channel tensor is determined based on first information, and the channel characteristics of the wireless channel associated with the first historical reference channel tensor are similar to the channel characteristics of the first channel.

[0293] The above are merely examples; for detailed steps or procedures, please refer to the descriptions in the foregoing embodiments.

[0294] Figure 13 This is another schematic block diagram of the communication device provided in the embodiments of this application. The communication device 1300 may be a chip, chip system, or processor, etc., in a terminal device or network device that implements the above-described method. The communication device 1300 can be used to implement the methods described in the above-described method embodiments; for details, please refer to the descriptions in the above-described method embodiments. It is understood that the communication device 1300 in the embodiments of this application and the communication device 1200 described above are two different communication devices in different embodiments, that is, communication device 1200 is... Figure 12 The communication device in the illustrated embodiment, communication device 1300 is Figure 13 The communication device in the illustrated embodiment.

[0295] like Figure 13 As shown, the communication device 1300 may include one or more processors 1310, which may also be referred to as processing units or processing modules, and can implement certain control functions. The processor 1310 may be a general-purpose processor or a dedicated processor, such as a baseband processor or a central processing unit. The baseband processor can be used to process communication protocols and communication data, while the central processing unit can be used to control the communication device 1300 (e.g., a base station, baseband chip, user, user chip), execute software programs, and process data from the software programs.

[0296] In an alternative design, the processor 1310 may also store instructions and / or data that can be executed by the processor 1310 to cause the communication device 1300 to perform the methods described in the above method embodiments.

[0297] In another alternative design, the communication device 1300 may include a communication interface 1320 for implementing receiving and transmitting functions. For example, the communication interface 1320 may be a transceiver circuit, interface, interface circuit, or transceiver. The transceiver circuit, interface, interface circuit, or transceiver for implementing receiving and transmitting functions may be separate or integrated. The aforementioned transceiver circuit, interface, interface circuit, or transceiver may be used for reading and writing code / data, or it may be used for transmitting or relaying signals.

[0298] Optionally, the communication device 1300 may include one or more memories 1330, which may store instructions that can be executed on the processor 1310, causing the communication device 1300 to perform the methods described in the above method embodiments. Optionally, the memories 1330 may also store data. Optionally, the processor 1310 may also store instructions and / or data. The processor 1310 and the memories 1330 may be provided separately or integrated together.

[0299] It should be understood that, in one possible design, the steps in the method embodiments provided in this application can be implemented by integrated logic circuits in the processor's hardware or by instructions in software form. The steps of the methods disclosed in the embodiments of this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are not provided here.

[0300] In one implementation, the communication device 1300 may correspond to the terminal device in the above method embodiments and may be used to execute the various steps and / or processes executed by the terminal device in the above method embodiments. The processor 1310 may be used to execute instructions stored in the memory 1330, and when the processor 1310 executes the instructions stored in the memory, the processor 1310 is used to execute the various steps and / or processes of the above method embodiments corresponding to the terminal device.

[0301] In another implementation, the communication device 1300 may correspond to the network device in the above method embodiments and may be used to execute the various steps and / or processes executed by the network device in the above method embodiments. The processor 1310 may be used to execute instructions stored in the memory 1330, and when the processor 1310 executes the instructions stored in the memory, the processor 1310 is used to execute the various steps and / or processes of the above method embodiments corresponding to the network device.

[0302] It should be understood that the aforementioned processing device can be one or more chips. For example, the processing device can be a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), a system-on-chip (SoC), a central processor unit (CPU), a network processor (NP), a digital signal processor (DSP), a microcontroller unit (MCU), a programmable logic device (PLD), or other integrated chips.

[0303] It is understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM). It should be noted that the memory used in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0304] According to the method provided in the embodiments of this application, this application also provides a chip system, which includes one or more processors for calling and executing instructions stored in memory, thereby causing the method described in the embodiments of this application to be executed. The chip system may be composed of chips or may include chips and other discrete devices.

[0305] The chip system may include input circuits or interfaces for transmitting information or data, and output circuits or interfaces for receiving information or data.

[0306] According to the method provided in the embodiments of this application, this application also provides a communication system, which includes the aforementioned network device and terminal device (such as the aforementioned first terminal device).

[0307] According to the method provided in the embodiments of this application, this application also provides a computer program product, which includes a computer program or instructions that, when the computer program is run (or executed), cause the method described in the embodiments of this application to be executed.

[0308] According to the method provided in the embodiments of this application, this application also provides a computer-readable storage medium storing a program or instructions, which, when executed, cause the method described in the embodiments of this application to be performed.

[0309] The computer-readable storage medium may be the aforementioned volatile memory or non-volatile memory, or it may include both volatile memory and non-volatile memory.

[0310] In the embodiments of this application, the terms and English abbreviations are exemplary examples given for ease of description and should not be construed as limiting the application in any way. This application does not preclude the possibility of defining other terms that can achieve the same or similar functions in existing or future agreements.

[0311] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When these computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated.

[0312] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0313] It should be understood that in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0314] In summary, the above description is merely a preferred embodiment of the technical solution of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A communication method, characterized in that, Applied to network devices, the method includes: Receive first information from a terminal device, the first information being used to indicate a first characteristic of a first channel, the first channel being a wireless channel between the terminal device and the network device; Downlink data is sent to the terminal device based on the target channel tensor; the target channel tensor is obtained by weighting the first channel tensor and the first historical reference channel tensor, or the target channel tensor is obtained by inputting the first channel tensor and the first historical reference channel tensor into a fusion network, the fusion network being used to perform nonlinear combination processing on the first channel tensor and the first historical reference channel tensor, the fusion network being deployed in the network device; the first channel tensor is determined based on the first information, and the channel characteristics of the wireless channel associated with the first historical reference channel tensor are similar to the channel characteristics of the first channel.

2. The method according to claim 1, characterized in that, In the fused network, the same shared weights are used for processing the antenna positions in the first channel tensor, the subcarrier positions in the first channel tensor, the antenna positions in the first historical reference channel tensor, and the subcarrier positions in the first historical reference channel tensor.

3. The method according to claim 1, characterized in that, The first information includes a first sequence, and the first channel tensor is determined based on the first information, including: The first channel tensor is obtained by decompressing the reconstructed channel feature sequence, which is determined based on the first sequence and the target first vector; Wherein, the target first vector is used to indicate the second feature of the first channel, and the degree to which the first feature indicated by the first sequence changes over time is greater than the degree to which the second feature changes over time; the first sequence is determined based on the second channel tensor, and the target first vector is determined based on the second channel tensor or the third channel tensor, and the second channel tensor and the third channel tensor are determined based on the first channel under different time periods.

4. The method according to claim 3, characterized in that, The reconstructed channel feature sequence is obtained by adding the reconstructed residual sequence and the reconstructed stable feature sequence. The reconstructed residual sequence is obtained by decoding the first sequence after performing a dimensionality increase operation, or the reconstructed residual sequence is obtained by decoding the first sequence. The reconstructed stable feature sequence is determined based on the target first vector, and the length of the reconstructed stable feature sequence is the same as the length of the reconstructed residual sequence.

5. The method according to claim 3, characterized in that, The first historical reference channel tensor is obtained based on the reconstructed channel feature sequence, the first historical channel feature sequence, and the second historical reference channel tensor; The first historical reference channel tensor corresponds to a first location, which is the location closest to the terminal device among multiple locations. Each of the multiple locations corresponds to at least one historical reference channel tensor, and the multiple locations are located within the beam coverage area of ​​the network device. Each channel tensor in the second historical reference channel tensor corresponds one-to-one with the historical channel feature sequence in the first historical channel feature sequence. The first historical reference channel tensor includes N historical reference channel tensors associated with historical channel feature sequences. The N historical channel feature sequences are the top N most similar channel feature sequences obtained by performing similarity calculations between the reconstructed channel feature sequence and each channel feature sequence in the first historical feature sequence, where N is a positive integer.

6. The method according to claim 3, characterized in that, Before sending downlink data to the terminal device based on the target channel tensor, the method further includes: The device receives second information from the terminal device, the second information including a first vector, the first vector being used to indicate the second feature, and the target first vector being the first vector.

7. The method according to claim 3, characterized in that, After receiving the first information from the terminal device, the method further includes: Based on the channel reconstruction error information, a third piece of information is sent to the terminal device. The channel reconstruction error information is used to indicate the difference between the channel represented by the target channel tensor and the current first channel. The third piece of information is used to adjust the fourth piece of information. The fourth piece of information is used to indicate the degree of vector dimension truncation processing performed on the first full sequence. The first full sequence is determined based on the second channel tensor. The first sequence is the sequence obtained after performing vector dimension truncation processing on the first full sequence.

8. The method according to claim 7, characterized in that, The channel reconstruction error information is determined based on the target channel tensor and the channel quality indicator (CQI).

9. The method according to any one of claims 1-4 and 6-8, characterized in that, The first historical reference channel tensor corresponds to a second location, which is the location closest to the terminal device among a plurality of locations. Each of the plurality of locations corresponds to at least one historical reference channel tensor, and the plurality of locations are located within the beam coverage area of ​​the network device.

10. A communication method, characterized in that, Applied to terminal devices, the method further includes: Send first information to the network device, the first information being used to indicate a first characteristic of a first channel measured by the terminal device, the first channel being a wireless channel between the terminal device and the network device; The network device receives downlink data, which is obtained based on a target channel tensor. The target channel tensor is obtained by weighting a first channel tensor and a first historical reference channel tensor, or by inputting the first channel tensor and the first historical reference channel tensor into a fusion network. The fusion network is used to perform nonlinear combination processing on the first channel tensor and the first historical reference channel tensor, and the fusion network is deployed in the network device. The first channel tensor is determined based on the first information, and the channel characteristics of the wireless channel associated with the first historical reference channel tensor are similar to the channel characteristics of the first channel.

11. The method according to claim 10, characterized in that, The first information includes a first sequence, wherein the degree to which the first feature indicated by the first sequence changes over time is greater than the degree to which the second feature of the first channel changes over time, and the second feature is indicated by a first vector; the first sequence and the first vector are determined based on an original channel feature sequence, which is obtained by compressing a second channel tensor obtained from the first channel measurement.

12. The method according to claim 11, characterized in that, The original channel feature sequence is obtained by first performing frequency domain compression on the second channel tensor, and then performing spatial domain compression.

13. The method according to claim 11, characterized in that, The first vector is obtained by inputting the original channel feature sequence into the stable split path in the two-stream decoupled autoencoder; the first sequence is obtained by inputting the original residual sequence into the fast-changing split path in the two-stream decoupled autoencoder, the original residual sequence is obtained by subtracting the original stable feature sequence from the original channel feature sequence, and the original stable feature sequence is obtained by copying the first vector according to the length of the original channel feature sequence.

14. The method according to claim 11, characterized in that, The first vector is obtained by inputting the original channel feature sequence into the stable split path in the dual-stream decoupled autoencoder; the first sequence is obtained by inputting the original residual sequence and the fourth information into the fast-change split path in the dual-stream decoupled autoencoder, the original residual sequence is obtained by subtracting the original stable feature sequence from the original channel feature sequence, the original stable feature sequence is obtained by copying the first vector according to the length of the original channel feature sequence, the fourth information is used to indicate the degree of vector dimension truncation processing performed on the first full sequence generated during the encoding process of the dual-stream decoupled autoencoder, and the first sequence is the sequence obtained after performing vector dimension truncation processing on the first full sequence.

15. The method according to claim 14, characterized in that, The two-stream decoupled autoencoder uses a nested strategy during training, which is used to sort the physical features of the second full sequence generated during the encoding process of the two-stream decoupled autoencoder according to their importance. The loss function used to train the two-stream decoupled autoencoder includes: a first component, a second component, and a third component; The value of the first component is positively correlated with the first difference degree, which is used to characterize the difference between the channel feature sequence recovered based on the second vector and the second sequence and the initial channel feature sequence. The weight of the first component is positively correlated with the degree of truncation of the second sequence. The value of the second component is positively correlated with the difference in correlation between the second vector in the time dimension. The value of the third component is positively correlated with the similarity between the second sequence and the second vector. The operation of the dual-stream decoupling autoencoder to truncate the vector dimension of the second full sequence is related to the operation of sorting the physical features of the second full sequence according to their importance.

16. The method according to claim 14, characterized in that, The fourth information is determined based on the third information from the network device, and the third information is used to adjust the fourth information.

17. The method according to any one of claims 13-16, characterized in that, The first vector is obtained by inputting the original channel feature sequence into the stable split path in the dual-stream decoupled autoencoder, including: when the original channel feature sequence is input into the stable split path in the dual-stream decoupled autoencoder, the original channel feature sequence is mapped to different subspaces by multiple attention heads in the stable split path using independent weight matrices, and then the feature vectors of multiple scenarios output by the multiple attention heads are fused.

18. The method according to any one of claims 11-16, characterized in that, Before receiving downlink data from the network device, the method further includes: When the channel aging degree is greater than or equal to a preset threshold, a second message is sent to the network device. The channel aging degree is used to indicate the degree of change of the current first channel compared to the first channel when the terminal device sends the third vector. The third vector is a vector sent by the terminal device to the network device most recently before sending the first message, which indicates a second feature of the first channel. The second message includes the first vector.

19. The method according to claim 18, characterized in that, The channel aging degree is related to the first coefficient, the first vector, and the third vector.

20. The method according to claim 19, characterized in that, The channel aging degree satisfies the following relationship with the first coefficient, the first vector, and the third vector: ； in, The channel aging degree is mentioned above. For the first coefficient, For the first vector, This is the third vector.

21. A communication device, characterized in that, The device includes at least one processor coupled to a memory storing a program or instructions, the processor executing the program or instructions to cause the device to perform the method as described in any one of claims 1 to 20.

22. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed, they cause the computer to perform the method as described in any one of claims 1 to 20.

23. A communication system, characterized in that, include: A network device and a terminal device, wherein the network device is configured to perform the method as described in any one of claims 1 to 9, and the terminal device is configured to perform the method as described in any one of claims 10 to 20.

Images