A communication method, a communication device, a system and a computer readable storage medium

Abstract

The application provides a communication method, a communication device, a system and a computer readable storage medium, which can realize joint compression and prediction functions based on a variable output Encoder and Decoder architecture of a TCN, adjust the length of a compression vector according to the time correlation of a channel, and determine the length of a prediction window. In addition, in the prediction process, a matching output layer can be selected from K output layers corresponding to the Decoder according to the variable-length compression vector, and a model algorithm associated with the output layer is used for prediction. Furthermore, in the training process, model parameters and configuration parameters can be optimized according to a loss function between predicted channel state information (CSI) and a corresponding prediction true value, so as to improve the prediction effect. The process does not need to call different AI / ML models for different scenarios, reduces the local storage overhead, and reduces the signaling overhead of the model switching process.

Description

Technical Field

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

[0002] To support higher user throughput, the channel bandwidth and the number of multiple input multiple output (MIMO) ports in 6th generation (6G) mobile communication systems will be further expanded. For example, the maximum channel bandwidth in 5th generation (5G) mobile communication systems is 100MHz, while 6G may support 800MHz. Initially, 5G only supported 32-port transmission; Rel-19's MIMO enhancements introduced 48, 64, and 128 ports. In the 6G era, it may support 256 ports. With the increase in bandwidth and the number of ports, the overhead of the traditional channel state information reference signal (CSI RS) increases significantly. Therefore, based on NR time-domain channel state information (CSI) prediction, further extensions have been made, introducing spatial, frequency, and time-domain CSI prediction based on Artificial Intelligence / Machine Learning (AI / ML) to reduce CSI RS overhead, improve channel estimation accuracy, and thus achieve higher throughput and spectral efficiency.

[0003] Through CSI prediction and CSI compression reporting, mobile networks can anticipate the state of the radio channel, enabling more efficient resource allocation, scheduling, and interference management. In the spatial-frequency channel state information compression (SF CSI compression) process, the terminal device measures the CSI RS to estimate the downlink channel and compresses the channel matrix (or precoding matrix) using an auto-encoder-decoder architecture, feeding it back to the network device. In the temporal domain CSI compression process, not only can historical CSI information be used for CSI compression, but compression can also be combined with prediction; that is, the target CSI for compression can be the current CSI or a future CSI.

[0004] In the AI / ML-based CSI compression and prediction process, different channel characteristics and different stages (such as CSI compression feedback, CSI prediction, etc.) correspond to different AI / ML models, which brings huge local storage and signaling overhead. Summary of the Invention

[0005] This application provides a communication method, communication device, system, and computer-readable storage medium. The method can implement joint compression and prediction functions based on a model architecture. This model architecture can adjust the length of the compression vector according to the time correlation of the channel matrix, so that the model can adapt to different prediction windows and prediction algorithms according to the length of the compression vector. For different scenarios, there is no need to call different AI / ML models, which reduces local storage overhead and signaling overhead during model switching.

[0006] Firstly, a communication method is provided. 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. The method includes: The system receives Channel State Information Reference Signal (CSI RS) from a network device, acquires the channel matrix and time series information of the CSI RS, outputs a first compressed vector with a target length, which is obtained by encoding the channel matrix based on the time series information, and sends the first compressed vector and first indication information to the network device. The first indication information is used to indicate the target length of the first compressed vector, and the target length of the first compressed vector is used to determine the target output layer. The model algorithm corresponding to the target output layer is used to perform decoding processing on the first compressed vector to obtain the predicted channel state information (CSI).

[0007] Optionally, in the embodiments of this application, the CSI RS can be an aperiodic RS. In other words, the CSI RS resources can be aperiodic, and the CSI RS for the next prediction window is sent only after the CSI RS of one prediction window has been sent. The embodiments of this application do not limit this.

[0008] It should be understood that the "first compressed vector" is the feedback vector obtained by the terminal device after performing encoding processing on the channel matrix. In the embodiments of this application, the first compressed vector is variable-length, that is, the encoder can perform variable-length feature compression using the temporal correlation contained in the time-series information of the CSI RS, thereby obtaining a first compressed vector with a target length. In other words, the encoder of the terminal device can determine the length of the feedback vector (CSI payload), i.e., the "target length," based on the temporal correlation. Therefore, in the description of the embodiments of this application, the "first compressed vector" can also be referred to as a "compressed vector," "variable-length compressed vector," "compressed codeword," etc.

[0009] Through the above-described scheme, the embodiments of this application can achieve joint compression and prediction functions based on the TCN's variable output encoder and decoder architecture. Furthermore, the length of the compression vector can be adjusted according to the temporal correlation of the channel matrix, allowing the model to adapt to different prediction windows and algorithms based on the length of the compression vector. This eliminates the need to call different AI / ML models for different scenarios, reducing local storage overhead and signaling overhead during model switching. In other words, the joint compression and prediction described above can perform compression and prediction on a series of temporally continuous channel matrices H as a whole, and can determine the length of the compression vector based on its temporal correlation to adapt to different prediction algorithms.

[0010] Furthermore, during the prediction process, the network device can select a matching output layer from the K output layers corresponding to the decoder based on the length of the compressed vector, and use the model algorithm associated with that output layer for prediction. This means that the variable-length CSI feedback information can be directly decompressed and restored to CSI prediction information, thereby improving the matching and accuracy of the prediction algorithm.

[0011] It should also be understood that network devices can accurately perform beamforming on terminal devices based on the predicted CSI, that is, determine which signal to send to the terminal device and in which direction, thereby suppressing interference from multiple users and improving downlink speed, coverage, and communication reliability. In addition, network devices can also use the predicted CSI to allocate time-frequency resources to terminal devices, assess the current compression accuracy, and determine the strength of channel time-varying characteristics, etc., which are not limited in the embodiments of this application.

[0012] In conjunction with the first aspect, in some implementations of the first aspect, the first compression vector is obtained by the encoder of the temporal convolutional network performing encoding processing on the channel matrix based on time series information, and the encoder of the temporal convolutional network is deployed on the terminal device.

[0013] Combining the first aspect and the above implementation methods, in some implementation methods of the first aspect, the first compressed vector is obtained by the encoder of the temporal convolutional network performing time-series modeling on the channel matrix to output the feature vector and gate vector of the maximum length, performing a masking operation on the feature vector based on the gate vector to obtain an intermediate vector, and then performing concatenation processing on the intermediate vector and the gate vector.

[0014] The process described above dynamically adjusts the effective compression dimension based on the time-varying nature of the channel and the required prediction accuracy, avoiding redundancy or insufficient accuracy in fixed-length compression. The gating vector can be learned by the TCN encoder, automatically learning the optimal compression length for different scenarios. This allows for the use of shorter compression vectors in low-speed scenarios with strong time correlation, reducing feedback overhead; and automatically extending the vector in high-speed scenarios with weak time correlation to ensure accuracy.

[0015] In conjunction with the first aspect and the above implementation methods, in some implementation methods of the first aspect, the method further includes: acquiring training samples, the training samples including multiple channel matrices and the predicted ground truth values ​​corresponding to the multiple channel matrices respectively; training the encoder, decoder and selector based on the training samples, the decoder corresponding to K output layers, each of the K output layers being associated with a different model algorithm and each output layer being able to process a compression vector of different lengths, and the selector being used to select a target output layer from the K output layers according to the target length of the first compression vector, where K is a positive integer.

[0016] In conjunction with the first aspect and the above implementation methods, in some implementation methods of the first aspect, the method further includes: receiving first configuration information from a network device, the first configuration information being used to configure model parameters of the encoder, decoder, and selector; and initializing the encoder according to the first configuration information, so that the encoder can output a second compression vector based on any second channel matrix included in the input training samples.

[0017] Combining the first aspect and the above implementation methods, in some implementation methods of the first aspect, the encoder, decoder and selector are trained based on training samples, including: obtaining the K channel state information (CSI) predicted by the K output layers of the decoder after performing decoding processing on the length of the second compression vector respectively; and adjusting the model parameters of the encoder, decoder and selector based on the loss function between the predicted K CSI and the predicted ground truth value corresponding to the second channel matrix.

[0018] In combination with the first aspect and the above implementation methods, in some implementation methods of the first aspect, the method further includes: sending second configuration information to the network device, the second configuration information being used to indicate the adjusted model parameters of the decoder and selector.

[0019] It should be understood that the model training process can be completed on the terminal device. That is, unlike the model usage process described above, the entire model training process can be completed only on the terminal device. The training of different modules such as the decoder and selector can be completed on the terminal device. After the training is completed, the terminal device can send second configuration information to the network device. The second configuration information is used to indicate the adjusted model parameters of the decoder and selector. The network device can update the model parameters related to the decoder and selector according to the second configuration information.

[0020] Through the training process described above, the network device decodes the compressed vector using the K output layers of the decoder, outputting the predicted channel state information (CSI) corresponding to the training samples. Based on the loss function between the predicted CSI and the ground truth values ​​corresponding to multiple channel matrices, the K output layers of the encoder and the K output layers of the decoder of the terminal device are fine-tuned. In other words, for each CSI sequence in the training samples, the model parameters and related configuration parameters of the encoder, decoder, and selector can be continuously optimized and adjusted based on the loss function between the predicted value obtained from each CSI sequence in the input training samples and the ground truth values ​​in the training samples, thereby improving the model's prediction performance and accuracy.

[0021] In summary, the embodiments of this application can achieve joint compression and prediction functions based on the variable output encoder and decoder architecture of TCN, and can adjust the length of the compression vector according to the temporal correlation of the channel matrix, thereby determining the length of the prediction window. In other words, the joint compression and prediction described above can compress and predict a series of temporally continuous channel matrices H as a whole, and can determine the length of the compression vector and the size of the prediction window based on their temporal correlation. Based on the temporal correlation of the channel matrix, a variable-length compression vector can be obtained, and then the length of the prediction window can be determined according to the length of the compression vector, thereby adapting to different prediction windows and prediction algorithms. That is, different prediction algorithms are selected according to temporal correlation. This process can adapt to different scenarios, and there is no need to call different AI / ML models for different scenarios, which greatly reduces the local storage requirements of the model and also reduces the signaling overhead caused by model switching.

[0022] Furthermore, during the prediction process, a matching output layer can be selected from the K output layers corresponding to the decoder based on the length of the compressed vector. The model algorithm associated with this output layer is then used for prediction. In other words, the variable-length CSI feedback information is directly decompressed and restored to CSI prediction information, which improves the matching and accuracy of the prediction algorithm.

[0023] During the training phase, the network device can pre-configure multiple prediction windows of different lengths. The model directly learns the temporal correlation between the prediction window and the CSI. Based on the loss function between the predicted value obtained from each CSI sequence in the input training samples and the true value in the training samples, the model continuously optimizes and adjusts each model parameter and related configuration parameter of the encoder, decoder, and selector, thereby improving the prediction performance of the model.

[0024] Secondly, a communication method is provided. This method can be executed by a network device, such as a base station or satellite as described in the embodiments of this application, or by a component (such as a circuit, chip, or chip system) configured in the base station or satellite. It can also be implemented by a logic module or software capable of implementing all or part of the functions of the base station or satellite; this application does not limit this. The following description uses a base station or satellite as an example. The method includes: The system sends a Channel State Information Reference Signal (CSI RS) to the terminal device; receives a first compressed vector and a first indication information from the terminal device. The first compressed vector is obtained by encoding the channel matrix of the CSI RS based on the time series information of the CSI RS, and the first indication information is used to indicate the target length of the first compressed vector; and outputs predicted Channel State Information (CSI). The CSI is obtained by decoding the first compressed vector based on the model algorithm corresponding to the target output layer, and the target output layer is determined based on the target length of the first compressed vector.

[0025] In conjunction with the second aspect, in some implementations of the second aspect, the decoder of the network device has K output layers, each output layer is associated with a different model algorithm and each output layer can process compression vectors of different lengths, where K is a positive integer.

[0026] Optionally, the first compression vector is obtained by the encoder of the temporal convolutional network deployed on the terminal device performing the encoding process on the channel matrix based on the time series information.

[0027] Combining the second aspect and the above implementation methods, in some implementation methods of the second aspect, the first compressed vector is obtained by the encoder of the temporal convolutional network performing time-series modeling on the channel matrix to output the feature vector and gate vector of the maximum length, performing a masking operation on the feature vector based on the gate vector to obtain an intermediate vector, and then performing concatenation processing on the intermediate vector and the gate vector.

[0028] In conjunction with the second aspect and the above implementation methods, in some implementation methods of the second aspect, the method further includes: sending first configuration information to the terminal device, the first configuration information being used to configure the model parameters of the encoder, decoder and selector, and the first configuration information being used to initialize the encoder, so that the encoder can train the encoder, decoder and selector based on the input training samples, the training samples including multiple channel matrices and the predicted ground truth values ​​corresponding to the multiple channel matrices respectively.

[0029] In combination with the second aspect and the above implementation, in some implementations of the second aspect, the selector is used to select a target output layer from the K output layers of the decoder according to the target length of the first compression vector.

[0030] In conjunction with the second aspect and the above implementation methods, in some implementation methods of the second aspect, the method further includes: receiving second configuration information from a terminal device, the second configuration information being used to indicate the model parameters of the decoder and selector after training and adjustment.

[0031] The second aspect is the implementation on the network device 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.

[0032] Thirdly, a communication device is provided, comprising a communication unit and a processing unit, the communication unit and the processing unit cooperating with each other to enable the communication device to perform the functions of the terminal device designed in the method of the first aspect. These functions can be implemented by hardware or by hardware executing corresponding software. The hardware or software includes one or more units corresponding to the above functions.

[0033] Fourthly, a communication device is provided, comprising a communication unit and a processing unit, the communication unit and the processing unit cooperating with each other to enable the communication device to perform the functions of the network device designed in the method of the second aspect described above. These functions can be implemented by hardware or by hardware executing corresponding software. The hardware or software includes one or more units corresponding to the above functions.

[0034] The third and fourth aspects are the implementation on the device side, which correspond to the first and second aspects. The explanations, supplements, and descriptions of the beneficial effects of the first and second aspects also apply to the third and fourth aspects, and will not be repeated here.

[0035] Fifthly, a communication device is provided, including a processor. The processor is coupled to a memory and can be used to execute instructions or data in the memory to implement the method in any of the possible implementations of the first aspect described above. Optionally, the communication device further includes a memory. Optionally, the communication device further includes a communication interface, and the processor is coupled to the communication interface.

[0036] In one implementation, the communication interface can be a transceiver, or an input / output interface.

[0037] In another implementation, the communication device is a chip configured in a terminal device. When the communication device is a chip configured in a terminal device, the communication interface can be an input / output interface.

[0038] In a sixth aspect, a communication device is provided, including a processor. The processor is coupled to a memory and can be used to execute instructions or data in the memory to implement the method in any possible implementation of the second aspect described above. 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 can be a transceiver, or an input / output interface.

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

[0041] In a seventh aspect, a processor is provided, comprising: an input circuit, an output circuit, and a processing circuit. The processing circuit is used to receive signals through the input circuit and to 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, for example, but not limited to, output to and transmitted by a transmitter. Furthermore, the input circuit and the output circuit can be the same circuit, which is used as both 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] Eighthly, a communication device is provided, including a processor and a memory. The processor is used to read instructions stored in the memory, receive signals via a receiver, and transmit signals via a transmitter to execute the method in any possible implementation of any of the preceding aspects.

[0044] Optionally, there may be one or more processors and one or more memories.

[0045] Ninthly, a computer program product is provided, comprising: a computer program (also referred to as code or instructions) that, when executed, causes a computer to perform a method in any possible implementation of any of the above aspects.

[0046] In a tenth aspect, a computer-readable storage medium is provided that stores a computer program (also referred to as code or instructions) that, when executed on a computer, causes the computer to perform the methods in any possible implementation of any of the above aspects.

[0047] Eleventhly, 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 possible implementations to be executed. The chip system may be composed of chips or may include chips and other discrete devices.

[0048] The chip system may include input circuits or interfaces for transmitting information or data, and output circuits or interfaces for receiving information or data.

[0049] In a twelfth aspect, a communication system is provided, including the aforementioned terminal device and network device. Optionally, the communication system may further include other devices that communicate with the terminal device and / or network device. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the architecture of a mobile communication system applicable to embodiments of this application.

[0051] Figure 2 This is a flowchart of an AI-based CSI compression and prediction process.

[0052] Figure 3 This is a schematic diagram of a communication method 300 provided in an embodiment of this application.

[0053] Figure 4 This is a schematic diagram of a sliding window for CSI prediction and feedback in this application.

[0054] Figure 5 This is an example of a variable-length compressed vector provided in the embodiments of this application. A schematic diagram illustrating the implementation process.

[0055] Figure 6 This is a schematic diagram illustrating a complete process of CSI inference based on an AI model, as provided in an embodiment of this application.

[0056] Figure 7 This is a schematic diagram of the training process of an example model provided in an embodiment of this application.

[0057] Figure 8 This is a schematic block diagram of a communication device provided in an embodiment of this application.

[0058] Figure 9 This is another schematic block diagram of the communication device 900 provided in the embodiments of this application. Detailed Implementation

[0059] In the embodiments of this application, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this embodiment, unless otherwise stated, "multiple" means two or more.

[0060] It should be noted that, in the description of the embodiments of this application, unless otherwise stated, " / " means "or", for example, A / B can mean A or B; "and / or" in this document describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, in the description of the embodiments of this application, "multiple" refers to two or more.

[0061] It should also be noted that in the embodiments of this application, "preset", "fixed value", etc. can be implemented by pre-saving the corresponding code, table or other means that can be used to indicate relevant information in the electronic device. This application does not limit the specific implementation method.

[0062] It should be understood that the methods, situations, categories, and classifications of embodiments in this application are for the convenience of description only and should not constitute a special limitation. Various methods, categories, situations, and features in embodiments can be combined with each other without contradiction.

[0063] It should also be understood that, in the description of this embodiment, unless otherwise stated, "multiple" means two or more. 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.

[0064] The technical solutions of this application can be applied to various communication systems, such as: Long Term Evolution (LTE) system, LTE Frequency Division Duplex (FDD) system, LTE Time Division Duplex (TDD) system, Universal Mobile Telecommunication System (UMTS), New Radio (NR) in 5th Generation (5G) mobile communication system, and future mobile communication systems, such as 6th Generation (6G) mobile communication system. This application does not limit these applications.

[0065] The technical solutions provided in this application can also be applied to machine-type communication (MTC), long-term evolution-machine (LTE-M) technology, device-to-device (D2D) networks, machine-to-machine (M2M) networks, Internet of Things (IoT) networks, or other networks. Among these, IoT networks may include, for example, vehicle-to-everything (V2X) networks. The communication methods in V2X systems are collectively referred to as vehicle-to-X (V2X), where X can represent anything. For example, V2X may include vehicle-to-vehicle (V2V) communication, vehicle-to-infrastructure (V2I) communication, vehicle-to-pedestrian (V2P) communication, or vehicle-to-network (V2N) communication, etc. This application does not limit these applications.

[0066] Figure 1 This is a schematic diagram of the architecture of a mobile communication system applicable to embodiments of this application.

[0067] like Figure 1As shown, the communication system includes a wireless access network 100, a core network 20, and an Internet 30. The wireless access network 100 may include at least one access network device (such as...). Figure 1 Access network devices 110a and 110b, collectively referred to as access network device 110, may also include at least one terminal (such as...). Figure 1 The devices 120a, 120b, 120c, 120d, 120e, 120f, 120g, 120h, 120i, and 120j are collectively referred to as Terminal 120. Terminals 120a-120j are connected wirelessly to access network devices 110a and 110b. Access network devices 110a and 110b are connected to the core network 20 wirelessly or via wired connection. The core network devices in the core network and the access network devices in the wireless access network can be different physical devices, or they can be the same physical device integrating core network logical functions and wireless access network logical functions, or they can be a single physical device integrating some core network device functions and some wireless access network device functions. Terminals can be interconnected via wired or wireless connections, and access network devices can be interconnected via wired or wireless connections. It should be understood that... Figure 1 This is just an illustration; the communication system may also include other network devices, such as wireless repeaters and / or wireless backhaul devices. Figure 1 Not shown in the image.

[0068] The access network device 110 in the wireless access network 100 of this application embodiment is sometimes also called an access node. The access network device 110 has wireless transceiver capabilities for communicating with the terminal 120. The access network device 110 includes, but is not limited to, base stations, evolved NodeBs (eNodeBs), transmission reception points (TRPs), next-generation NodeBs (gNBs) in 5G mobile communication systems, next-generation base stations in 6th-generation (6G) mobile communication systems, access network devices or modules of access network devices in Open RAN (ORAN) systems, base stations in future mobile communication systems, or access nodes in WiFi systems. The access network device 110 can also be a module or unit capable of implementing some of the functions of a base station. For example, the access network device 110 can be a central unit (CU), a distributed unit (DU), a CU-control plane (CP), a CU-user plane (UP), or a radio unit (RU), as described below. In the ORAN system, CU can also be called O-CU, DU can also be called open (O)-DU, CU-CP can also be called O-CU-CP, CU-UP can also be called O-CUP-UP, and RU can also be called O-RU. Access network equipment 110 can be a macro base station (such as...) Figure 1 110a), micro base stations or indoor stations (such as Figure 1 Access network device 110 (110b) can be a relay node or donor node, or a wireless controller in a cloud radio access network (CRAN) scenario. Optionally, access network device 110 can also be a server, wearable device, or vehicle-mounted device, etc. For example, in vehicle-to-everything (V2X) technology, access network device 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 or through relay stations. Terminals can communicate with multiple base stations in different access technologies. The embodiments of this application do not limit the specific technology or specific device form adopted by access network device 110.

[0069] In this application, the access network device is abbreviated as "network device". Unless otherwise specified, the network device refers to the access network device in this application. For example, the first network device and the second network device in the following embodiments. For example, the first network device can be a base station in a 5G communication system, and the second network device can be a base station in a 6G communication system. This application does not limit the specific network device.

[0070] It should be understood that the first network device and the second network device can be coexisting entities or independent entities, and this application embodiment does not limit this.

[0071] It should also be understood that the first network device and the second network device can communicate or exchange information through core network interconnection and internal interface collaboration. For example, core network interconnection can serve as a collaborative method, allowing core network 20 to simultaneously manage both 5G access network devices and 6G access network devices. Both 5G base stations and 6G base stations can connect to core network 20 via the backhaul network. Alternatively, the interface between base stations (e.g., the Xn interface) can support real-time information exchange between 5G and 6G base stations; this embodiment of the application does not limit this approach.

[0072] A terminal can also be called a terminal device, user equipment (UE), mobile station, mobile terminal, etc. Terminals can be widely used in various communication scenarios, such as device-to-device (D2D) communication, vehicle-to-everything (V2X) communication, machine-type communication (MTC), Internet of Things (IoT), virtual reality, augmented reality, industrial control, autonomous driving, telemedicine, smart grids, smart furniture, smart offices, smart wearables, smart transportation, or smart cities. Terminals can be mobile phones, tablets, computers with wireless transceiver capabilities, wearable devices, vehicles, drones, helicopters, airplanes, ships, robots, robotic arms, or smart home devices, etc. This application does not limit the device form of the terminal.

[0073] For example, such as Figure 1The devices 120a-120i shown can be understood as communication devices with terminal functions. Specifically, 120a is a smartphone; 120b is an in-vehicle device; 120c is a mobile power bank charging station; 120d is a smart home device; 120e is a wearable device; 120f is a smart point-of-sale device or smartphone; 120g is a tablet, laptop, or PDA; 120h is a smart meter or smart printer; 120i is an in-flight device, hotspot device, or mobile device (e.g., a smartphone); 120i is a mobile device (e.g., a smartphone).

[0074] In this application, the terminal device can be any of the devices listed above.

[0075] Access network devices and / or terminals can be fixed or mobile. They can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; they can also be deployed on water; and they can be deployed in the air on aircraft, balloons, and satellites. This application does not limit the application scenarios of the access network devices and terminals. Furthermore, access network devices and terminals can be deployed in the same or different scenarios; for example, both can be deployed on land; or the access network device can be deployed on land and the terminal on water, and so on.

[0076] Taking a base station as a network device as an example, the roles of the base station and the terminal can be relative, for example, Figure 1 The helicopter or drone 120i can be configured as a mobile base station. For terminals 120j accessing the wireless access network 100 via 120i, 120i can be a base station; however, for 110a, 120i is a terminal. 110a and 120i communicate via a wireless air interface protocol, or they can communicate via a base station-to-base station interface protocol. In this case, relative to 110a, 120i is also a base station. Therefore, both base stations and terminals can be collectively referred to as communication devices. Figure 1 The 110a and 110b in the text can be referred to as communication devices with base station functions. Figure 1 The 120a-120j in the text can be referred to as communication devices with terminal functions.

[0077] In this embodiment, the communication device with access network device functionality can be an access network device, a module within an access network device (such as a chip, chip system, or software module), or a control subsystem containing access network device functionality. For example, a control subsystem containing access network device functionality can be a control center in scenarios where terminals can be applied, such as smart grids, industrial control, intelligent transportation, or smart cities.

[0078] In the embodiments of this application, the communication device with terminal function can be a terminal, or a module in the terminal (such as a chip, chip system, modem, or software model, etc.), or a device that includes terminal function. In the embodiments of this application, for ease of description, subsequent embodiments will be described using network devices (e.g., base stations) and terminals (UEs) as examples.

[0079] Communication between base stations and terminals, between base stations, and between terminals can be conducted using licensed spectrum, unlicensed spectrum, or both simultaneously. Communication can be conducted using spectrum below 6 GHz, spectrum above 6 GHz, or both simultaneously. The embodiments of this application do not limit the spectrum resources used for wireless communication.

[0080] The technical solutions provided in this application can be applied to wireless communication between communication devices. Wireless communication between communication devices can include: wireless communication between a base station and a terminal, wireless communication between base stations, and wireless communication between terminals. It is understood that in the embodiments of this application, the physical downlink share channel (PDSCH), physical downlink control channel (PDCCH), and physical uplink share channel (PUSCH) are merely examples of downlink data channels, downlink control channels, and uplink data channels, respectively. In different systems and scenarios, data channels and control channels may have different names, and the embodiments of this application do not limit this. In this application, the base station sends downlink signals or downlink information to the terminal, and the downlink information is carried on the downlink channel; the terminal sends uplink signals or uplink information to the base station, and the uplink information is carried on the uplink channel. In order to communicate with the base station, the terminal needs to establish a wireless connection with the cell controlled by the base station. The cell with which the terminal has established a wireless connection is called the terminal's "serving cell." When the terminal communicates with the serving cell, it is also subject to interference from signals from neighboring cells.

[0081] 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 be found in the 3GPP standard protocols. It should be understood that the technical terms in this application are for illustrative purposes only and not as limiting. For example, as technology evolves, technical terms may also change; where the technical meaning remains the same, other technical terms should also apply to this application.

[0082] 1. Beam and reference signal (RS) A beam is a communication resource. A beam can be wide, narrow, or other types of beams. The technology used to form a beam can be beamforming or other techniques. Beamforming technology can specifically be digital beamforming, analog beamforming, or hybrid digital / analog beamforming. Different beams can be considered different resources. The same or different information can be transmitted through different beams. Optionally, multiple beams with the same or similar communication characteristics can be considered as a single beam. A beam can include one or more antenna ports for transmitting data channels, control channels, and detection signals, etc. For example, a transmit beam can refer to the signal strength distribution in different directions of space after a signal is transmitted through an antenna, and a receive beam can refer to the signal strength distribution in different directions of space of a wireless signal received from an antenna. It is understood that one or more antenna ports forming a beam can also be considered as a set of antenna ports.

[0083] Beams can be divided into transmit beams and receive beams of network devices, and transmit beams and receive beams of terminals. The transmit beam of a network device describes the beamforming information transmitted by the network device, and the receive beam of a network device describes the beamforming information received by the network device. The transmit beam of a terminal describes the beamforming information transmitted by the terminal, and the receive beam of a terminal describes the beamforming information received by the terminal. In other words, beams are used to describe beamforming information.

[0084] Beams can correspond to time resources and / or spatial resources and / or frequency domain resources.

[0085] Alternatively, the beam can also correspond to a reference signal resource or beamforming information.

[0086] Optionally, the beam can also correspond to information associated with reference signal resources of the network device. The reference signal can be a channel state information reference signal (CSIRS), a synchronous signal / physical broadcast channel block (SSB), a demodulation reference signal (DMRS), a phase tracking reference signal (PTRS), a tracking reference signal (TRS), etc. The information associated with the reference signal resource can be a reference signal resource identifier or quasi-co-location (QCL) information, etc. The reference signal resource identifier corresponds to a transmit / receive beam pair previously established based on measurements of that reference signal resource. Through this reference signal resource index, the terminal can infer the beam information.

[0087] Communication systems typically use different types of reference signals: one type is used to estimate the channel, enabling coherent demodulation of received signals containing control information or data; another type is used to measure channel state or channel quality, thereby enabling terminal scheduling. Terminal devices obtain channel state information (CSI) based on channel quality measurements of the CSI RS. This CSI information can be transmitted by the terminal to network devices via the physical uplink control channel (PUCCH) or the physical uplink shared channel (PUSCH).

[0088] 2. Channel State Information (CSI) During the journey of a wireless signal from the transmitter to the receiver through a wireless channel, fading occurs due to scattering, reflection, and energy attenuation with distance. Furthermore, the wireless signal may be interfered with by other signals at the receiver, affecting reception. Signal attenuation and interference characteristics can be characterized using signal attenuation indexing (CSI).

[0089] Optionally, the CSI may include at least one of the following: CSI RS resource indicator (CRI), rank indicator (RI), precoding matrix indicator (PMI), channel quality indicator (CQI), synchronization signal and physical broadcast channel block (SSB) resource indicator (SSBRI), layer indicator (LI), L1-reference signal received power (RSRP), and L1-signal to interference plus noise ratio (SINR).

[0090] Optionally, CSI can be sent from the terminal to the network device via PUCCH or PUSCH.

[0091] During communication, network devices and terminals obtain the Channel Sounding Reference Signal (SRS) by measuring the Channel Sounding Reference Signal (SRS), and then perform data transmission and reception based on the obtained SRS. For example, the channel or interference can be measured by sending an RS with a known sequence. In NR, the downlink channel is typically measured using the CSI RS, and the uplink channel is measured using the Channel Sounding Reference Signal (SRS).

[0092] 3. Channel State Information Report Configuration (CSI-ReportConfig) This is primarily used to configure parameters related to channel state reporting, such as the reporting type and the measured metrics being reported. Specifically, the reporting configuration identifier (reportConfigId) is the identity (Id) number for this CSI-ReportConfig, used to identify the CSI-ReportConfig; Channel Measurement Resources (resources for Channel Measurement) configures the Channel State Information - Reference Signal (CSI RS) resources for channel measurements, and is associated with the resource configuration through CSI-ResourceConfigId; Interference Measurement Resources (CSI-IM-Resources for Interference) configures the CSI RS resources used for interference measurements, and is associated with the resource configuration through CSI-ResourceConfigId.

[0093] Optionally, the CSI report configuration parameters may include CSI report type (reportConfigType), CSI report quantity (report Quantity), etc. CSI report type can be divided into periodic, semi-static (semi-Persistent On PUCCH or semi-Persistent On PUSCH), and aperiodic reporting. Network devices can configure different report quantities to allow terminal devices to report different CSIs, including CSI RS resource indicator (CSIRS resource indicator, CRI), rank indicator (RI), pre-coding matrix indicator (PMI), channel quality indicator (CQI), SINR, RSRP, etc.

[0094] 4. Channel State Information Resource Configuration (CSI - ResourceConfig) Information related to resources used for configuring CSI measurements. This may include a reported resource identifier (CSI-ResourceConfigId) and / or a resource set list (CSI RS-Resource set list). The CSI-ResourceConfigId is used to identify the CSI-ResourceConfig; the CSI RS-Resource set list may include resource sets for channel measurements and resource sets for interference measurements.

[0095] 5. CSI Report CSI report feedback is a crucial means of ensuring the communication quality of wireless communication links. CSI report feedback refers to the process where a network device sends an RS (Receiving Signal) to a terminal device, and the terminal device, after measuring and calculating the RS, feeds back the CSI (Content Quality Index) to the network device via a CSI report. Based on the measured values, i.e., the CSI, fed back by the terminal device, the network device adjusts its modulation and / or coding to ensure communication speed.

[0096] CSI reports are sent from terminal devices to network devices to inform the network devices of the channel status when they send downlink information to the terminal devices. One CSI report instructs the terminal device to send back one CSI. Different CSIs can correspond to different frequency bands, different transmission assumptions, or different reporting modes.

[0097] Generally, a CSI report can be associated with one reference signal resource for channel measurement, and also with one or more reference signal resources for interference measurement. One CSI report corresponds to one transmission resource, that is, the time-frequency resource used by the terminal device to transmit the CSI.

[0098] The CSI report in this application may also be referred to as "CSI reporting".

[0099] CSI reports include three types: periodic CSI reports, semi-continuous CSI reports, and non-periodic CSI reports.

[0100] For periodic CSI reports, the terminal periodically sends channel state information. For example, the terminal can send channel state information at intervals of T time units, where the time unit can be a time slot, millisecond, subframe, etc.

[0101] For semi-persistent CSI reports, the terminal can periodically send channel state information within a first time period. For example, after receiving the activation MAC CE or PDCCH from a semi-persistent CSI report, the terminal sends channel state information at intervals of T time units. Afterward, after receiving the deactivation MAC CE or PDCCH from a semi-persistent CSI report, the terminal stops sending channel state information.

[0102] For aperiodic CSI reports, after receiving the activation signaling for an aperiodic CSI report, the terminal reports channel state information and stops reporting channel state information after the reporting ends.

[0103] 6. Artificial intelligence / machine learning processing unit (AI / ML PU) The AI / ML PU, also known as the APU, is a new module specifically designed to process CSI reports related to AI / ML. It aims to optimize the performance of wireless communication systems by leveraging advanced algorithms and technologies to improve the accuracy and efficiency of channel state information.

[0104] It should be understood that the definitions of the above terms can be referenced from the prior art. However, as technology continues to develop, the above definitions may also change, and the embodiments of this application are not limited thereto.

[0105] To support higher user throughput, 6G channel bandwidth and the number of multiple input multiple output (MIMO) ports will be further expanded. For example, NR's maximum channel bandwidth is 100MHz, while 6G may be 800MHz. NR initially only supports 32 ports, but Rel-19's MIMO enhancements introduced 48, 64, and 128 ports. In the 6G era, it may support 256 ports. With the increase in bandwidth and port count, the overhead of traditional CSI RS increases significantly. Therefore, based on NR's time-domain CSI prediction, further extensions have been made, introducing spatial, frequency, and time-domain CSI prediction based on Artificial Intelligence (AI) to reduce CSI RS overhead, improve channel estimation accuracy, and thus achieve higher throughput and spectral efficiency.

[0106] AI / ML-based CSI prediction only requires transmitting CSI RS for a portion of the spatial resources (e.g., ports, antennas) and / or a portion of the frequency resources (e.g., subcarriers, resource blocks, subbands). Terminal devices can then predict the complete CSI using the CSI RS. Through CSI prediction and CSI compression reporting, network devices can anticipate the state of the wireless channel, enabling more efficient resource allocation, scheduling, and interference management.

[0107] The following describes several different CSI compression use cases: 1. Spatial-frequency channel state information compression (SF CSI compression) SF CSI compression is a core technology for MIMO and 5G / 6G. By deeply exploring the strong correlation between the channel in the spatial domain (antenna) and frequency domain (subcarrier), it achieves high compression ratio, low complexity, and high precision CSI feedback and transmission. Typical processing models can include convolutional neural networks (CNN), spatial-frequency networks (SF-Net), and transformer models.

[0108] In the SF CSI compression use case, the terminal device can measure the CSI RS and estimate the downlink channel, then compress the channel matrix (or precoding matrix) and feed it back to the network device (NW) via an encoder-decoder architecture. The encoder on the terminal device side inputs the high-dimensional CSI matrix (real part + imaginary part) into the SF-Net, extracts global / local features, and compresses it into low-dimensional codewords. The decoder on the network side receives the codewords, reconstructs the high-precision CSI, restores the spatial-frequency domain features, and then predicts the future CSI in the time domain, supporting large-scale MIMO precoding and multi-user scheduling.

[0109] 2. Temporal domain CSI compression Temporal domain CSI compression leverages the strong correlation of CSI over time, such as slow time-varying, smooth changes, and high similarity between adjacent time points, to remove redundancy, reduce dimensionality, and encode multiple consecutive frames of CSI, achieving high compression ratio, low latency, and high-precision reconstruction. It is mainly used in scenarios such as high-speed movement, continuous CSI feedback, V2X, and real-time behavior recognition.

[0110] For example, typical processing models may include long short-term memory network (LSTM) encoder-decoder processing models, temporal convolutional network (TCN) transformer encoder-decoder processing models, differential prediction processing models, etc.

[0111] In LSTM, the encoder encodes temporal features from historical CSI sequences, compressing the variable-length input sequence into a fixed-dimensional context vector. The decoder then uses this vector to progressively generate variable-length output sequences, predicting future CSIs frame by frame. In TCN, the encoder extracts local temporal features, which are then captured by the Transformer to capture long-range temporal dependencies. The decoder utilizes a TransformerDecoder with temporal location encoding, supporting multi-step parallel prediction. This process eliminates the need for frame-by-frame autoregression, directly outputting all predicted frames and reducing latency.

[0112] In the temporal domain CSI compression use case, the terminal device can not only use historical CSI information for CSI compression, but also combine compression with prediction, that is, the target CSI for compression can be the current CSI or the future CSI.

[0113] In the Encoder-Decoder architecture for CSI prediction described above, the core of CSI prediction is to extract temporal or spatial frequency domain features through the Encoder and then predict the CSI at future times through the Decoder. The overall architecture needs to balance feature extraction capability, temporal modeling accuracy, and low complexity, and be compatible with terminal devices.

[0114] Figure 2 This is a flowchart illustrating AI-based CSI compression and prediction. This process enables CSI estimation, prediction, encoding, and compression to be completed on the terminal device side, while decoding and reconstruction are performed on the network side, thereby reducing feedback overhead and improving beamforming accuracy.

[0115] For example, Figure 2 The CSI compression and prediction method 200 shown includes the following process: S21, the network device sends CSI RS to the terminal device for the terminal device to perform downlink channel estimation.

[0116] S22, the terminal device estimates the original CSI matrix at the current time based on CSI RS, and the original CSI matrix may contain amplitude and phase information.

[0117] Optionally, the network device sends a CSI RS, and the terminal device, after receiving the CSI RS, estimates the downlink channel matrix according to the following formulas (1) and (2): Formula (1) in, This indicates the signal received by the terminal device. Let M be the receiving antenna and H be the subcarrier. H represents the CSI matrix to be estimated. , X is the transmitting antenna. X is the pilot signal with known CSI RS, and N is Gaussian white noise.

[0118] Estimate the downlink channel matrix using least squares (LS) : Formula (2) Finally, the CSI of the discrete time series is obtained, labeled as .

[0119] S23, the prediction module (predictor) of the terminal device uses the historical CSI sequence to predict the CSI at future times, compensating for the channel time-varying error caused by the feedback delay.

[0120] Input historical CSI sequences Predict the future based on the following formula (3) frame: Formula (3) Where L is the length of the history window. To predict the step size. For time series prediction models, such as any one of the previously introduced LSTM, TCN, and Transformer. Then calculate the loss function according to the following formula (4), which can be understood as the prediction accuracy: Formula (4) S24, the Encoder of the terminal device extracts and compresses the predicted CSI to generate low-dimensional codewords.

[0121] The predicted CSI future can be calculated using the following formula (5). frame Compressed into low-dimensional codewords : Formula (5) in, d (2 represents the real part + the imaginary part).

[0122] S25, the network device receives the compressed codeword reported by the terminal device.

[0123] S26, The network device's decoder is based on compressed codewords. Decode and reconstruct the complete CSI for use in precoding, beamforming, and resource scheduling at the base station.

[0124] Network device receives codewords According to the following formula (6), the Decoder can decode the CSI and reconstruct the CSI: Formula (6) It should be understood that It is a decoding network symmetrical to the Encoder, i.e., the decoding algorithm of the Decoder. Furthermore, the loss function can be reconstructed according to the following formula (7): Formula (7) The total loss of the process is calculated according to the following formula (8). ,in, This is used to balance the accuracy of predicting and reconstructing CSI.

[0125] Formula (8) The method described above, 200, achieves an integrated design for CSI prediction and compression. First, the predictor module in S23 compensates for channel time-varying characteristics, and then the encoder in S24 compresses redundancy, improving CSI timeliness and significantly reducing feedback overhead. This process places the complex estimation, prediction, and encoding tasks on the terminal device side, requiring only lightweight decoding on the network device side, which aligns with the computing power allocation principles of 5G or 6G distributed architectures.

[0126] However, in traditional AI / ML-based CSI prediction and compression methods, different channel characteristics correspond to different AI models. For example, the strength of event correlation in a channel may correspond to different AI models; or, different use cases, such as different processing steps for CSI compression feedback and CSI prediction, may correspond to different AI models. This results in both the terminal device and network device sides needing to store multiple sets of model parameters, leading to significant local storage overhead. Furthermore, during scenario switching, it is necessary to negotiate and switch model versions and configure hyperparameters (such as window length and compression ratio) via network-side signaling. The frequency of signaling interactions increases linearly with scenario complexity, consuming air interface resources and resulting in substantial signaling overhead.

[0127] To address the aforementioned issues, this application provides a communication method that can adapt to CSI predictions with different time correlations in different scenarios, significantly reducing model storage requirements and minimizing signaling overhead during model switching.

[0128] 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 (e.g., terminal devices, 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 (e.g., terminal devices, 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 logic modules or software that can implement all or part of the functions of the device.

[0129] 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.

[0130] Figure 3 This is a schematic diagram of an example communication method 300 provided in an embodiment of this application. It should be understood that... Figure 3 The terminal device in the middle can be Figure 1 Any terminal device in the context of network equipment can refer to any component within a terminal device (such as a processor, chip, or chip system). Network equipment can be... Figure 1 Any access network device, such as a base station; or it can refer to a device within the access network device (such as a processor, chip, or chip system).

[0131] like Figure 3 As shown, method 300 includes the following steps: S301, the network device sends a Channel State Information Reference Signal (CSI RS) to the terminal device, and the terminal device receives the CSI RS from the network device.

[0132] It should be understood that in the embodiments of this application, the network device can send CSI RS to the terminal device, for example, in the time and frequency domain according to a fixed pattern. The terminal device knows the time period during which the network device sends the CSI RS, and the CSI RS is the basis for the terminal device to perform downlink channel estimation.

[0133] Optionally, in this embodiment, CSI RS can be sent aperiodically, with the CSI RS for the next prediction window being sent only after the CSI RS for one prediction window has finished being sent. This embodiment does not limit this.

[0134] Figure 4 This is a schematic diagram of a sliding window for CSI prediction and feedback in this application. Combined with... Figure 4 Examples illustrate several different window ranges for interaction between terminal devices and network devices.

[0135] For example, such as Figure 4 As shown, the rectangle filled with slanted lines and shaded areas represents the "observation window," in which network devices send CSI RS to terminal devices. The "observation window" can be understood as the period during which network devices send CSI RS to terminal devices for downlink channel estimation; the CSI RS within this period is the basis data used by the terminal devices for CSI prediction.

[0136] The black-filled rectangle illustrates the "CSI feedback" process, which is the process by which the terminal device reports a CSI report to the network device. Specifically, the terminal device can report compressed CSI codewords to the network device.

[0137] The white rectangle shows the dynamic prediction window. The CSI in this dynamic prediction window is predicted by the terminal device based on the model and the CSI RS received in the observation window. The window range of the dynamic prediction window is the prediction range. The window length of the dynamic prediction window can be dynamically adjusted according to the time correlation of the channel.

[0138] It should be understood that, in the embodiments of this application, "channel temporal correlation" means that there are no instantaneous changes between wireless channels, and the difference between the CSI measured at two adjacent moments (e.g., the 1st frame, the 2nd frame, the 3rd frame, etc.) is very small. That is, the channel changes slowly. This pattern of highly similar, mutually bound, and mutually deductive CSI at consecutive moments is called "channel temporal correlation." In addition, in the embodiments of this application, "temporal correlation" can also be called "temporal domain correlation," which refers to the inherent channel characteristics of highly similar, slowly changing, and highly repetitive CSI across multiple consecutive frames within a continuous time period. This is the core of temporal CSI compression and temporal CSI prediction.

[0139] For example, such as Figure 4 As shown in the diagram, the white rectangle indicates that a longer dynamic prediction window indicates a stronger temporal correlation of the channel; conversely, a shorter dynamic prediction window indicates a weaker temporal correlation of the channel. The time-domain dependence of the channel between adjacent time points can be represented as: .in, Let be the channel matrix at time t. Here is the channel matrix at time t-1. The time-domain correlation coefficient ( The closer the correlation is to 1, the stronger the correlation. This refers to the micro-domain residual or change.

[0140] It should also be understood that temporal correlation information can be the result of model learning, i.e., based on the initial matrix. Learn and determine the time series of the current CSI, that is, determine the time correlation information.

[0141] S302, the terminal device obtains the channel matrix and time series information of CSI RS.

[0142] S303, the terminal device outputs a first compressed vector with a target length, which is obtained by encoding the channel matrix based on time series information.

[0143] It should be understood that the "first compression vector" is the feedback vector obtained by the terminal device after encoding the channel matrix. In the embodiments of this application, the first compression vector is variable-length, that is, the encoder can use the time correlation contained in the time series information of CSI RS to perform variable-length feature compression, thereby obtaining a first compression vector with a target length.

[0144] It should also be understood that "time series information" is the same as the "time correlation information" described above. In other words, the encoder of the terminal device can determine the length of the feedback vector (CSI payload), i.e., the "target length," based on the time correlation information. Therefore, in the description of the embodiments of this application, "first compression vector" can also be referred to as "compression vector," "variable-length compression vector," "compression codeword," etc.

[0145] In one possible implementation, the terminal device can encode the channel matrix based on time-series information using an encoder of a temporal convolutional network (TCN) to obtain a first compressed vector with a target length. The encoder of the temporal convolutional network is deployed on the terminal device.

[0146] For example, the terminal device-side model can be the encoder of a TCN, which can use temporal correlation to perform variable-length feature compression to obtain a first compressed vector with a target length.

[0147] For example, the network device sends a CSI RS, and the terminal device, after receiving the CSI RS, estimates the downlink channel matrix according to the formulas (1) and (2) described above. Finally, the CSI of the discrete time series is obtained, labeled as ,Right now Furthermore, the terminal device can also obtain the time-series information of the CSI RS, and then the encoder of the terminal device can encode the input downlink channel matrix and determine the target length (CSI payload) of the feedback vector obtained after encoding based on the time-series information.

[0148] It should also be understood that temporal correlation information can be the result of model learning, i.e., based on the initial matrix. Learn and determine the time series of the current CSI, i.e., determine the time correlation information and feedback vector. The target length.

[0149] In variable-length compression vectors In implementation, a gated output dimension method can be used, meaning the encoder's output actually consists of two parts. , , , , The following is combined with Figure 5 Let me introduce it.

[0150] Figure 5 This is an example of a variable-length compressed vector provided in the embodiments of this application. A schematic diagram illustrating the implementation process.

[0151] For example, such as Figure 5 As shown, the CSI variable-length compression coding process based on the TCN encoder can achieve adaptive compression length through gating and masking, minimizing feedback overhead while ensuring accuracy.

[0152] Specifically, Figure 5 The process shown may include: (1) Temporal modeling process: The TCN encoder performs temporal modeling on the input CSI sequence and outputs a feature vector with a fixed maximum length. and gate vector The TCN encoder can output two signals: the maximum length candidate feature vector. and gate vector It should be understood that the TCN encoder is parameterized. The temporal convolutional network is capable of extracting the temporal features of CSI.

[0153] (2) Gating mask operation: based on gating vector right Perform a masking operation to retain the valid dimensions. It should be understood that the gating vector... Each element in the vector is either 0 or 1, used to indicate the feature vector in the vector. Is the corresponding dimension valid? The masking operation transforms the feature vector... middle Set the dimension to zero. This yields... .

[0154] It should be understood that eigenvectors The fixed-length design facilitates hardware implementation, and the masking operation only takes effect at the logic layer without changing the underlying transmission format, thus improving the model's compatibility.

[0155] (3) Vector concatenation process: Finally, by and Together they form a variable-length compressed vector. , used for CSI feedback. It can achieve adaptive length compression.

[0156] The process described above can dynamically adjust the effective compression dimension based on the time-varying nature of the channel and the required prediction accuracy, avoiding redundancy or insufficient accuracy in fixed-length compression. The optimal compression length is learned by the TCN encoder and automatically learns the optimal compression length for different scenarios. This allows for the use of shorter compression vectors in low-speed scenarios with strong time correlation, reducing feedback overhead; and the vectors are automatically extended in high-speed scenarios with weak time correlation to ensure accuracy.

[0157] S304, the terminal device sends a first compression vector and a first indication information to the network device, and the network device receives the first compression vector and the first indication information sent by the terminal device.

[0158] The first compression vector is obtained by encoding the channel matrix of the CSI RS based on the time series information of the CSI RS, and the first indication information is used to indicate the target length of the first compression vector.

[0159] Optionally, the first indication information may indicate the target length of the first compression vector in an explicit or implicit manner.

[0160] For example, when the first indication information implicitly indicates the target length of the first compression vector, the first indication information can indicate the position of the non-zero component in the first compression vector, so that the network device can restore the first compression vector according to the position of the non-zero component and determine the target length of the first compression vector according to the restored first compression vector.

[0161] For example, the first compression vector reported by the terminal device contains "11111", and the first indication information indicates the position of each 1. For instance, if the first indication information indicates that the 1 is located in the 3rd, 4th, 6th, 8th, and 11th positions respectively, then it can be determined that the current first compression vector is actually {0011 0101 0010}. Furthermore, the network device can determine the target length of the first compression vector based on this actual first compression vector.

[0162] Alternatively, when the first indication information explicitly indicates the target length of the first compression vector, the first indication information can directly indicate the length value, such as X bits.

[0163] S305, the network device determines the target length of the first compression vector and the target output layer associated with the target length based on the first instruction information.

[0164] S306, the network device decodes the first compression vector based on the model algorithm corresponding to the target output layer and outputs the predicted channel state information (CSI).

[0165] Optionally, the decoder deployed on the network device has K output layers, each associated with a different model algorithm, and each output layer can process compressed vectors of different lengths, where K is a positive integer. In other words, each output layer can correspond to a different prediction window length. The network device can determine the prediction window length based on the target length of the first compressed vector, and then select a target output layer that fits the prediction window length, outputting the index of the target output layer. This allows the model to call the algorithm of the target output layer corresponding to that index to perform the prediction process and obtain the CSI.

[0166] Figure 6 This is a schematic diagram illustrating a complete process of CSI inference based on an AI model, provided in an embodiment of this application. Figure 6 The process shown can correspond to the aforementioned Figure 3 S301-S306 in the series.

[0167] For example, such as Figure 6 As shown, the process described above can be based on Figure 6 The processing is implemented using several processing modules, which may include an input module and an encoder on the terminal device side, and a selector and a decoder on the network device side. The decoder may have K output layers. The selector is used to select a target output layer from the K output layers of the decoder based on the target length of the first compression vector.

[0168] Specifically, the process includes the following: (1) The terminal device receives the CSI RS in time slot 1-T sent by the network device. The terminal device can estimate the channel matrix H based on the time series of the CSI RS: And input the entire row into the matrix into the encoder. (2) The encoder can output a compressed vector of the target length based on the channel matrix H. and compress the vector and compression vector The first instruction information is reported to the network equipment; (3) The network device receives the compressed vector and compression vector After receiving the first instruction information, based on the compression vector The first indication information determines the compression vector The target length, respectively, will compress the vector. The input decoder compresses the vector. The target length is input into the selector. (4) The selector is based on the compression vector. Includes temporal correlation information, decision prediction window length, and the output layer index of the output decoder. ; (5) The decoder uses the index of the selector. The corresponding output layer model algorithm predicts the CSI of the future prediction window, denoted as a high-dimensional vector. : .

[0169] Through the above scheme, the embodiments of this application can realize joint compression and prediction functions based on the TCN variable output encoder and decoder architecture, and can adjust the length of the compression vector according to the temporal correlation of the channel matrix, thereby determining the length of the prediction window. In other words, the joint compression and prediction described above can compress and predict a series of temporally continuous channel matrices H as a whole, and can determine the length of the compression vector and the size of the prediction window based on their temporal correlation.

[0170] Furthermore, during the prediction process, a matching output layer can be selected from the K output layers corresponding to the decoder based on the length of the compressed vector. The model algorithm associated with this output layer is then used for prediction. In other words, the variable-length CSI feedback information is directly decompressed and restored to CSI prediction information, which improves the matching and accuracy of the prediction algorithm.

[0171] It should be understood that network devices can accurately perform beamforming on terminal devices based on the predicted CSI, that is, determine which signal to send to the terminal device and in which direction, thereby suppressing interference from multiple users and improving downlink speed, coverage, and communication reliability. Furthermore, network devices can also use the predicted CSI to allocate time-frequency resources to terminal devices, assess current compression accuracy, and determine the strength of channel time-varying characteristics, etc., but this application does not limit these aspects.

[0172] In addition, regarding such Figure 6 The CSI inference model shown in this application embodiment also provides a model training method that can improve the model's adaptability to different scenarios. The training process of the model is described below.

[0173] In one possible implementation, the network device can send first configuration information to the terminal device, and correspondingly, the terminal device can receive the first configuration information from the network device. This first configuration information is used to configure the model parameters of the encoder, decoder, and selector. For example, the first configuration information can configure the lengths of K different prediction windows, where K is a positive integer. The window length can be denoted as... That is, the first configuration information is used to configure K items. Each The values ​​can correspond to different encoder output layers. The terminal device can initialize K output layers according to the first configuration information, so that the terminal device can train the encoder, decoder and selector based on training samples.

[0174] In this embodiment, the "first configuration information" may include one or more parameters related to the AI / ML model configured by the network device for the terminal device, used by the terminal device to perform CSI prediction based on the AI / ML model. For example, Figures 5 to 7 In the encoder, “θ” is used to represent encoder-related model parameters, such as weight information and bias parameters, and “Ψ” is used to represent decoder-related model parameters. The embodiments of this application do not limit the content of the first configuration information.

[0175] Optionally, the first configuration information may be carried in radio resource control (RRC) signaling sent by the network device to the terminal device, such as RRC ReconfigurationComplete signaling.

[0176] It should be understood that the model training process can be completed on the terminal device, which is different from the aforementioned Figure 3The model described herein can be trained entirely on the terminal device. Training of different modules, such as the decoder and selector, can be completed on the terminal device. Once training is complete, the terminal device can send second configuration information to the network device. This second configuration information indicates the adjusted model parameters of the decoder and selector. The network device can update the relevant model parameters of the decoder and selector based on the second configuration information.

[0177] Optionally, the second configuration information may be RRC signaling sent by the network device to the terminal device, etc., which is not limited in this embodiment of the application.

[0178] In one possible implementation, the terminal device can acquire training samples, which include input values ​​of multiple channel matrices and predicted ground values ​​corresponding to the multiple channel matrices respectively; and train the encoder of the terminal device based on the training samples.

[0179] Optionally, the training dataset on the terminal device side can be constructed based on the channel matrix H estimated by the CSI RS sent by the network device side, and each training sample can contain input and output: Input: The input value is used to input the model on the terminal device side to start the training process. The input value can be a CSI sequence that has been actually measured over a period of time.

[0180] Truth value (output): The true value can be understood as the "standard answer", which is the real channel obtained accurately through CSI RS. Based on the true value and the model prediction value, the prediction error is calculated, thereby continuously correcting and fine-tuning the model to make the predicted value output by the model more accurate.

[0181] Figure 7 This is a schematic diagram of the training process of an example model provided in an embodiment of this application.

[0182] In one possible implementation, the encoder on the terminal device side can output a second compressed vector based on the input in the input training samples, that is, for any second channel matrix among multiple channel matrices, and then transmit the second compressed vector to the network device side.

[0183] Furthermore, the K Channel State Information (CSI) values ​​predicted by the K output layers of the decoder after performing decoding processing on the length of the second compression vector can be obtained; and the model parameters of the encoder, decoder, and selector can be adjusted based on the loss function between the predicted K CSI values ​​and the predicted ground truth values ​​corresponding to the second channel matrix.

[0184] For example, such as Figure 7 As shown, the model training process can be based on and Figure 6 The architecture is the same, consisting of an input module, encoder, selector, and decoder, where the decoder can have K output layers. Specifically, it includes the following processes: (1) For each CSI sequence in the input training dataset, the terminal device can estimate the channel matrix H based on the time series of each CSI RS: And input the entire row into the matrix into the encoder. (2) The encoder can output a second compression vector based on each channel matrix H. and the second compression vector and compression vector The first instruction information is reported to the network equipment; (3) The network device receives the second compression vector and compression vector After receiving the first instruction information, based on the compression vector The first indication information determines the second compression vector The length of the second compression vector is respectively... Input the K output layers of the decoder to obtain K CSIs output by the K output layers; (4) Based on the loss function between the K CSIs and the predicted true values ​​corresponding to the second channel matrix in the training samples, the model parameters of the encoder, decoder and selector are adjusted, such as the weights and bias parameters of the encoder, decoder and selector.

[0185] During the training process, the high-dimensional vector of the output can be obtained according to the following formula (9). The loss function between the true value (output) and the actual value:

[0186] Formula (9) The loss function of the above formula (9) consists of two parts. The first part is the selection probability of each output layer. The corresponding prediction errors The weighted summation, this part is included in the cost item. This can prevent the model from blindly selecting a shorter window. The prediction window length corresponding to the output layer k is available. Function fitting.

[0187] Among them, prediction error The calculation can be performed according to the following formula (10): Formula (10) It should be understood that for each input, there is only one corresponding output layer, and the prediction error of that output layer is guaranteed. Minimum, meaning the best predictive performance.

[0188] In formula (9) It is a cost item, and .in, The length of the prediction window is determined by the length of the output layer; the smaller the prediction window, the better. The larger the value, the better; conversely, the longer the prediction window, the better. The smaller. It should be understood that this cost item... It can be used to balance prediction errors and prevent the model from blindly selecting the output layer with the shortest length.

[0189] The second part of the Loss function in the above formula (9) is a variable-length compressed vector. Feedback overhead cost , measured in bits or effective vector length. That is, a variable-length compressed vector. The bit length is related to the "time correlation" mentioned above. A stronger time correlation results in a longer bit length, while a weaker time correlation results in a shorter bit length.

[0190] In variable-length compression vectors In its implementation, this scheme adopts a gated output dimension method: the encoder output actually consists of two parts. , , , , An encoder can... Each component has an "importance" evaluation logits, which is the parameter position index during training. , At this time, the cost of expenses It can be fitted as ,in, Right now Each component, The number of quantization bits (a preset constant).

[0191] In the initial training phase, the temperature parameter t can be set relatively large, and then gradually decreased. A larger t results in a smoother sigmoid output; a smaller t results in an output of 0 or 1, meaning that during inference, the output is directly binarized. .

[0192] The input to the decoder during training is During deployment, variable-length compression vectors also need to be processed. To perform a "remove zero" operation, use Indicator index, Decoder receives ,according to Padding with zeros will yield a high-dimensional vector. .

[0193] Through the training process described above, the network device decodes the compressed vector using the K output layers of the decoder, outputting the predicted channel state information (CSI) corresponding to the training samples. Based on the loss function between the predicted CSI and the ground truth values ​​corresponding to multiple channel matrices, the K output layers of the encoder and the K output layers of the decoder of the terminal device are fine-tuned. In other words, for each CSI sequence in the training samples, the model parameters and related configuration parameters of the encoder, decoder, and selector can be continuously optimized and adjusted based on the loss function between the predicted value obtained from each CSI sequence in the input training samples and the ground truth values ​​in the training samples, thereby improving the model's prediction performance.

[0194] In summary, the embodiments of this application can achieve joint compression and prediction functions based on the TCN's variable output encoder and decoder architecture, and can adjust the length of the compression vector according to the temporal correlation of the channel matrix, thereby determining the length of the prediction window. In other words, the joint compression and prediction described above can compress and predict a series of temporally continuous channel matrices H as a whole, and can determine the length of the compression vector and the size of the prediction window based on their temporal correlation. Based on the temporal correlation of the channel matrix, different prediction windows can be adapted, that is, different prediction algorithms can be selected according to the temporal correlation. This process can adapt to different scenarios, and different AI / ML models do not need to be called for different scenarios, reducing local storage overhead and signaling overhead caused by model switching.

[0195] Furthermore, during the prediction process, a matching output layer can be selected from the K output layers corresponding to the decoder based on the length of the compressed vector. The model algorithm associated with this output layer is then used for prediction. In other words, the variable-length CSI feedback information is directly decompressed and restored to CSI prediction information, which improves the matching and accuracy of the prediction algorithm.

[0196] During the training phase, network devices can pre-configure multiple prediction windows of different lengths. The model directly learns the temporal correlation between the prediction window and the CSI (Constant Sign Indicator Sequence). Based on the loss function between the predicted value obtained from each CSI sequence in the input training samples and the ground truth value in the training samples, the model continuously optimizes and adjusts each model parameter and related configuration parameter of the encoder, decoder, and selector, thereby improving the model's prediction performance. This process can adapt to different temporal correlations in different scenarios, greatly reducing storage requirements and minimizing signaling overhead caused by model switching.

[0197] It should be understood that the process described above can be applied to scenarios of short-term prediction and shared perception in autonomous driving or V2X, such as compressing the intermediate representation after perception fusion into... The selector can determine the amount of data to send based on the current speed, scenario complexity, and link quality. With prediction window In emergency situations, more information and longer-term forecasts can be sent first, allowing the cloud or other vehicles to receive them and make more accurate trajectory predictions, collision warnings, or environmental mapping.

[0198] It can also be applied to event-driven reporting and prediction scenarios in low-power IoT or sensor networks. For example, sensors can locally compress short-term historical data as necessary, and the selector can decide on the appropriate selection based on event importance, remaining power, and communication budget. (Send a more detailed summary or only send a coarse alarm); the network-side decoder can use the received partial information to confirm the event, predict the future state, or trigger actions, etc., but this application embodiment does not limit this.

[0199] It should be understood that Figures 1 to 7 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 1 to 7 The examples in the document can be transformed into equivalent ways to obtain more implementations.

[0200] The above text combined Figures 1 to 7 This document describes in detail the communication method provided in the embodiments of this application. The following will combine... Figures 8 to 9The 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 communication 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.

[0201] In the embodiments described above, the terminal device may execute some or all of the steps in each embodiment; the network device may execute some or all of the steps in each embodiment. These steps or operations are merely examples, and the embodiments of this application may also perform other operations or variations thereof. Furthermore, the steps may 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.

[0202] Figure 8 This is a schematic block diagram of a communication device provided in an embodiment of this application. Figure 8 As shown, the communication device 800 may include a communication unit 820. The communication unit 820 can implement corresponding communication functions, which can be internal communication functions of the communication device 800 or communication functions between the communication device 800 and other devices. Optionally, the communication unit 820 may also be referred to as a communication interface or transceiver unit.

[0203] Optionally, the communication device 800 may further include a processing unit 810, which can perform corresponding processing functions.

[0204] Optionally, the communication device 800 may further include a storage unit, which can be used to store instructions and / or data; the processing unit 810 can read the instructions and / or data in the storage unit so that the communication device 800 can implement the aforementioned method embodiments.

[0205] In one possible design, the communication device 800 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 800 can be used to execute the steps or processes performed by the terminal device in any of the above method embodiments.

[0206] Alternatively, the communication device 800 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 800 may be used to execute the steps or processes performed by the network device in any of the above method embodiments.

[0207] For example, the communication unit 820 is used to implement Figure 3The processes S301 and S304 described herein are implemented by the processing unit 810. Figure 3 The processes S302 and S303 described herein, or those used to implement Figure 3 For the sake of simplicity, the relevant processes of S305 and S306 will not be described in detail here.

[0208] The above are merely examples; for detailed steps or procedures, please refer to the descriptions in the foregoing embodiments.

[0209] Figure 9 This is another schematic block diagram of the communication device 900 provided in the embodiments of this application. The communication device 900 may be a chip, chip system, or processor, etc., in a terminal device or network device that implements the above-described methods. The communication device 900 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.

[0210] like Figure 9 As shown, the communication device 900 may include one or more processors 910, which may also be referred to as processing units or processing modules, and can implement certain control functions. The processor 910 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 900 (e.g., a base station, baseband chip, user, user chip), execute software programs, and process data from the software programs.

[0211] In an alternative design, the processor 910 may also store instructions and / or data, which can be executed by the processor 910 to cause the communication device 900 to perform the methods described in the above method embodiments.

[0212] In another alternative design, the communication device 900 may include a communication interface 920 for implementing receiving and transmitting functions. For example, the communication interface 920 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.

[0213] Optionally, the communication device 900 may include one or more memories 930, which may store instructions that can be executed on the processor 910 to cause the communication device 900 to perform the methods described in the above method embodiments. Optionally, the memories 930 may also store data.

[0214] Optionally, instructions and / or data may also be stored in the processor 910. The processor 910 and the memory 930 may be configured separately or integrated together.

[0215] 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.

[0216] In one implementation, the communication device 900 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 910 may be used to execute instructions stored in the memory 930, and when the processor 910 executes the instructions stored in the memory, the processor 910 is used to execute the various steps and / or processes of the above method embodiments corresponding to the terminal device.

[0217] In another implementation, the communication device 900 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 910 may be used to execute instructions stored in the memory 930, and when the processor 910 executes the instructions stored in the memory, the processor 910 is used to execute the various steps and / or processes of the above method embodiments corresponding to the network device.

[0218] 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.

[0219] 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.

[0220] 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.

[0221] The chip system may include input circuits or interfaces for transmitting information or data, and output circuits or interfaces for receiving information or data.

[0222] 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.

[0223] According to the method provided in the embodiments of this application, this application also provides a computer program product, which includes: computer program code, which, when run on a computer, causes the computer to execute the various steps or processes executed by the network device or terminal device in any of the foregoing method embodiments.

[0224] According to the method provided in the embodiments of this application, this application also provides a computer-readable storage medium storing program code, which, when run on a computer, causes the computer to execute the various steps or processes executed by the network device or terminal device in any of the foregoing method embodiments.

[0225] 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.

[0226] 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.

[0227] 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. A computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated.

[0228] 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.

[0229] 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.

[0230] In summary, the above are merely preferred embodiments of the technical solutions of this application and are 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 a terminal device, the method includes: Receive Channel State Information Reference Signal (CSI RS) from network devices, and obtain the channel matrix and time series information of the CSI RS; Output a first compressed vector with a target length, the first compressed vector being obtained by encoding the channel matrix based on the time series information, the first compressed vector being variable in length, and the target length being determined based on the time series information; The network device sends a first compression vector and first indication information. The first indication information is used to indicate the target length of the first compression vector. The target length of the first compression vector is used to determine the target output layer. The model algorithm corresponding to the target output layer is used to perform decoding processing on the first compression vector to obtain the predicted channel state information (CSI). The decoder of the network device has K output layers, and the target output layer is selected from the K output layers.

2. The method according to claim 1, characterized in that, The first compression vector is obtained by the encoder of the temporal convolutional network performing the encoding process on the channel matrix based on the time series information, and the encoder of the temporal convolutional network is deployed on the terminal device.

3. The method according to claim 2, characterized in that, The first compressed vector is obtained by performing a masking operation on the feature vector with the maximum length after the encoder of the temporal convolutional network performs time-series modeling on the channel matrix and outputs a gated vector, and then performing a concatenation process on the intermediate vector and the gated vector to obtain an intermediate vector.

4. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Obtain training samples, which include multiple channel matrices and the predicted ground values ​​corresponding to the multiple channel matrices respectively; Based on the training samples, the encoder, decoder, and selector are trained. The decoder corresponds to K output layers. Each of the K output layers is associated with a different model algorithm and each output layer can process compression vectors of different lengths. The selector is used to select the target output layer from the K output layers according to the target length of the first compression vector, where K is a positive integer.

5. The method according to claim 4, characterized in that, The method further includes: Receive first configuration information from the network device, the first configuration information being used to configure the model parameters of the encoder, the decoder and the selector; Based on the first configuration information, the encoder is initialized so that the encoder can output a second compression vector based on any one of the second channel matrices included in the input training samples.

6. The method according to claim 5, characterized in that, The training of the encoder, decoder, and selector based on the training samples includes: After the K output layers of the decoder perform decoding processing on the length of the second compression vector, the K channel state information (CSI) are predicted. Based on the loss function between the predicted K CSIs and the predicted ground truth values ​​corresponding to the second channel matrix, the model parameters of the encoder, the decoder, and the selector are adjusted.

7. The method according to claim 5 or 6, characterized in that, The method further includes: Send second configuration information to the network device, the second configuration information being used to indicate the adjusted model parameters of the decoder and the selector.

8. A communication method, characterized in that, Applied to network devices, the method includes: Send Channel State Information Reference Signal (CSI RS) to the terminal device; The terminal device receives a first compression vector and a first indication information. The first compression vector is obtained by encoding the channel matrix of the CSI RS according to the time series information of the CSI RS. The first indication information is used to indicate the target length of the first compression vector. The first compression vector is variable in length, and the target length is determined based on the time series information. The predicted channel state information (CSI) is output. The CSI is obtained by decoding the first compressed vector based on the model algorithm corresponding to the target output layer. The target output layer is determined according to the target length of the first compressed vector. The decoder of the network device has K output layers, and the target output layer is selected from the K output layers.

9. The method according to claim 8, characterized in that, Each output layer is associated with a different model algorithm and each output layer can process compressed vectors of different lengths, where K is a positive integer.

10. The method according to claim 8 or 9, characterized in that, The first compression vector is obtained by performing time-series modeling on the channel matrix to output the feature vector and gate vector of the maximum length, performing a masking operation on the feature vector based on the gate vector to obtain an intermediate vector, and then performing concatenation processing on the intermediate vector and the gate vector.

11. The method according to claim 10, characterized in that, The method further includes: Send first configuration information to the terminal device. The first configuration information is used to configure the model parameters of the encoder, decoder and selector, and is used to initialize the encoder so that the encoder can train the encoder, the decoder and the selector based on the input training samples. The training samples include multiple channel matrices and the predicted ground truth values ​​corresponding to the multiple channel matrices.

12. The method according to claim 11, characterized in that, The selector is used to select the target output layer from the K output layers of the decoder based on the target length of the first compression vector.

13. The method according to claim 11 or 12, characterized in that, The method further includes: The device receives second configuration information from the terminal device, the second configuration information being used to indicate the model parameters of the decoder and the selector after training and adjustment.

14. A communication device, characterized in that, The communication device includes a module or unit for performing the method according to any one of claims 1 to 13.

15. A communication device, characterized in that, The device includes a processor and an interface circuit, the interface circuit being used to receive signals from other devices and transmit them to the processor or to send signals from the processor to other devices, the processor being used to implement the method as described in any one of claims 1 to 13 via logic circuits or executing code instructions.

16. A system, characterized in that, The system includes: a network device for performing the method of any one of claims 1 to 7 and a terminal device for performing the method of any one of claims 8 to 13.

17. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when the code corresponding to the computer program is run on a computer, enables the computer to execute the communication method according to any one of claims 1 to 13.

Images