WO2023088138A1 - 信道状态信息处理方法及装置 - Google Patents

信道状态信息处理方法及装置 Download PDF

Info

Publication number
WO2023088138A1
WO2023088138A1 PCT/CN2022/130632 CN2022130632W WO2023088138A1 WO 2023088138 A1 WO2023088138 A1 WO 2023088138A1 CN 2022130632 W CN2022130632 W CN 2022130632W WO 2023088138 A1 WO2023088138 A1 WO 2023088138A1
Authority
WO
WIPO (PCT)
Prior art keywords
communication device
channel state
state information
information
graph
Prior art date
Application number
PCT/CN2022/130632
Other languages
English (en)
French (fr)
Inventor
李梦圆
李榕
王坚
Original Assignee
华为技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Publication of WO2023088138A1 publication Critical patent/WO2023088138A1/zh
Priority to US18/666,833 priority Critical patent/US20240322883A1/en

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
    • H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
    • H04B7/06—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station
    • H04B7/0613—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station using simultaneous transmission
    • H04B7/0615—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station using simultaneous transmission of weighted versions of same signal
    • H04B7/0619—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station using simultaneous transmission of weighted versions of same signal using feedback from receiving side
    • H04B7/0621—Feedback content
    • H04B7/0626—Channel coefficients, e.g. channel state information [CSI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B17/00—Monitoring; Testing
    • H04B17/30—Monitoring; Testing of propagation channels
    • H04B17/391—Modelling the propagation channel
    • H04B17/3913—Predictive models, e.g. based on neural network models
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
    • H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
    • H04B7/0413—MIMO systems
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
    • H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
    • H04B7/06—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/08—Modifications for reducing interference; Modifications for reducing effects due to line faults ; Receiver end arrangements for detecting or overcoming line faults

Definitions

  • the present application relates to the field of communications, and in particular to a method and device for processing channel state information.
  • a terminal device needs to report channel state information (CSI) to a network device periodically or aperiodically.
  • the CSI may include at least one CSI report, and the CSI report is used to indicate the channel state information of the downlink channel.
  • the terminal device may acquire and report the above at least one CSI report by measuring the downlink reference signal sent by the network device.
  • the network device may allocate corresponding downlink transmission resources to the terminal device according to the CSI report reported by the terminal device.
  • the current terminal equipment usually does not report the actual CSI to the network equipment, but indicates the precoding matrix through the precoding matrix indicator (PMI) transmitted in the CSI report, so as to realize the precoding and downlink of the network equipment side data transmission.
  • the precoding matrix indicated by the PMI is not the actual precoding matrix corresponding to the CSI, thus causing communication performance loss.
  • Embodiments of the present application provide a channel state information processing method and device, which can ensure the effectiveness of CSI feedback and improve communication performance.
  • the present application provides a method for processing channel state information, and the method can be applied to a first communication device, where the first communication device is, for example, a terminal device or a network device.
  • the method includes: firstly, the first communication device determines the first graph model corresponding to the first channel state information, and the first channel state information is the channel state information from the second communication device to the first communication device; then, the first communication device Using the first neural network to process the first graph model to obtain first information, the first information is used by the second communication device to restore the first channel state information; finally, the first communication device sends the first information to the second communication device.
  • the first communication device can retain important feature information in the first graphical model and omit unimportant feature information in the first graphical model during the process of using the first neural network to process the first graphical model.
  • Feature information using the first neural network to compress the first channel state information, reducing the CSI feedback overhead while ensuring the effectiveness of CSI feedback, so that the channel state information recovered by the second communication device is closer to the original channel, thereby improving communication performance .
  • the first communication device determining the first graph model corresponding to the first channel state information includes: the first communication device determining the first graph model corresponding to the first channel state information according to the transmitting and receiving antenna configuration information.
  • the transmitting and receiving antenna configuration information includes receiving antenna configuration information of the first communication device and transmitting antenna configuration information of the second communication device.
  • the first information includes auxiliary information
  • the auxiliary information is used to determine the second graph model
  • the second graph model is used to restore the first channel state information.
  • the auxiliary information can help the second communication device recover the first channel state information, and improve the accuracy of the first channel state information recovered by the second communication device.
  • the auxiliary information includes index information of some nodes of the first graph model. It can be understood that when the first communication device uses the first neural network to process the first graph model, unimportant feature information in the first graph model will be omitted, for example, feature information of some nodes in the first graph model will be omitted.
  • the first communication device may use the auxiliary information to indicate the discarded nodes in the first graph model, or use the auxiliary information to indicate the retained nodes in the first graph model, which can improve the accuracy of the second communication device in recovering the first channel state information .
  • auxiliary information may also include mean values of features of some nodes. In this way, it can further help the second communication device recover the first channel state information, and improve the accuracy of the first channel state information recovered by the second communication device.
  • the first information includes second channel state information and auxiliary information
  • the second channel state information includes processed first channel state information
  • the first neural network may include a first image pooling layer and a compression layer.
  • the first graph pooling layer is used to determine the third graph model and auxiliary information according to the first graph model
  • the compression layer is used to determine the second channel state information according to the third graph model.
  • the input of the first graph pooling layer may be graph models of different structures, that is to say, for the first communication device and the second communication device with different antenna configurations (hereinafter collectively referred to as the two ends of sending and receiving CSI), the first Each communication device may determine a first graph model corresponding to the first channel state information, and input the first graph model into the same first neural network.
  • the design and training of the first neural network does not depend on the configuration of the antennas at both ends of the sending and receiving CSI. In this way, when the configuration of the antennas at both ends of the sending and receiving CSI changes, redesign and training of the neural network can be avoided, and it is suitable for dynamic multiple input multiple output (MIMO) systems or multiple MIMO systems with different antenna configurations.
  • MIMO multiple input multiple output
  • the first graph pooling layer may be a graph pooling layer using a self-attention mechanism. Since the graph pooling layer of the self-attention mechanism needs fewer parameters to be trained, and can fully extract the feature information of the nodes of the first graph model and the topology information of the graph, the training efficiency and processing capacity of the neural network can be improved.
  • the first neural network may include a first graph convolution layer, a first graph pooling layer, and a compression layer.
  • the first graph convolution layer is used to determine the first graph model after convolution according to the first graph model
  • the first graph pooling layer is used to determine the third graph model and auxiliary information according to the first graph model after convolution
  • the compression layer is used to determine the second channel state information according to the third graph model.
  • the design and training of the first neural network does not depend on the configuration of the antennas at both ends of the sending and receiving CSI. In this way, when the configurations of the antennas at both ends of the sending and receiving CSI change, redesign and training of the neural network can be avoided, and it is applicable to a dynamic MIMO system or multiple MIMO systems with different antenna configurations.
  • the compression layer can be a fully connected layer.
  • the first channel state information includes delay-angle domain channel state information between the transmitting antenna of the second communication device and the receiving antenna of the first communication device.
  • the delay-angle domain channel state information includes channel state information between R receiving angles and T transmitting angles, R and T are both positive integers, and at least one of R and T is greater than 1.
  • the first graph model includes multiple nodes, and the characteristics of the nodes include channel state information between the i-th transmitting angle and the j-th receiving angle, 1 ⁇ i ⁇ T, 1 ⁇ j ⁇ R.
  • the first graph model also includes at least one edge. Wherein, each edge is connected to two nodes, and each edge indicates that the two nodes connected to the edge correspond to two adjacent receiving angles or correspond to two adjacent transmitting angles.
  • the channel state information included in the characteristics of each node includes C elements in the time domain, where C is less than or equal to the number of subcarriers.
  • C due to the sparsity of the channel state information in the time domain, appropriately reducing the number of elements included in the channel state information included in the characteristics of each node in the time domain will not affect the accuracy of the CSI.
  • the data processing amount of the communication device can be reduced and the processing rate of the communication device can be increased without affecting the accurate recovery of the first channel state information by the second communication device.
  • the receiving antenna configuration information of the first communication device includes one or more of the following: the number of receiving antennas of the first communication device, the type of receiving antenna front, and the arrangement of receiving antenna units.
  • the transmitting antenna configuration information of the second communication device includes one or more of the following items: the number of transmitting antennas of the second communication device, the type of transmitting antenna array, and the arrangement of transmitting antenna units.
  • the first communication device determining the first channel state information includes: the first communication device according to the space-frequency domain channel state information between the transmitting antenna of the second communication device and the receiving antenna of the first communication device , to determine the first channel state information.
  • the first channel state information is delay-angle domain channel state information.
  • the first neural network is determined according to the training data set, and the training data set includes a plurality of fourth channel state information for the first neural network, and the fourth channel state information is information from the second communication device to the first Channel state information of the communication device.
  • the present application provides a method for processing channel state information, and the method can be applied to a second communication device, where the second communication device is, for example, a network device or a terminal device.
  • the method includes: firstly, the second communication device receives the first information sent by the first communication device, the first information is used for the second communication device to restore the first channel state information, and the first channel state information is the communication between the second communication device and the The channel state information of the first communication device. Then, the second communication device uses the second neural network to process the first information to obtain the second graphical model. Finally, the second communication device determines third channel state information according to the second graph model, where the third channel state information is the recovered first channel state information.
  • the determining, by the second communications device, the third channel state information according to the second map model includes: determining, by the second communications device, the third channel state according to the second map model and the configuration information of the transmitting and receiving antennas.
  • the transmitting and receiving antenna configuration information includes receiving antenna configuration information of the first communication device and transmitting antenna configuration information of the second communication device.
  • the first information includes auxiliary information
  • the auxiliary information is used to determine the second graph model.
  • the auxiliary information includes index information of some nodes of the first graph model, where the first graph model is a graph model corresponding to the first channel state information.
  • auxiliary information may also include mean values of features of some nodes.
  • the first information includes second channel state information and auxiliary information
  • the second channel state information includes processed first channel state information
  • the second neural network may include a decompression layer and a second graph convolution layer.
  • the decompression layer is used to determine the fourth graph model according to the second channel state information
  • the second graph convolution layer is used to determine the second graph model according to the auxiliary information and the fourth graph model.
  • the second graph convolutional layer may include multiple graph convolutional layers, and direct connection channels are included between the multiple graph convolutional layers. In this way, the training efficiency and stability of the neural network can be improved.
  • the decompression layer can be a fully connected layer.
  • the first channel state information includes delay-angle domain channel state information between the transmitting antenna of the second communication device and the receiving antenna of the first communication device.
  • the delay-angle domain channel state information includes channel state information between R receiving angles and T transmitting angles, R and T are both positive integers, and at least one of R and T is greater than 1.
  • the second graph model includes a plurality of nodes, and the characteristics of the nodes include channel state information between the i-th transmitting angle and the j-th receiving angle, 1 ⁇ i ⁇ T, 1 ⁇ j ⁇ R.
  • the second graph model also includes at least one edge. Among them, each edge is connected to two nodes, and each edge indicates that the two nodes connected to the edge correspond to two adjacent receiving angles or correspond to two adjacent transmitting angles.
  • the channel state information included in the characteristics of each node includes C elements in the time domain, where C is less than or equal to the number of subcarriers.
  • the receiving antenna configuration information of the first communication device includes one or more of the following: the number of receiving antennas of the first communication device, the type of receiving antenna front, and the arrangement of receiving antenna units.
  • the transmitting antenna configuration information of the second communication device includes one or more of the following items: the number of transmitting antennas of the second communication device, the type of transmitting antenna array, and the arrangement of transmitting antenna units.
  • the second neural network is determined according to the training data set, the training data set includes a plurality of fourth channel state information for the second neural network, and the fourth channel state information is the information from the second communication device to the first Channel state information of the communication device.
  • a first communication device in a third aspect, includes a processing module and a transceiver module.
  • the processing module is configured to determine the first graph model corresponding to the first channel state information, where the first channel state information is the channel state information from the second communication device to the first communication device.
  • the processing module is further configured to use the first neural network to process the first graphical model to obtain first information, and the first information is used by the second communication device to restore the first channel state information.
  • the transceiver module is configured to send the first information to the second communication device.
  • the processing module is further configured to determine the first graph model corresponding to the first channel state information according to the configuration information of the transmitting and receiving antennas.
  • the transmitting and receiving antenna configuration information includes receiving antenna configuration information of the first communication device and transmitting antenna configuration information of the second communication device.
  • the first information includes auxiliary information
  • the auxiliary information is used to determine the second graph model
  • the second graph model is used to restore the first channel state information.
  • the auxiliary information includes index information of some nodes of the first graph model.
  • auxiliary information may also include mean values of features of some nodes.
  • the first information includes second channel state information and auxiliary information
  • the second channel state information includes processed first channel state information
  • the first neural network may include a first image pooling layer and a compression layer.
  • the first graph pooling layer is used to determine the third graph model and auxiliary information according to the first graph model
  • the compression layer is used to determine the second channel state information according to the third graph model.
  • the first graph pooling layer may be a graph pooling layer using a self-attention mechanism.
  • the first neural network may include a first graph convolution layer, a first graph pooling layer, and a compression layer.
  • the first graph convolution layer is used to determine the first graph model after convolution according to the first graph model
  • the first graph pooling layer is used to determine the third graph model and auxiliary information according to the first graph model after convolution
  • the compression layer is used to determine the second channel state information according to the third graph model.
  • the compression layer can be a fully connected layer.
  • the first channel state information includes delay-angle domain channel state information between the transmitting antenna of the second communication device and the receiving antenna of the first communication device.
  • the delay-angle domain channel state information includes channel state information between R receiving angles and T transmitting angles, R and T are both positive integers, and at least one of R and T is greater than 1.
  • the first graph model includes a plurality of nodes, and the characteristics of the nodes include the channel state information between the i-th transmitting angle and the j-th receiving angle, 1 ⁇ i ⁇ T, 1 ⁇ j ⁇ R.
  • the first graph model also includes at least one edge. Wherein, each edge is connected to two nodes, and each edge indicates that the two nodes connected to the edge correspond to two adjacent receiving angles or correspond to two adjacent transmitting angles.
  • the channel state information included in the characteristics of each node includes C elements in the time domain, where C is less than or equal to the number of subcarriers.
  • the receiving antenna configuration information of the first communication device includes one or more of the following: the number of receiving antennas of the first communication device, the type of receiving antenna front, and the arrangement of receiving antenna units.
  • the transmitting antenna configuration information of the second communication device includes one or more of the following items: the number of transmitting antennas of the second communication device, the type of transmitting antenna array, and the arrangement of transmitting antenna units.
  • the processing module is further configured to determine the first channel state information according to space-frequency domain channel state information between the transmitting antenna of the second communication device and the receiving antenna of the first communication device.
  • the first channel state information is delay-angle domain channel state information.
  • the first neural network is determined according to the training data set, and the training data set includes a plurality of fourth channel state information for the first neural network, and the fourth channel state information is information from the second communication device to the first Channel state information of the communication device.
  • the transceiver module may include a receiving module and a sending module.
  • the receiving module is used to realize the receiving function of the first communication device described in the third aspect
  • the sending module is used to realize the sending function of the first communication device described in the third aspect.
  • the first communication device may further include a storage module, where programs or instructions are stored in the storage module.
  • the processing module executes the program or instruction
  • the first communication device can execute the method described in the first aspect.
  • the first communication device described in the third aspect may be a terminal device or a network device, or a chip (system) or other components or components set in a terminal device or a network device, or may include a terminal A device or a communication device of a network device, which is not limited in this application.
  • a second communication device in a fourth aspect, includes: a processing module and a transceiver module.
  • the transceiver module is used to receive the first information sent by the first communication device, the first information is used by the second communication device to restore the first channel state information, and the first channel state information is the information from the second communication device to the first communication device.
  • Channel state information of a communication device The processing module is configured to use the second neural network to process the first information to obtain the second graphical model.
  • the processing module is further configured to determine third channel state information according to the second graph model, where the third channel state information is the recovered first channel state information.
  • the processing module is further configured to determine a third channel state according to the second graph model and configuration information of the transmitting and receiving antennas.
  • the transmitting and receiving antenna configuration information includes receiving antenna configuration information of the first communication device and transmitting antenna configuration information of the second communication device.
  • the first information includes auxiliary information
  • the auxiliary information is used to determine the second graph model.
  • the auxiliary information includes index information of some nodes of the first graph model, where the first graph model is a graph model corresponding to the first channel state information.
  • auxiliary information may also include mean values of features of some nodes.
  • the first information includes second channel state information and the auxiliary information
  • the second channel state information includes processed first channel state information
  • the second neural network may include a decompression layer and a second graph convolution layer.
  • the decompression layer is used to determine the fourth graph model according to the second channel state information
  • the second graph convolution layer is used to determine the second graph model according to the auxiliary information and the fourth graph model.
  • the second graph convolutional layer may include multiple graph convolutional layers, and direct connection channels are included between the multiple graph convolutional layers.
  • the decompression layer can be a fully connected layer.
  • the first channel state information includes delay-angle domain channel state information between the transmitting antenna of the second communication device and the receiving antenna of the first communication device.
  • the delay-angle domain channel state information includes channel state information between R receiving angles and T transmitting angles, R and T are both positive integers, and at least one of R and T is greater than 1.
  • the second graph model includes a plurality of nodes, and the characteristics of the nodes include channel state information between the i-th transmitting angle and the j-th receiving angle, 1 ⁇ i ⁇ T, 1 ⁇ j ⁇ R.
  • the second graph model also includes at least one edge. Wherein, each edge is connected to two nodes, and each edge indicates that the two nodes connected to the edge correspond to two adjacent receiving angles or correspond to two adjacent transmitting angles.
  • the channel state information included in the characteristics of each node includes C elements in the time domain, where C is less than or equal to the number of subcarriers.
  • the receiving antenna configuration information of the first communication device includes one or more of the following: the number of receiving antennas of the first communication device, the type of receiving antenna front, and the arrangement of receiving antenna units.
  • the transmitting antenna configuration information of the second communication device includes one or more of the following items: the number of transmitting antennas of the second communication device, the type of transmitting antenna array, and the arrangement of transmitting antenna units.
  • the second neural network is determined according to the training data set, the training data set includes a plurality of fourth channel state information for the second neural network, and the fourth channel state information is the information from the second communication device to the first Channel state information of the communication device.
  • the transceiver module may include a receiving module and a sending module.
  • the receiving module is used to realize the receiving function of the second communication device described in the fourth aspect
  • the sending module is used to realize the sending function of the second communication device described in the fourth aspect.
  • the second communication device may further include a storage module, where programs or instructions are stored in the storage module.
  • the processing module executes the program or instruction
  • the second communication device can execute the method described in the second aspect.
  • the second communication device described in the fourth aspect may be a network device or a terminal device, or a chip (system) or other components or components set in the network device or terminal device, or may include a network
  • the device or the communication device of the terminal device is not limited in this application.
  • a communication device configured to execute the channel state information processing method described in any one of the implementation manners of the first aspect to the second aspect.
  • the communication device described in the fifth aspect may be the first communication device or the second communication device, or may be a chip (system) or other components or components set in the first communication device or the second communication device , may also be a communication device including a first communication device or a second communication device, which is not limited in this application.
  • the first communication device is configured to execute the channel state information processing method described in any possible implementation of the first aspect
  • the second communication device is configured to execute the method described in any possible implementation of the second aspect. A method for processing channel state information.
  • the communication device described in the fifth aspect includes corresponding modules, units, or means (means) for implementing the channel state information processing method described in any one of the first to second aspects above.
  • the means can be realized by hardware, software, or by executing corresponding software by hardware.
  • the hardware or software includes one or more modules or units for performing the functions involved in the above channel state information processing method.
  • a communication device in a sixth aspect, includes: a processor, configured to execute the channel state information processing method described in any one possible implementation manner of the first aspect to the second aspect.
  • the communication device described in the sixth aspect may further include a transceiver.
  • the transceiver may be a transceiver circuit or an interface circuit.
  • the transceiver can be used for the communication device described in the sixth aspect to communicate with other communication devices.
  • the communication device described in the sixth aspect may further include a memory.
  • the memory can be integrated with the processor or set separately.
  • the memory may be used to store computer programs and/or data involved in the channel state information processing method described in any one of the first aspect to the second aspect.
  • the communication device described in the sixth aspect may be the first communication device or the second communication device, or may be a chip (system) or other components or components set in the first communication device or the second communication device , may also be a communication device including a first communication device or a second communication device, which is not limited in this application.
  • the first communication device is configured to execute the channel state information processing method described in any possible implementation of the first aspect
  • the second communication device is configured to execute the method described in any possible implementation of the second aspect. A method for processing channel state information.
  • a communication device in a seventh aspect, includes: a processor, the processor is coupled with the memory, and the processor is used to execute the computer program stored in the memory, so that the communication device executes any one of the possible implementation manners in the first aspect to the second aspect.
  • the channel state information processing method includes: a processor, the processor is coupled with the memory, and the processor is used to execute the computer program stored in the memory, so that the communication device executes any one of the possible implementation manners in the first aspect to the second aspect.
  • the communication device described in the seventh aspect may further include a transceiver.
  • the transceiver may be a transceiver circuit or an interface circuit.
  • the transceiver can be used for the communication device described in the seventh aspect to communicate with other communication devices.
  • the communication device described in the seventh aspect may be the first communication device or the second communication device, or may be a chip (system) or other components or components set in the first communication device or the second communication device , may also be a communication device including a first communication device or a second communication device, which is not limited in this application.
  • the first communication device is configured to execute the channel state information processing method described in any possible implementation of the first aspect
  • the second communication device is configured to execute the method described in any possible implementation of the second aspect. A method for processing channel state information.
  • a communication device in an eighth aspect, includes: a processor and an interface circuit. Wherein, the interface circuit is used to receive code instructions and transmit them to the processor.
  • the processor is configured to run the above code instructions to execute the channel state information processing method described in any one of the implementation manners of the first aspect to the second aspect.
  • the communication device described in the eighth aspect may further include a memory.
  • the memory can be integrated with the processor or set separately.
  • the memory may be used to store computer programs and/or data involved in the channel state information processing method described in any one of the first aspect to the second aspect.
  • the communication device described in the eighth aspect may be the first communication device or the second communication device, or it may be a chip (system) or other components or components set in the first communication device or the second communication device , may also be a communication device including a first communication device or a second communication device, which is not limited in this application.
  • the first communication device is configured to execute the channel state information processing method described in any possible implementation of the first aspect
  • the second communication device is configured to execute the method described in any possible implementation of the second aspect. A method for processing channel state information.
  • a communication device in a ninth aspect, includes a processor and a storage medium, and the storage medium stores instructions, and when the instructions are executed by the processor, the channel state information processing method described in any one of the possible implementations from the first aspect to the second aspect is executed. accomplish.
  • the communication device described in the ninth aspect may be the first communication device or the second communication device, or it may be a chip (system) or other components or components set in the first communication device or the second communication device , may also be a communication device including a first communication device or a second communication device, which is not limited in this application.
  • the first communication device is configured to execute the channel state information processing method described in any possible implementation of the first aspect
  • the second communication device is configured to execute the method described in any possible implementation of the second aspect. A method for processing channel state information.
  • a processor configured to execute the channel state information processing method described in any one possible implementation manner of the first aspect to the second aspect.
  • a communication system in an eleventh aspect, includes a first communication device or a second communication device.
  • the first communication device is configured to execute the channel state information processing method described in any possible implementation of the first aspect
  • the second communication device is configured to execute the method described in any possible implementation of the second aspect.
  • a method for processing channel state information is provided.
  • a computer-readable storage medium includes a computer program or an instruction, and when the computer program or instruction is executed by a processor, any one of the first aspect to the second aspect Possible Implementation Modes The described channel state information processing method is implemented.
  • a computer program product in a thirteenth aspect, includes an instruction, and when the instruction is executed by a processor, the channel state information described in any one of the possible implementations of the first aspect to the second aspect is made The processing method is implemented.
  • a chip in a fourteenth aspect, includes a processing logic circuit and an interface circuit.
  • the number of processing logic circuits may be one or more, and the number of interface circuits may be more than one.
  • the interface circuit is used to receive code instructions and transmit them to the processing logic circuit.
  • the processing logic circuit is configured to run the above code instructions to execute the channel state information processing method described in any one of the implementation manners of the first aspect to the second aspect.
  • the chip may include a memory, and the memory may be integrated with the processing logic circuit or provided separately.
  • the memory may be used to store computer programs and/or data involved in the channel state information processing method described in any one of the first aspect to the second aspect.
  • the chip described in the fourteenth aspect may be located in the first communication device or the second communication device, and may be located in the first communication device or the second communication device in a communication system.
  • the chip when the chip is located in the first communication device, it is used to execute the channel state information processing method described in any possible implementation manner in the first aspect, and when the chip is located in the second communication device, it is used to execute any possible implementation in the second aspect.
  • the channel state information processing method described in the implementation manner when the chip is located in the second communication device, it is used to execute any possible implementation in the second aspect.
  • FIG. 1 is a schematic diagram of a single-array antenna configuration provided in an embodiment of the present application
  • FIG. 2 is a schematic diagram of a multi-array antenna configuration provided in an embodiment of the present application
  • FIG. 3 is a schematic diagram of a fully connected neural network provided in an embodiment of the present application.
  • FIG. 4 is a schematic diagram of the optimization of the loss function provided by the embodiment of the present application.
  • FIG. 5 is a schematic diagram of error backpropagation provided by the embodiment of the present application.
  • FIG. 6 is a schematic diagram of the hierarchical structure of the graph neural network provided by the embodiment of the present application.
  • FIG. 7 is a schematic structural diagram of a communication system provided by an embodiment of the present application.
  • FIG. 8 is a schematic structural diagram of a communication device provided by an embodiment of the present application.
  • FIG. 9 is an interactive schematic diagram of a channel state information processing method provided in an embodiment of the present application.
  • FIG. 10 is a schematic diagram of a first sub-graph model provided by an embodiment of the present application.
  • FIG. 11 is a schematic diagram of connection of a transceiver antenna provided by an embodiment of the present application.
  • Fig. 12 is a schematic diagram 1 of the first graph model provided by the embodiment of the present application.
  • Fig. 13 is a schematic diagram 2 of the first graph model provided by the embodiment of the present application.
  • FIG. 14 is a first schematic structural diagram of the neural network provided by the embodiment of the present application.
  • Fig. 15 is a structural schematic diagram II of the neural network provided by the embodiment of the present application.
  • FIG. 16 is a schematic diagram of the third structure of the neural network provided by the embodiment of the present application.
  • Fig. 17 is a structural schematic diagram 4 of the neural network provided by the embodiment of the present application.
  • FIG. 18 is a schematic structural diagram of a communication device provided by an embodiment of the present application.
  • MIMO technology is usually used to increase system capacity, that is, multiple antennas are used for communication at the sending end and receiving end at the same time.
  • multiple antennas are used for communication at the sending end and receiving end at the same time.
  • using multiple antennas combined with space division multiplexing can double the system capacity and increase the communication rate.
  • the use of multiple antennas also brings about the problem of interference enhancement. Therefore, it is necessary to perform certain processing on the signal to suppress the impact of interference.
  • This method of suppressing interference through signal processing can be implemented at the receiving end as well as at the transmitting end.
  • the sending end may perform preprocessing on the signal to be sent, and then send it through the MIMO channel. This method is called precoding.
  • precoding is an introduction to precoding.
  • the data to be sent is x
  • the receiving end can use ⁇ - 1UT for decoding to obtain multiple one-to-one channels without interference.
  • the standard provides a series of V matrices, that is, a codebook (codebook), which is coordinated by the transmitting and receiving ends to select an appropriate precoding matrix.
  • codebook codebook
  • the current terminal equipment usually does not report the actual CSI to the network equipment, but the terminal equipment feeds back the PMI to indicate a V that can maximize the H capacity in the codebook given by the standard, so as to realize the network equipment side.
  • the precoding process of this method is equivalent to implicitly indicating the actual CSI.
  • the V indicated by the PMI is not the actual precoding matrix corresponding to the CSI, thus causing performance loss.
  • types of antenna arrays may include single-panel and multi-panel. The following are introduced respectively:
  • a fully connected neural network is also called a multilayer perceptron (MLP).
  • MLP multilayer perceptron
  • Figure 3 is a schematic diagram of a fully connected neural network provided by the embodiment of the present application, an MLP may include an input layer (including x 1 ⁇ x 4 in Figure 3), an output layer (including Figure 3 y 1 ⁇ y 6 in ), and multiple hidden layers.
  • Each hidden layer can contain several nodes (black circles in Figure 3), called neurons. Among them, neurons in two adjacent layers are connected in pairs.
  • the following is a brief introduction to the implementation principle of the fully connected neural network.
  • the output h of the neurons of the next layer is the weighted sum of all the neurons x connected to it through the activation function, expressed in a matrix, you can refer to the following formula:
  • x is a neuron
  • w is a weight matrix
  • b is a bias vector
  • f is an activation function
  • a neural network can be understood as a mapping relationship from an input data set to an output data set.
  • neural networks are initialized randomly, that is, w and b are random numbers. The process of obtaining this mapping relationship from random w and b with existing data is called neural network training.
  • the specific method of training may include: using a loss function (loss function) to evaluate the output of the neural network, backpropagating the error, and iteratively optimizing w and b through gradient descent until the loss function reaches the minimum value.
  • a loss function loss function
  • FIG. 4 is a schematic diagram of the optimization of the loss function provided by the embodiment of the present application.
  • the expression of the gradient descent process can be as follows:
  • FIG. 5 is a schematic diagram of error backpropagation provided by the embodiment of the present application.
  • the formula of error backpropagation can be expressed as:
  • L is the loss function
  • w ij is the weight of node j (that is, the neuron in Figure 3) connected to node i
  • s i is the weighted sum of inputs on node i.
  • GNN is a neural network model specially proposed for graph data (hereinafter collectively referred to as graph model), which has good characteristics independent of graph scale, that is, the same GNN can handle graph models of different structures and/or different scales.
  • FIG. 6 is a schematic diagram of the hierarchical structure of GNN provided by the embodiment of the present application.
  • GNN takes a graph model as input, and outputs a p-dimensional vector representation of a graph, node, edge or subgraph.
  • Graph convolution operations can be performed in each layer of GNN, such as the first graph convolution layer in Figure 6. Similar to traditional convolution operations, graph convolution operations can be considered as aggregation operations on neighboring point information.
  • the formula expression can be as follows:
  • k represents the kth layer graph convolution
  • x v represents the characteristics of node v
  • Agg (k) ( ⁇ , ⁇ ; ⁇ ) represents the aggregation function of the kth layer, which is shared by all nodes
  • ⁇ is to be Parameters for training.
  • the current terminal equipment usually does not report the actual CSI to the network equipment, but indicates the precoding matrix through the PMI in the CSI report, so as to realize the selection of the precoding matrix.
  • the precoding matrix indicated by the PMI is not the actual precoding matrix corresponding to the CSI, thus causing performance loss. Therefore, how to transmit the actual CSI becomes an urgent problem to be solved.
  • an embodiment of the present application provides a technical solution, and the technical solution includes a communication system, a method for processing channel state information applied to the communication system, a communication device, and the like.
  • the wireless communication system can be a fourth generation (4th generation, 4G) communication system (for example, a long term evolution system (long term evolution, LTE) system), the fifth Generation (5th generation, 5G) communication system (for example, new air interface (new radio, NR) system), mobile communication system evolved after 5G (for example, 6G communication system), and narrow band-internet of things system (narrow band-internet of things , NB-IoT) etc.
  • 4G fourth generation
  • LTE long term evolution
  • 5th generation, 5G for example, new air interface (new radio, NR) system
  • mobile communication system evolved after 5G for example, 6G communication system
  • narrow band-internet of things system narrow band-internet of things system
  • the technical solutions of the embodiments of the present application can also be applied to a satellite communication system or a non-terrestrial network (non-terrestrial network, NTN) communication system, where the satellite communication system or the NTN communication system can be integrated with the wireless communication system.
  • the technical solutions of the embodiments of the present application can also be applied to satellite inter-satellite link communication systems, wireless screen projection systems, virtual reality (virtual reality, VR) communication systems, and integrated access and backhaul (IAB) systems , wireless fidelity (wireless fidelity, Wi-Fi) communication system, optical communication system, etc., which are not limited.
  • the present application presents various aspects, embodiments or features in terms of a system that can include a number of devices, components, modules and the like. It is to be understood and appreciated that the various systems may include additional devices, components, modules, etc. and/or may not include all of the devices, components, modules etc. discussed in connection with the figures. Additionally, combinations of these schemes can also be used.
  • the network architecture and business scenarios described in the embodiments of the present application are for more clearly illustrating the technical solutions of the embodiments of the present application, and do not constitute limitations on the technical solutions provided by the embodiments of the present application.
  • the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.
  • An embodiment of the present application provides a communication system, and the communication system may be suitable for communication between a first communication device and a second communication device.
  • the communication system provided by the embodiment of the present application may include one or more first communication devices and one or more second communication devices.
  • the respective numbers of the first communication device and the second communication device in the communication system Not limited.
  • the first communication device is, for example, a terminal device or a network device
  • the second communication device is, for example, a network device or a terminal device.
  • the first communication device is used as a terminal device
  • the second communication device is used as a network device as an example to describe the solution provided by the embodiment of the present application, which will be described in a unified manner here and will not be described in detail below.
  • FIG. 7 is a schematic structural diagram of a communication system provided by the embodiment of the present application.
  • the communication system may include network devices and terminal devices, and the network device and terminal devices may be connected via way to connect. Data and/or control signaling can be exchanged between the network device and the terminal device.
  • the network device in this embodiment of the present application is a device for connecting a terminal device to a wireless network.
  • the network device may be a node in a radio access network, may also be called a base station, and may also be called a radio access network (radio access network, RAN) node (or device), wherein the base station may be a distributed antenna system, It may be a radio frequency head end of the base station that communicates with a certain terminal device.
  • RAN radio access network
  • the network equipment may include an evolved base station (evolved Node B, eNB or eNodeB) in an LTE system or an evolved LTE system (LTE-Advanced, LTE-A), such as a traditional macro base station eNB and a heterogeneous network scenario
  • the micro base station eNB or it can also include the next generation node B (next generation node B, gNB) in the 5G NR system, or it can also include the transmission receiving point (transmitting and receiving point, TRP), the transmitting point (transmitting point, TP ), home base station (for example, home evolved NodeB, or home Node B, HNB), base band unit (base band unit, BBU), base band pool BBU pool, or Wi-Fi access point (access point, AP), mobile switching
  • a network device may serve as a Layer 1 (L1) relay, or as a base station, or as a DU, or as an IAB node, which is not limited in this embodiment of the present application.
  • the network device may also be a node in the core network.
  • the terminal device in the embodiment of the present application may be a device for implementing a wireless communication function, such as a terminal or a chip that may be used in the terminal.
  • the terminal can be user equipment (user equipment, UE), access terminal, terminal unit, terminal station, mobile station, mobile station, remote station, remote terminal, mobile device, wireless communication in the 5G network or the future evolved PLMN.
  • An access terminal may be a cellular telephone, a cordless telephone, a session initiation protocol (SIP) telephone, a wireless local loop (WLL) station, a personal digital assistant (PDA), a Functional handheld devices, computing devices or other processing devices connected to wireless modems, vehicle-mounted devices or wearable devices, VR terminal devices, augmented reality (augmented reality, AR) terminal devices, wireless terminals in industrial control (industrial control), Wireless terminals in self driving, drones, sensors, actuators, satellite terminals, wireless terminals in remote medical, wireless terminals in smart grid, transportation security Safety), wireless terminals in smart city, wireless terminals in smart home, etc.
  • SIP session initiation protocol
  • WLL wireless local loop
  • PDA personal digital assistant
  • Functional handheld devices computing devices or other processing devices connected to wireless modems, vehicle-mounted devices or wearable devices
  • VR terminal devices augmented reality (augmented reality, AR) terminal devices
  • Wireless terminals in self driving drones, sensors, actuators, satellite terminals, wireless terminals
  • the terminal may be a terminal in vehicle-to-everything (V2X) (such as a vehicle-to-everything device), a terminal in device-to-device (Device to Device) communication, or a machine-to-machine (M2M) Communication terminals, etc.
  • V2X vehicle-to-everything
  • M2M machine-to-machine
  • the network devices and terminal devices in the embodiments of this application can be deployed on land, including indoor or outdoor, handheld or vehicle-mounted; they can also be deployed on water; they can also be deployed on airplanes, balloons and artificial satellites in the air .
  • the embodiments of the present application do not limit the application scenarios of the network device and the terminal device.
  • the embodiment of the present application does not specifically limit the specific structure of the execution subject of the method provided in the embodiment of the present application, as long as the program that records the code of the method provided in the embodiment of the present application can be executed according to the method provided in the embodiment of the present application Communication is enough.
  • the execution body of the channel state information processing method provided in the embodiment of the present application may be the first communication device or the second communication device, or a function capable of calling and executing programs in network equipment or terminal equipment. module.
  • the relevant functions of the first communication device or the second communication device in the embodiment of the present application can be realized by one device, or by multiple devices, or by one or more functional modules in one device , which is not specifically limited in this embodiment of the present application. It can be understood that the above functions can be network elements in hardware devices, software functions running on dedicated hardware, or a combination of hardware and software, or instantiated on a platform (for example, a cloud platform) virtualization capabilities.
  • a platform for example, a cloud platform
  • FIG. 8 is a schematic structural diagram of a communication device 800 provided by an embodiment of the present application.
  • the communication device 800 may include one or more processors 801, a communication line 802, and at least one communication interface (the communication interface 804 and a processor 801 are included as an example in FIG.
  • a memory 803 may also be included.
  • the processor 801 can be a central processing unit (central processing unit, CPU), a microprocessor, a specific application integrated circuit (application-specific integrated circuit, ASIC), or one or more integrated circuits used to control the execution of the program program of this application. circuit.
  • CPU central processing unit
  • ASIC application-specific integrated circuit
  • Communication line 802 may include a path for connecting between different components.
  • the communication line 802 may be a bus, such as an address bus, a data bus, a control bus, and the like.
  • the communication interface 804 may be a transceiver module, and may be used to communicate with other devices or a communication network.
  • the transceiving module may be a device such as a transceiver or a transceiver.
  • the communication interface 804 may also be a transceiver circuit located in the processor 801 to realize signal input and signal output of the processor.
  • the storage 803 may be a device having a storage function.
  • it may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (random access memory, RAM) or other types of memory that can store information and instructions
  • a dynamic storage device can also be an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or can be used to carry or store desired program code in the form of instructions or data structures and can be stored by a computer Any other medium, but not limited to.
  • the memory may exist independently and be connected to the processor through the communication line 802 . Memory can also be integrated with the processor.
  • the memory 803 is used to store computer-executed instructions for implementing the solution of the present application, and the execution is controlled by the processor 801 .
  • the processor 801 is configured to execute computer-executed instructions stored in the memory 803, so as to implement the channel state information processing method provided in the embodiment of the present application.
  • the processor 801 may also perform processing-related functions in the channel state information processing method provided in the following embodiments of the present application, and the communication interface 804 is responsible for communicating with other devices or communication networks.
  • the example does not specifically limit this.
  • the computer-executable instructions in the embodiment of the present application may also be referred to as application code, which is not specifically limited in the embodiment of the present application.
  • the processor 801 may include one or more CPUs, for example, CPU0 and CPU1 in FIG. 8 .
  • the communications apparatus 800 may include multiple processors, for example, the processor 801 and the processor 808 in FIG. 8 .
  • Each of these processors may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor.
  • a processor herein may refer to one or more devices, circuits, and/or processing cores for processing data (eg, computer program instructions).
  • the communication apparatus 800 may further include an output device 805 and an input device 806 .
  • Output device 805 is in communication with processor 801 and can display information in a variety of ways.
  • FIG. 9 is an interactive schematic diagram of a method for processing channel state information provided in an embodiment of the present application.
  • the method for processing channel state information may be applied to the above-mentioned communication system, and may be executed by a network device or a terminal device in the above-mentioned communication system.
  • the channel state information processing method can ensure the validity of the CSI and improve the communication performance.
  • the method may include S901-S905, which will be described in sequence below.
  • the terminal device determines a first graph model corresponding to the first channel state information.
  • the first channel state information and the first graph model are introduced respectively below.
  • the first channel state information may include channel state information between the network device and the terminal device. Specifically, the first channel state information represents CSI between the network device and the terminal device.
  • the first channel state information may include delay-angle domain channel state information between the transmitting antenna of the network device and the receiving antenna of the terminal device.
  • the delay-angle domain channel state information may include channel state information between N r receiving angles and N t transmitting angles, where both N r and N t are positive integers, and at least one of R and T is greater than 1.
  • N r can be denoted as R and N t can be denoted as T.
  • the channel state information in the delay-angle domain may be represented by a matrix, that is, the channel state information in the delay-angle domain may be a channel matrix in the delay-angle domain.
  • N r may represent the number of receiving antennas of the terminal device
  • N t may represent the number of transmitting antennas of the network device, which are uniformly described here and will not be described in detail below.
  • the terminal device may determine first channel state information.
  • the implementation of "the terminal device determines the first channel state information" and the first channel state information will be described in detail below.
  • the terminal device determining the first channel state information may include: the terminal device determining the first channel state information according to the space-frequency domain channel state information between the transmitting antenna of the network device and the receiving antenna of the terminal device.
  • the channel state information in the space-frequency domain may be represented by a matrix, that is, the channel state information in the space-frequency domain may be a channel matrix in the space-frequency domain.
  • the terminal device may receive a reference signal sent by the network device, and perform CSI estimation based on the reference signal, to obtain a space-frequency domain channel matrix (denoted as ).
  • the reference signal sent by the network device may be a channel state information-reference signal (channel state information-reference signal, CSI-RS).
  • the end device can Perform discrete Fourier transform (discrete fourier transform, DFT) to obtain the channel matrix in the delay-angle domain (denoted as ), That is, the delay-angle domain channel state information, that is, the first channel state information.
  • DFT discrete Fourier transform
  • the transmitting antenna of the network device includes N t transmitting antennas
  • the receiving antenna of the terminal device includes N r receiving antennas
  • the space-frequency domain channel matrix includes N t transmitting antennas and N r receiving antennas
  • Channel state information the channel state information between the N t transmission angles and the N r reception angles in the angle domain can be obtained after the discrete Fourier transform is performed on the space domain of the space-frequency domain channel matrix.
  • the process may include the following several implementation modes (mode 1 to mode 3).
  • Mode 1 for each receiving angle, determine a corresponding delay-angle domain channel matrix (denoted as ), i is a positive integer, i ⁇ N r . including N r
  • the terminal device can estimate the channel between the i-th receiving antenna and the N t transmitting antennas among the N r receiving antennas according to the reference signal, and obtain the space between the i-th receiving antenna and the N t transmitting antennas -
  • the frequency domain channel matrix (denoted as ), is the number of subcarriers, Represents a complex field.
  • the end device can Perform two-dimensional DFT to obtain the channel matrix in the delay-angle domain corresponding to the i-th receiving angle (that is, ). It can be seen that the method 1 obtained
  • N r pieces each Corresponding to a receiving angle, all the is the channel matrix in the delay-angle domain (that is, ).
  • It can represent the number of subcarriers, which will be described in a unified manner here, and will not be described in detail below.
  • method 1 can represent the channel state information between one receiving angle and N t transmitting angles, all can represent the channel state information between N r receiving angles and N t transmitting angles, that is to say, Channel state information between N r receiving angles and N t transmitting angles may be included.
  • Mode 2 for each transmission angle, determine a corresponding delay-angle domain channel matrix (denoted as ), j is a positive integer, j ⁇ N t . including N t
  • the terminal device can estimate the channel between the j-th transmit antenna and the N r receive antennas among the N t transmit antennas according to the reference signal, and obtain the space between the j-th transmit antenna and the N r receive antennas - The frequency domain channel matrix (denoted as ), Then, the end device can Perform two-dimensional DFT to obtain the channel matrix in the delay-angle domain corresponding to the jth transmission angle (that is, ). It can be seen that the method 2 obtained There are N t pieces, each Corresponding to a transmit antenna, all the is the channel matrix in the delay-angle domain (that is, ).
  • method 2 can represent the channel state information between one transmit angle and N r receive angles, all can represent the channel state information between N r receiving angles and N t transmitting angles, that is to say, Channel state information between N r receiving angles and N t transmitting angles may be included.
  • Mode 3 for all transmit angles and all receive angles, determine the corresponding delay-angle domain channel matrix (denoted as ). that is
  • the terminal device can estimate the channels between the N r receiving antennas and the N t transmitting antennas according to the reference signal, and obtain the space-frequency domain channel matrix between the N r receiving antennas and the N t transmitting antennas (denote for ), Then, the end device can Perform three-dimensional DFT to obtain the channel matrix in the delay-angle domain (that is, ).
  • mode 3 can represent the channel state information between N r receiving angles and N t transmitting angles, that is to say, Channel state information between N r receiving angles and N t transmitting angles may be included.
  • the first graph model may include a plurality of nodes, and the characteristics of a node may include channel state information between the j-th transmission angle among the N t transmission angles and the i-th reception angle among the N r reception angles, i and j are all positive integers, i ⁇ N r , j ⁇ N t .
  • the features of multiple nodes may include channel state information between N t transmit angles and N r receive angles.
  • the first graph model may also include at least one edge. Wherein, each edge may be connected to two nodes, and each edge may indicate that the two nodes connected to the edge correspond to two adjacent receiving angles or correspond to two adjacent transmitting angles.
  • the terminal device determining the first graph model corresponding to the first channel state information may include: the terminal device determining the first graph model corresponding to the first channel state information according to the transmitting and receiving antenna configuration information.
  • the transmitting and receiving antenna configuration information may include: receiving antenna configuration information of the terminal device and transmitting antenna configuration information of the network device.
  • the receiving antenna configuration information of the terminal device may include one or more of the following: the number of receiving antennas of the terminal device, the type of receiving antenna array, and the arrangement of receiving antenna units.
  • the transmit antenna configuration information of the network device may include one or more of the following: the number of transmit antennas of the network device, the type of transmit antenna front, and the arrangement of transmit antenna units.
  • the above-mentioned transmitting and receiving antenna configuration information may be predefined by the protocol; it may also be indicated by the network device to the terminal device through signaling.
  • the network device may also send the transmitting and receiving antenna configuration information to the terminal device, correspondingly , the terminal device receives the transmitting and receiving antenna configuration information from the network device.
  • the embodiment of the present application does not limit the specific implementation manner for the terminal device to acquire configuration information of the transmitting and receiving antennas.
  • the terminal device determines the first graph model corresponding to the first channel state information according to the transmitting and receiving antenna configuration information, which may include the following several implementation manners (mode 4 to mode 6).
  • Mode 4 for each receiving angle, a corresponding first sub-graph model is determined, and all first sub-graph models are the first graph model.
  • the terminal device can use the above method 1 to obtain the sub-graph model corresponding to the i-th receiving angle Then, the terminal device determines the The corresponding first subgraph model.
  • the angle domain may include N 1 ⁇ N 2 emission angles.
  • the j-th angle among the N 1 ⁇ N 2 emission angles can include a vector in the time domain dimension (denoted as Xj, j is a positive integer, j ⁇ N t ), and Xj includes elements, that is, the length of Xj is Xj can represent the channel state information between the j-th transmission angle and the i-th reception angle, and all Xj can represent the channel state information between the N t transmission angles and the i-th reception angle.
  • FIG. 10 is a schematic diagram of a first sub-graph model provided by an embodiment of the present application.
  • the terminal device can determine the N 1 ⁇ N 2 angles as nodes in the first sub-graph model, that is, the first sub-graph model includes N 1 ⁇ N 2 nodes, and the feature of each node is that the node corresponds to A vector (ie, Xj) of emission angles comprised in the time-domain dimension.
  • Terminal equipment can also be based on Identify edges in the first subgraph model.
  • an end device can be based on For the emission angles corresponding to the N 1 ⁇ N 2 angles included in the angle domain, it is determined that there is an edge between two nodes corresponding to the adjacent emission angles in the first subgraph model. In other words, each edge can represent two adjacent emission angles corresponding to two nodes connected to the edge.
  • the terminal device can be based on The angle domain includes the adjacency relationship between N 1 ⁇ N 2 angles, and it is determined that there is an edge between adjacent nodes in the first subgraph model, and there is no edge between non-adjacent nodes.
  • each edge may indicate that two nodes connected to the edge are adjacent in the first subgraph model. Since the correlation between non-adjacent angles is generally small, the implementation of an edge between adjacent nodes in the first subgraph model can not only reduce the data processing amount of the device, but also ensure the accuracy of the channel state information. effectiveness.
  • the terminal device can be based on For the emission angles corresponding to the N 1 ⁇ N 2 angles included in the angle domain, it is determined that there is an edge between two nodes corresponding to the same reception angle in the first subgraph model.
  • Xj is Dimension complex vector
  • Xj can also be written as Represents all elements of the j-th emission angle in the time domain dimension, It can also be simply expressed as H[:,j].
  • the imaginary part and real part of H[:, j] can be concatenated into dimensions
  • the vector of , that is, the imaginary part and real part of H[:, j] are processed separately.
  • the first subgraph model can be represented by an adjacency matrix and a node characteristic matrix.
  • the terminal device can pass the adjacency matrix of N 1 ⁇ N 2 rows and N 1 ⁇ N 2 columns Represents the connection relationship between nodes in the first subgraph model.
  • the terminal device can determine the first subgraph model shown in 1 in Figure 12; for The terminal device may determine the first subgraph model as shown in 2 in FIG. 12 .
  • the corresponding first subgraph model may include 3 nodes (including nodes 1 to 3) and 2 edges (including edges 1 and 2).
  • Node 1 to node 3 all correspond to receiving angle 1
  • node 1 also corresponds to transmitting angle 1
  • node 2 also corresponds to transmitting angle 2
  • node 3 also corresponds to transmitting angle 3
  • edge 1 connects node 1 and node 2
  • edge 2 connects node 2, node 3.
  • the corresponding first subgraph model is with The corresponding first subgraph model is similar, and 2 in FIG. 12 will not be repeated here.
  • the corresponding first subgraph model can be represented by the following adjacency matrix A and node feature matrix B:
  • Mode 5 for each emission angle, determine a corresponding second sub-graph model, and all the second sub-graph models are the first graph model.
  • the terminal device can use the above method 2 to obtain the Then, the terminal device determines the The corresponding second subgraph model.
  • the angle domain may include N 1 ⁇ N 2 angles.
  • the i-th angle among the N 1 ⁇ N 2 angles can include a vector in the time domain dimension (denoted as Ui, i is a positive integer, i ⁇ N r ), Ui includes elements, that is, the length of Ui is Ui can represent the channel state information between the i-th receiving angle and the j-th transmitting angle, and all Ui can represent the channel state information between the N r receiving angles and the j-th transmitting angle.
  • the terminal device can determine the N 1 ⁇ N 2 angles as nodes in the second sub-graph model, that is, the second sub-graph model includes N 1 ⁇ N 2 nodes,
  • the feature of each node is the vector (ie, Ui) included in the time domain dimension of the receiving angle corresponding to the node.
  • Terminal equipment can also be based on Identify edges in the second subgraph model.
  • an end device can be based on For receiving angles corresponding to the N 1 ⁇ N 2 angles included in the angle field, it is determined that there is an edge between two nodes corresponding to adjacent receiving angles in the second subgraph model. In other words, each edge can represent two adjacent acceptance angles corresponding to two nodes connected to the edge.
  • the terminal device can be based on The angle domain includes the adjacency relationship between N 1 ⁇ N 2 angles, and it is determined that there is an edge between adjacent nodes in the second subgraph model, and there is no edge between non-adjacent nodes.
  • each edge may indicate that two nodes connected to the edge are adjacent in the second subgraph model. Since the correlation between non-adjacent angles is generally small, the implementation of an edge between adjacent nodes in the second subgraph model can not only reduce the data processing amount of the device, but also ensure the accuracy of the channel state information. effectiveness.
  • the terminal device can be based on For receiving angles corresponding to the N 1 ⁇ N 2 angles included in the angle domain, it is determined that there is an edge between two nodes corresponding to the same transmitting angle in the second subgraph model.
  • Ui is Dimension complex vector
  • Ui can also be written as Represents all elements of the i-th receiving angle in the time domain dimension, It can also be simply expressed as H[:,i].
  • the imaginary part and real part of H[:,i] can be concatenated into dimensions
  • the vector of , that is, the imaginary part and real part of H[:,i] are processed separately.
  • the second subgraph model can be represented by an adjacency matrix and a node characteristic matrix.
  • the terminal device can pass the adjacency matrix of N 1 ⁇ N 2 rows and N 1 ⁇ N 2 columns Represents the connection relationship between nodes in the second subgraph model.
  • Mode 5 For a detailed description of the implementation process of Mode 5, reference may be made to the relevant descriptions in the above Mode 4 in conjunction with FIG. 11 and FIG. 12 , and details are not repeated here.
  • the terminal device can use the above method 3 to obtain Then according to the transceiver antenna configuration information, determine Corresponding first graph model.
  • the terminal device determines according to the configuration information of the transmitting and receiving antennas: the network device includes N t transmitting antennas, and the terminal device includes N r receiving antennas, then the terminal device can determine That is to say, The angle domain may include N t ⁇ N r angle pairs.
  • the N t ⁇ N r angle pairs are represented by a matrix, that is, the N t ⁇ N r angle pairs include N r rows and N t columns of angle pairs
  • the i-th row of the N t ⁇ N r angle pairs , the j-th column angle pair can include a vector (denoted as V(i, j) in the time domain dimension, i and j are both positive integers, i ⁇ N r , j ⁇ N t )
  • V(i, j) includes elements, that is, the length of V(i, j) is G(i, j) corresponds to the j-th transmit angle and corresponds to the i-th receive angle.
  • V(i, j) can represent the channel state information between the j-th transmitting angle and the i-th receiving angle
  • all V(i, j) can represent the distance between N t transmitting angles and N r receiving angles.
  • the terminal device can determine these N t ⁇ N r angle pairs as nodes in the first graph model, that is, the first graph model includes N t ⁇ N r nodes, each The feature of a node is the vector (ie, V(i, j)) included in the time domain dimension by the angle pair corresponding to the node.
  • Terminal equipment can also be based on Identify edges in the first subgraph model. For example, an end device can be based on The angle pair included in the angle domain corresponds to the emission angle and the reception angle, and it is determined that an edge exists between two nodes corresponding to adjacent reception angles or corresponding to adjacent emission angles in the first graph model. In other words, each edge may indicate that the two nodes connected to the edge correspond to two adjacent receiving angles or correspond to two adjacent transmitting angles.
  • the terminal device can be based on The adjacency relationship between the angles included in the angle domain determines that there is an edge between adjacent nodes in the first graph model, and there is no edge between non-adjacent nodes.
  • each edge may indicate that two nodes connected to the edge are adjacent in the first graph model. Since the correlation between non-adjacent angles is generally small, the implementation of an edge between adjacent nodes in the first graph model can not only reduce the data processing amount of the device, but also ensure the effective channel state information sex.
  • the terminal device can be based on In the transmitting angle and receiving angle corresponding to the angle included in the angle field, it is determined that an edge exists between two nodes corresponding to the same transmitting angle or corresponding to the same receiving angle in the second subgraph model.
  • V(i, j) is dimensional complex vector
  • V(i, j) can also be written as represents all elements of the angle pair in the time domain dimension, It can also be simply expressed as H[:,i,j].
  • the imaginary part and real part of H[:,i,j] can be concatenated into dimensions
  • the vector of , that is, the imaginary part and real part of H[:,i,j] are processed separately.
  • the first graph model can be represented by an adjacency matrix and a node feature matrix.
  • the terminal device can pass the adjacency matrix of N t ⁇ N r rows and N t ⁇ N r columns Represents the connection relationship between nodes in the first graph model.
  • the terminal device can also pass the node feature matrix Represents the characteristics of each node.
  • the i-th row and j-th column element of B is V(i, j).
  • the terminal device can obtain the Then determine the For the specific implementation manner of the corresponding first graph model, refer to the relevant description in the foregoing manner 6.
  • the terminal device may determine the first graphic model as shown in FIG. 13 .
  • the corresponding first graph model may include 6 nodes (including nodes 1 to 6) and 7 edges (including edges 1 to 7).
  • the transmission angles and reception angles corresponding to each node, and the connection relationship between nodes can be referred to as shown in FIG. 13 , which will not be repeated here.
  • the channel state information included in the characteristics of each node may include in the time domain elements.
  • the channel state information included in the characteristics of each node may include C elements in the time domain, and C is less than or equal to the number of subcarriers, that is, In other words, in this embodiment of the present application, all elements included in the channel state information included in the characteristics of each node in the time domain may be retained, or a part may be discarded and the remaining part may be used as the characteristics of each node.
  • the terminal device may determine the first subgraph model corresponding to Hi according to the transmitting and receiving antenna configuration information. For the specific method, refer to the above method 4, which will not be repeated here.
  • N c can represent the time domain dimension to the length of the time domain dimension after the truncation of the time delay-angle domain channel matrix, in This unified description will not be described in detail below.
  • the imaginary part and the real part of H[:,i,j] are processed separately, then the dimension of Xj in the above method 4 is 2N c .
  • the data processing amount of the communication device can be reduced and the processing rate of the communication device can be increased without affecting the accurate recovery of the first channel state information by the second communication device.
  • the terminal device uses the first neural network to process the first graphical model to obtain first information.
  • the first information is used for the network device to restore the first channel state information.
  • the first information may include first channel state information processed by the first neural network, so that the network device can restore the first channel state information according to the first information.
  • the first neural network can compress the feature information contained in the first graphical model.
  • important feature information may be retained, and unimportant feature information may be discarded (or omitted). Since unimportant feature information in the feature information contained in the first graph model may be discarded, in order to improve the accuracy of the network device in recovering the first channel state information based on the first information, in some possible embodiments, the first information Auxiliary information may be included, and the auxiliary information may be used to determine a second graph model, and the second graph model may be used to restore the first channel state information.
  • the first information may include the second channel state information and auxiliary information
  • the auxiliary information is used to determine the second graph model means that the second graph model can be restored by combining the auxiliary information with the second channel state information.
  • the network device can use the auxiliary information to determine the second graph model, and restore the first channel state information according to the second graph model.
  • the auxiliary information can help the second communication device recover the first channel state information, and improve the accuracy of the first channel state information recovered by the second communication device.
  • the auxiliary information may include index information of some nodes of the first graph model.
  • the auxiliary information may be used to indicate part of feature information in the first graphical model (such as discarded or reduced feature information).
  • the terminal device can use the auxiliary information to indicate the indexes of some nodes in the first graph model that are discarded due to compression, so as to help the network device recover the first channel state information and improve the accuracy of the first channel state information recovered by the network device. sex.
  • the auxiliary information may include index information of discarded nodes in the first graph model, or the auxiliary information may include index information of retained nodes in the first graph model.
  • the network device can be further helped to restore the first channel state information, and the accuracy of the first channel state information restored by the network device can be improved.
  • the auxiliary information may also include mean values of features of some nodes.
  • the auxiliary information may also include the mean value of the features of the nodes discarded during the processing of the first graph model. In this way, the network device can be further helped to restore the first channel state information, and the accuracy of the first channel state information restored by the network device can be improved.
  • the index information of a node may also be referred to as a serial number of a node, an identifier of a node, or location information of a node, etc., which are not limited here.
  • the following describes an implementation manner in which in S902, the first neural network is used to process the first graphical model to obtain the first information.
  • the first neural network can be used to determine the first information according to the first graph model, the first information can include the second channel state information and the above-mentioned auxiliary information, the second channel
  • the state information may include processed first channel state information.
  • FIG. 15 is a second schematic structural diagram of a neural network provided by an embodiment of the present application.
  • the first neural network may include a first image pooling layer and a compression layer.
  • the first graph pooling layer can be used to determine the third graph model and auxiliary information according to the first graph model
  • the compression layer can be used to determine the second channel state information according to the third graph model.
  • the input of the first graph pooling layer can be graph models of different structures and/or different scales, that is to say, for terminal devices and network devices with different antenna
  • a first graphical model corresponding to the first channel state information may be determined, and the first graphical model is input into the same first neural network.
  • the first graph pooling layer can retain a fixed number of important nodes in the first graph model, and discard unimportant nodes in the first graph model, so that the determined structure of the third graph model (or called size, size) is fixed, that is, the third graph model has nothing to do with the structure and/or scale of the input first graph model. In this way, for the first graph models of different structures and/or different scales, the same first graph pooling layer and compression layer can be used for processing.
  • the same first graph pooling layer and compression layer can be used for the first graph model.
  • Channel state information is processed.
  • the design and training of the first neural network does not depend on the configuration of the antennas at both ends of the sending and receiving CSI. In this way, when the configurations of the antennas at both ends of the sending and receiving CSI change, redesign and training of the neural network can be avoided, and it is applicable to dynamic MIMO systems and MIMO systems with different antenna configurations.
  • the first graph pooling layer may be a graph pooling layer using a self-attention mechanism. Since the graph pooling layer of the self-attention mechanism needs fewer parameters to be trained, and can fully extract the feature information of the nodes of the first graph model and the topological structure information of the graph, the "first graph pooling layer adopts the self-attention mechanism
  • the "Graph Pooling Layer” can improve the training efficiency and processing power of the neural network.
  • FIG. 16 is a schematic structural diagram of a neural network provided in the embodiment of the present application.
  • the third neural network may include a first graph convolution layer, a first graph pooling layer, and Compression layer.
  • the first graph convolution layer is used to determine the first graph model after convolution according to the first graph model
  • the first graph pooling layer is used to determine the third graph model and auxiliary information according to the first graph model after convolution
  • the compression layer is used to determine the second channel state information according to the third graph model.
  • the first graph convolutional layer may include multiple graph convolutional layers.
  • the first graph convolution layer can be any type of graph convolution layer such as graph convolution network (graph convolution network, GCN) or graph sample and aggregate (GraphSage) graph convolution layer.
  • the first graph convolution layer can extract and retain the features of important nodes in the first graph model, and omit the features of unimportant nodes. Therefore, in the first neural network shown in Figure 16, the first graph convolution Layer enables preliminary channel information compression.
  • the design and training of the first neural network shown in FIG. 16 does not depend on the antenna configurations at both ends of the sending and receiving CSI. In this way, when the configurations of the antennas at both ends of the sending and receiving CSI change, redesign and training of the neural network can be avoided, and it is applicable to dynamic MIMO systems and MIMO systems with different antenna configurations.
  • the compression layer in the first neural network shown in FIG. 15 and the first neural network shown in FIG. 16 can compress the third graph model into data of a certain length, that is, the second channel state information .
  • the second channel state information may include compressed first channel state information.
  • the compression layer in the above-mentioned first neural network may be a fully connected layer.
  • the terminal device sends the first information to the network device.
  • the network device receives the first information from the terminal device.
  • the first information can be carried in an uplink channel, such as a physical uplink control channel (physical uplink control channel, PUCCH), a physical uplink shared channel (physical uplink shared channel, PUSCH), other physical uplink channels, etc., such as future may define
  • PUCCH physical uplink control channel
  • PUSCH physical uplink shared channel
  • the physical channel that can be used to carry the first information is not limited in this embodiment of the present application.
  • the network device uses the second neural network to process the first information to obtain a second graph model.
  • the first information may include auxiliary information and second channel state information.
  • auxiliary information and the second channel state information For relevant descriptions about the auxiliary information and the second channel state information, reference may be made to the relevant descriptions in S902 above, and details are not repeated here.
  • the second neural network may be used to determine the second graph model according to the second channel state information and auxiliary information.
  • the second neural network may include a decompression layer and a second graph convolution layer.
  • the decompression layer may be used to determine the fourth graph model according to the second channel state information.
  • the decompression layer may decompress the second channel state information into data of a certain length, which is used to determine the fourth graph model.
  • the decompression layer can be a fully connected layer.
  • the second graph convolution layer may be used to determine the second graph model according to the auxiliary information and the fourth graph model.
  • the second graph convolution layer is used to determine the second graph model according to the auxiliary information and the fourth graph model, which may include: the network device restores the missing nodes in the fourth graph model according to the auxiliary information to obtain the fifth graph model; then , input the fifth graph model into the second graph convolution layer, and the second graph convolution layer is used to determine the second graph model according to the fifth graph model.
  • the second graph convolution layer can be used to perform convolution on the fifth graph model to obtain the second graph model.
  • the fifth graph model can be regarded as the fourth graph model that complements and discards some nodes.
  • the auxiliary information may also include the mean value of the features of the nodes discarded during the processing of the first graph model, in other words, the auxiliary information may also include the mean value of the nodes discarded in the first graph model, the description of the first graph model refers to the above , which will not be repeated here.
  • the second graph convolutional layer may include multiple graph convolutional layers.
  • direct channels may be included between the multiple graph convolution layers included in the second graph convolution layer. In this way, the training efficiency and stability of the neural network can be improved.
  • the design and training of the second neural network does not depend on the configuration of the antennas at both ends of the sending and receiving CSI. In this way, when the configurations of the antennas at both ends of the sending and receiving CSI change, redesign and training of the neural network can be avoided, and it is applicable to dynamic MIMO systems and MIMO systems with different antenna configurations.
  • the network device determines third channel state information according to the second graph model.
  • the third channel state information is the restored first channel state information.
  • the third channel state information includes channel state information between the network device and the terminal device.
  • the third channel state information represents the CSI between the network device and the terminal device.
  • the network device determining the third channel state information according to the second graph model may include: the network device determining the third channel state according to the second graph model and the transmitting and receiving antenna configuration information.
  • the transmitting and receiving antenna configuration information for relevant descriptions of the transmitting and receiving antenna configuration information, reference may be made to relevant descriptions in S901-S902 above, and details are not repeated here.
  • S905 can be regarded as the inverse process of the above S901, so that the network device determines the third channel state according to the second graph model and the configuration information of the transmitting and receiving antennas, correspondingly refer to the above S901, which will not be repeated here.
  • FIG. 17 is a fourth structural diagram of the neural network provided by the embodiment of the present application.
  • the first neural network can include a 2-layer GraphSage graph convolution layer (including the first layer GraphSage graph convolution layer, the second layer GraphSage graph convolution layer), the graph pooling layer and the fully connected layer of the self-attention mechanism (denoted as first fully connected layer).
  • the second neural network may include a fully connected layer (referred to as the second fully connected layer) and a 4-layer GraphSage convolution layer (including the third layer to the sixth layer GraphSage convolution layer).
  • FIG. 17 For the cascading relationship between each network layer in the first neural network and the second neural network, reference may be made to FIG. 17 , which will not be repeated here.
  • the first layer of GraphSage graph convolution layer the second layer of GraphSage graph convolution layer.
  • the terminal device can input the first graph model into the first layer of GraphSage graph convolution layer, and after convolution of the first layer of GraphSage graph convolution layer and the second layer of GraphSage graph convolution layer, the first graph model after convolution is obtained.
  • the calculation expression of the kth layer GraphSage graph convolution layer can be as follows:
  • the convolution of the first layer GraphSage graph convolution layer and the second layer GraphSage graph convolution layer you can learn the relationship between nodes, the correlation between each node attribute and other node attributes, and get the first graph A point embedding feature vector for each node in the model.
  • the dimension of the point embedding feature vector can be smaller than the dimension of the original feature (such as 2N c in the above S901), therefore, the first layer of GraphSage graph convolution layer and the second layer of GraphSage graph convolution layer can achieve preliminary Channel information compression.
  • the parameters ⁇ W 1 , W 2 ⁇ involved in the above operation are applicable to all nodes, so the operation of this part is independent of the structure and scale of the input first graph model, that is, independent of the antenna configuration.
  • the terminal device can input the first convolutional graph model to the graph pooling layer of the self-attention mechanism, and obtain the third graph model after being pooled by the graph pooling layer of the self-attention mechanism.
  • the processing process of the graph pooling layer of the self-attention mechanism can be considered as a downsampling process, that is, only some points in the graph model and the connection relationship between them are retained, and the feature information of the nodes of the first graph model can be fully extracted. and graph topology information.
  • the calculation expression of the graph pooling layer of the self-attention mechanism can be as follows:
  • A is the adjacency matrix
  • H (2) is the matrix, including the point embedding feature vectors of all nodes output by the second layer of GraphSage graph convolution layer
  • GNN(H (2) ,A) represents the graph based on the self-attention mechanism
  • the convolution operation is to calculate the score of each node, and the score can indicate the importance of the node.
  • the terminal device can process the third graph model according to the following formula to obtain the first vector:
  • s can also be called the reserved dimension
  • the length of the embedded feature vector can also be called the embedded dimension, which will be described uniformly here and will not be described in detail below.
  • the terminal device can use indexes 1-15, 20, 22, 25, 30, and 31 as auxiliary information to indicate the nodes reserved in the first graph model of the network device, or use indexes 16-19, 21, and 23 , 24, 26-29, 32, to indicate the discarded nodes in the first graph model of the network device.
  • the above-mentioned auxiliary information may also include the mean value of the features (ie point embedding feature vectors) of all discarded nodes.
  • the mean of the features of the dropped nodes is the mean of the features of node 1 and the features of node 2.
  • s in the calculation expression of the graph pooling layer of the above self-attention mechanism may be predefined by a protocol, or may be indicated to the terminal device by the network device through signaling, which is not limited.
  • the terminal device may input the first vector obtained based on the third graph model into the first fully connected layer, and obtain the second channel state information after being compressed by the first fully connected layer.
  • the second fully connected layer may decompress the second channel state information to obtain a second vector.
  • the network device may use the second fully connected layer to decompress the second channel state information, and obtain the second vector based on the decompression result.
  • the network device may convert the second vector into a fourth graph model.
  • the second fully connected layer can output a real number vector with a length of 2sNc , and divide it into s subvectors with a length of 2Nc , and then determine the s subvectors as the fourth graph model.
  • each sub-vector is a feature of a node in the fourth graph model.
  • the network device may fill in the discarded nodes in the fourth graph model according to the auxiliary information to obtain the fifth graph model.
  • the network device may recover the discarded nodes in the fourth graph model according to the auxiliary information and the configuration information of the transmitting and receiving antennas. Specifically, the network device can determine the positions of the discarded nodes (that is, the adjacency relationship with other nodes) in the fourth graph model according to the auxiliary information and the configuration information of the transmitting and receiving antennas, and then fill the positions of these discarded nodes with zero vectors, thereby Restores the discarded nodes in the fourth graph model. The dimensionality of the zero vector is consistent with the dimensionality of the features of the nodes retained in the fourth graph model. This padding method may be called zero padding.
  • the network device can fill the first vector at the positions of these discarded nodes, thereby restoring the discarded nodes in the fourth graph model.
  • the first vector is determined by the mean value of the features of all discarded nodes, or in other words, the size of all elements in the first vector is the mean value of the features of all discarded nodes.
  • the dimensionality of the first vector is consistent with the dimensionality of the features of the nodes retained in the fourth graph model. This filling method can be called mean filling.
  • the network device may fill the second vector at the positions of these discarded nodes, so as to restore the discarded nodes in the fourth graph model.
  • the second vector is a vector estimated according to the characteristics of adjacent nodes, for example, the second vector is estimated by an interpolation method.
  • the dimensionality of the second vector is consistent with the dimensionality of the features of the nodes retained in the fourth graph model. This filling method may be called estimated filling.
  • the network device can input the fifth graph model into the third-layer GraphSage graph convolution layer, and obtain the second graph model after being convolved by the third-layer GraphSage graph convolution layer to the sixth-layer GraphSage graph convolution layer.
  • the calculation expressions of the third layer GraphSage graph convolution layer to the sixth layer GraphSage graph convolution layer can refer to the calculation expressions of the first layer GraphSage graph convolution layer and the second layer GraphSage graph convolution layer, in This will not be repeated here.
  • first direct connection channel between the input of the third layer GraphSage convolutional layer in the second neural network and the output of the fourth layer GraphSage convolutional layer, that is to say, the input layer 3
  • the data of the GraphSage convolutional layer and the data output by the fourth layer of the GraphSage convolutional layer are added and then input to the fifth layer of the GraphSage convolutional layer.
  • second direct connection channel between the input of the 5th layer GraphSage convolution layer and the output of the 6th layer GraphSage convolution layer.
  • the implementation principle is similar to the first direct channel, and will not be repeated here.
  • the 4-layer GraphSage graph convolution layer shown in Figure 17 can also be replaced with fewer graph convolution layers (such as 2-layer GraphSage graph convolution layers), or replaced with more graph convolution layers ( For example, a 6-layer GraphSage graph convolution layer), there is no limit to this. It has been verified that compared with fewer graph convolution layers, the performance of the 4-layer GraphSage graph convolution layer can meet the requirements; compared with more graph convolution layers, the performance of the 4-layer GraphSage graph convolution layer meets the requirements. Basics are easier to train.
  • the channel state information processing method shown in S901 to S905 above may be applied to a single front, or to multiple fronts.
  • the terminal device can use the same first neural network to process the channel state information (see S901-S903 for the specific process), The first information corresponding to each front is obtained, and the first information corresponding to each front is fed back to the network device.
  • the terminal device can also use the first neural network to process the channel state information of all the fronts in the multi-front at the same time, for example, determine the first channel state information corresponding to all the fronts in the multi-front, and determine The first graph model corresponding to the first channel state information, and then use S902 to S903 to obtain the first information corresponding to all the fronts, and feed back the first information corresponding to all the fronts to the network device.
  • the first neural network does not need to be adjusted and retrained each time it is used.
  • the application method of the network device is similar to that of the terminal device, and will not be repeated here.
  • the terminal device when the terminal device uses the first neural network to process the first graph model, it can retain important feature information in the first graph model, omit unimportant feature information in the first graph model, and use
  • the first neural network compresses the first channel state information to reduce the CSI feedback overhead while ensuring the effectiveness of the CSI feedback, so that the channel state information restored by the network device is closer to the original channel, thereby improving communication performance.
  • a training method for the first neural network and the second neural network is also provided, specifically as follows: a training data set including a plurality of training samples is obtained, wherein each training sample includes a fourth The channel state information, the fourth channel state information is the channel state information from the network device to the terminal device; the plurality of fourth channel state information is processed by the first neural network and the second neural network to obtain a plurality of fifth channel state information, so The fifth channel state information is the channel state information after the fourth channel state information is processed by the first neural network and the second neural network.
  • a loss function is determined based on the fourth plurality of channel state information and the fifth plurality of channel state information.
  • Gradients of the loss function to parameters of the first neural network and the second neural network are determined, and parameters are updated based on the gradients until a convergence condition is reached. For example, referring to the training process shown in FIG. 4 above, the gradient of the loss function to the parameters of the first neural network and the second neural network can be determined, and each parameter can be updated based on the gradient until the convergence condition is reached.
  • the implementation process of processing a plurality of fourth channel state information through the first neural network and the second neural network to obtain a plurality of fifth channel state information may include: taking each fourth channel state information as the first channel state information And through the above S901 ⁇ S902 processing, the first information corresponding to each fourth channel state information is obtained; after the first information corresponding to each fourth channel state information is processed by S903 (that is, through channel feedback), and then passed The above steps S904 to S905 are processed to obtain the fifth channel state information corresponding to each fourth channel state information.
  • the above training process can be performed offline or online, and the training process will not be repeated here.
  • the present application also simulates the above-mentioned method embodiments.
  • the specific simulation scenarios and simulation results are as follows:
  • Simulation scenario Using the Deep MIMO data set, 100,000 (recorded as training data set), 30,000 (denoted as verification data set) and 10,000 data sets (denoted as test data set) were generated for the training and verification of the above neural network and testing, these datasets are both normalized between training.
  • E ⁇ x ⁇ represents the expectation of x, for example, x is or H represents the original channel matrix (that is, the actual channel matrix); Indicates the channel matrix recovered after compressed feedback, that is, the first channel state information recovered through the above S901-S905; h n is the nth column vector in H; for The nth column vector in .
  • the simulation test is as follows:
  • N t the number of transmitting antennas of the network equipment.
  • Table 1 shows the performance of the method shown in Figure 9 when Nt is varied. Referring to Table 1, it can be seen that in the technical solution provided by the embodiment of the present application, when the number of antennas increases, the recovery accuracy rate only drops by about one percentage point. Therefore, the technical solutions provided by the embodiments of the present application have good scalability when the number of antennas changes.
  • Table 2 shows the performance of the method shown in Figure 9 when the antenna array arrangement is changed. Referring to Table 2, it can be seen that even if the antenna configuration is changed, the correlation degree is still higher than 94%. Therefore, the performance of the technical solution provided by the embodiment of the present application is relatively stable when the arrangement of the antenna array is changed.
  • the channel state information processing method provided by the embodiment of the present application is described in detail above with reference to FIGS. 1 to 17 .
  • the communication device configured to execute the channel state information processing method provided in the embodiment of the present application will be described in detail below with reference to FIG. 18 .
  • an embodiment of the present application provides a communication device 1800 .
  • the communication device 1800 may be a network device or a terminal device, or a device in the network device or the terminal device, or a device that can be matched with the network device or the terminal device.
  • the communication device 1800 may include modules or units corresponding to one-to-one execution of the methods/operations/steps/actions performed by the network device or the terminal device in the above method embodiments.
  • the unit may be a hardware circuit, or However, software may also be realized by combining hardware circuits with software.
  • the communications apparatus 1800 may include: a processing module 1801 and a transceiver module 1802 .
  • FIG. 18 shows only the main components of the communication device.
  • the communication device 1800 is applicable to the communication system shown in FIG. 7 , and performs the functions of the terminal equipment in the channel state information processing method shown in FIG. 9 .
  • the communication device 1800 It can be understood as the first communication device.
  • the processing module 1801 is configured to determine a first graph model corresponding to the first channel state information, where the first channel state information is channel state information from the second communication device to the first communication device.
  • the processing module 1801 is further configured to use the first neural network to process the first graphical model to obtain first information, and the first information is used by the second communication device to restore the first channel state information.
  • the transceiver module 1802 is configured to send the first information to the second communication device.
  • the processing module 1801 is further configured to determine the first graph model corresponding to the first channel state information according to the sending and receiving antenna configuration information.
  • the transmitting and receiving antenna configuration information includes receiving antenna configuration information of the first communication device and transmitting antenna configuration information of the second communication device.
  • the first information includes auxiliary information
  • the auxiliary information is used to determine the second graph model
  • the second graph model is used to restore the first channel state information.
  • the auxiliary information includes index information of some nodes of the first graph model.
  • auxiliary information may also include mean values of features of some nodes.
  • the first information includes second channel state information and auxiliary information
  • the second channel state information includes processed first channel state information
  • the first neural network may include a first image pooling layer and a compression layer.
  • the first graph pooling layer is used to determine the third graph model and auxiliary information according to the first graph model
  • the compression layer is used to determine the second channel state information according to the third graph model.
  • the first graph pooling layer may be a graph pooling layer using a self-attention mechanism.
  • the first neural network may include a first graph convolution layer, a first graph pooling layer, and a compression layer.
  • the first graph convolution layer is used to determine the first graph model after convolution according to the first graph model
  • the first graph pooling layer is used to determine the third graph model and auxiliary information according to the first graph model after convolution
  • the compression layer is used to determine the second channel state information according to the third graph model.
  • the compression layer can be a fully connected layer.
  • the first channel state information includes delay-angle domain channel state information between the transmitting antenna of the second communication device and the receiving antenna of the first communication device.
  • the delay-angle domain channel state information includes channel state information between R receiving angles and T transmitting angles, R and T are both positive integers, and at least one of R and T is greater than 1.
  • the first graph model includes multiple nodes, and the characteristics of the nodes include channel state information between the i-th transmitting angle and the j-th receiving angle, 1 ⁇ i ⁇ T, 1 ⁇ j ⁇ R.
  • the first graph model also includes at least one edge. Wherein, each edge is connected to two nodes, and each edge indicates that the two nodes connected to the edge correspond to two adjacent receiving angles or correspond to two adjacent transmitting angles.
  • the channel state information included in the characteristics of each node includes C elements in the time domain, where C is less than or equal to the number of subcarriers.
  • the receiving antenna configuration information of the first communication device includes one or more of the following: the number of receiving antennas of the first communication device, the type of receiving antenna front, and the arrangement of receiving antenna units.
  • the transmitting antenna configuration information of the second communication device includes one or more of the following items: the number of transmitting antennas of the second communication device, the type of transmitting antenna array, and the arrangement of transmitting antenna units.
  • the processing module 1801 is further configured to determine the first channel state information according to the space-frequency domain channel state information between the transmitting antenna of the second communication device and the receiving antenna of the first communication device.
  • the first channel state information is delay-angle domain channel state information.
  • the first neural network is determined according to the training data set, and the training data set includes a plurality of fourth channel state information for the first neural network, and the fourth channel state information is information from the second communication device to the first Channel state information of the communication device.
  • the communication device 1800 may be applicable to the communication system shown in FIG. 7, and perform the function of the network device in the channel state information processing method shown in FIG. 9. In this case, the communication device 1800 may be understood as a second communication device.
  • the transceiver module 1802 is configured to receive the first information sent by the first communication device, the first information is used by the second communication device to restore the first channel state information, and the first channel state information is the information sent by the second communication device to the Channel state information of the first communication device.
  • the processing module 1801 is configured to use the second neural network to process the first information to obtain a second graphical model.
  • the processing module 1801 is further configured to determine third channel state information according to the second graph model, where the third channel state information is the recovered first channel state information.
  • the processing module 1801 is further configured to determine a third channel state according to the second graph model and configuration information of the transmitting and receiving antennas.
  • the transmitting and receiving antenna configuration information includes receiving antenna configuration information of the first communication device and transmitting antenna configuration information of the second communication device.
  • the first information includes auxiliary information
  • the auxiliary information is used to determine the second graph model.
  • the auxiliary information includes index information of some nodes of the first graph model, where the first graph model is a graph model corresponding to the first channel state information.
  • auxiliary information may also include mean values of features of some nodes.
  • the first information includes second channel state information and the auxiliary information
  • the second channel state information includes processed first channel state information
  • the second neural network may include a decompression layer and a second graph convolution layer.
  • the decompression layer is used to determine the fourth graph model according to the second channel state information
  • the second graph convolution layer is used to determine the second graph model according to the auxiliary information and the fourth graph model.
  • the second graph convolutional layer may include multiple graph convolutional layers, and direct connection channels are included between the multiple graph convolutional layers.
  • the decompression layer can be a fully connected layer.
  • the first channel state information includes delay-angle domain channel state information between the transmitting antenna of the second communication device and the receiving antenna of the first communication device.
  • the delay-angle domain channel state information includes channel state information between R receiving angles and T transmitting angles, R and T are both positive integers, and at least one of R and T is greater than 1.
  • the second graph model includes a plurality of nodes, and the characteristics of the nodes include channel state information between the i-th transmitting angle and the j-th receiving angle, 1 ⁇ i ⁇ T, 1 ⁇ j ⁇ R.
  • the second graph model also includes at least one edge. Wherein, each edge is connected to two nodes, and each edge indicates that the two nodes connected to the edge correspond to two adjacent receiving angles or correspond to two adjacent transmitting angles.
  • the channel state information included in the characteristics of each node includes C elements in the time domain, where C is less than or equal to the number of subcarriers.
  • the receiving antenna configuration information of the first communication device includes one or more of the following: the number of receiving antennas of the first communication device, the type of receiving antenna front, and the arrangement of receiving antenna units.
  • the transmitting antenna configuration information of the second communication device includes one or more of the following items: the number of transmitting antennas of the second communication device, the type of transmitting antenna array, and the arrangement of transmitting antenna units.
  • the second neural network is determined according to the training data set, the training data set includes a plurality of fourth channel state information for the second neural network, and the fourth channel state information is the information from the second communication device to the first Channel state information of the communication device.
  • the transceiver module 1802 may include a receiving module and a sending module (not shown in FIG. 18 ).
  • the sending module is used to realize the sending function of the communication device 1800
  • the receiving module is used to realize the receiving function of the communication device 1800 .
  • the communication device 1800 may further include a storage module (not shown in FIG. 18 ), where programs or instructions are stored in the storage module.
  • the processing module 1801 executes the program or instruction
  • the communication apparatus 1800 can execute the function of the network device or the terminal device in the channel state information processing method shown in FIG. 9 .
  • the processing module 1801 involved in the communication device 1800 may be implemented by a processor or a processor-related circuit component, and may be a processor or a processing unit;
  • the transceiver module 1802 may be implemented by a transceiver or a transceiver-related circuit component, and may be a transceiver Transceiver or Transceiver Unit.
  • the transceiver module 1802 in the communication device 1800 can correspond to the input and output of the chip, for example, the receiving module in the transceiver module 1802 corresponds to the input of the chip, and the sending module in the transceiver module 1802
  • the output of the corresponding chip is not limited in this application.
  • the embodiment of the present application also provides a chip system, including: a processor, the processor is coupled with a memory, and the memory is used to store programs or instructions, and when the programs or instructions are executed by the processor, the The system on chip implements the method in any one of the foregoing method embodiments.
  • processors in the chip system there may be one or more processors in the chip system.
  • the processor can be realized by hardware or by software.
  • the processor may be a logic circuit, an integrated circuit, or the like.
  • the processor may be a general-purpose processor implemented by reading software codes stored in a memory.
  • the memory can be integrated with the processor, or can be set separately from the processor, which is not limited in this application.
  • the memory can be a non-transitory processor, such as a read-only memory ROM, which can be integrated with the processor on the same chip, or can be respectively arranged on different chips.
  • the setting method of the processor is not specifically limited.
  • the chip system may be a field programmable gate array (field programmable gate array, FPGA), an ASIC, a system on chip (SoC), a CPU, or a network processing A network processor (NP), a digital signal processor (DSP), a microcontroller (micro controller unit, MCU), or a programmable logic device (PLD) ) or other integrated chips.
  • FPGA field programmable gate array
  • SoC system on chip
  • NP network processor
  • DSP digital signal processor
  • MCU microcontroller
  • PLD programmable logic device
  • An embodiment of the present application provides a communication system.
  • the communication system includes network equipment and terminal equipment.
  • the combination of the network device and the terminal device can execute the foregoing method embodiments, and reference can be made to the foregoing method embodiments for the specific execution process, which will not be repeated here.
  • the present application also provides a computer-readable storage medium on which a computer program is stored, and when the computer-readable storage medium is executed by a computer, the functions of any one of the above method embodiments are realized.
  • the present application also provides a computer program product, which implements the functions of any one of the above method embodiments when executed by a computer.
  • processor in the embodiment of the present application may be a CPU, and the processor may also be other general-purpose processors, DSP, ASIC, FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. .
  • a general-purpose processor may be a microprocessor, or the processor may be any conventional processor, or the like.
  • the memory in the embodiments of the present application may be a volatile memory or a nonvolatile memory, or may include both volatile and nonvolatile memories.
  • the non-volatile memory may be ROM, programmable read-only memory (programmable ROM, PROM), erasable programmable read-only memory (erasable PROM, EPROM), EEPROM or flash memory.
  • Volatile memory can be RAM, which acts as external cache memory.
  • RAM random access memory
  • SRAM static random access memory
  • DRAM dynamic random access memory
  • SDRAM synchronous dynamic random access memory
  • Double data rate synchronous dynamic random access memory double data rate SDRAM, DDR SDRAM
  • enhanced SDRAM enhanced synchronous dynamic random access memory
  • SLDRAM synchronous connection dynamic random access memory
  • direct rambus RAM direct rambus RAM
  • the above-mentioned embodiments may be implemented in whole or in part by software, hardware (such as circuits), firmware, or other arbitrary combinations.
  • the above-described embodiments may be implemented in whole or in part in the form of computer program products.
  • the computer program product comprises one or more computer instructions or computer programs. When the computer instruction or computer program is loaded or executed on the computer, the processes or functions according to the embodiments of the present application will be generated in whole or in part.
  • the computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable devices.
  • the computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website, computer, server or data center Transmission to another website site, computer, server or data center by wired (such as infrared, wireless, microwave, etc.).
  • the computer-readable storage medium may be any available medium that can be accessed by a computer, or a data storage device such as a server or a data center that includes one or more sets of available media.
  • the available media may be magnetic media (eg, floppy disk, hard disk, magnetic tape), optical media (eg, DVD), or semiconductor media.
  • the semiconductor medium may be a solid state drive.
  • At least one means one or more, and “multiple” means two or more.
  • At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items.
  • at least one item (unit) in a, b, c can represent: a, b, c, a-b, a-c, b-c, or a-b-c, wherein a, b, c can be single or multiple.
  • sequence numbers of the above-mentioned processes do not mean the order of execution, and the execution order of the processes should be determined by their functions and internal logic, and should not be used in the embodiments of the present application.
  • the implementation process constitutes any limitation.
  • the disclosed systems, devices and methods may be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the units is only a logical function division. In actual implementation, there may be other division methods.
  • multiple units or components can be combined or May be integrated into another system, or some features may be ignored, or not implemented.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • each functional unit in each embodiment of the present application may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit.
  • the functions described above are realized in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. According to such an understanding, the technical solution of the present application can be embodied in the form of software products, which are stored in a storage medium and include several instructions to make a computer device (which can be a personal computer, a server, or a network equipment, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application.
  • the aforementioned storage medium includes: various media capable of storing program codes such as U disk, mobile hard disk, ROM, RAM, magnetic disk or optical disk.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Theoretical Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Electromagnetism (AREA)
  • Power Engineering (AREA)
  • Radio Transmission System (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本申请提供一种信道状态信息处理方法及装置，涉及通信领域。该信道状态信息处理方法包括：首先，第一通信装置确定第一信道状态信息对应的第一图模型，第一信道状态信息为第二通信装置到第一通信装置的信道状态信息；然后，第一通信装置利用第一神经网络处理第一图模型，得到第一信息，所述第一信息用于第二通信装置恢复第一信道状态信息；最后，第一通信装置向第二通信装置发送第一信息。这样，能够保证信道状态信息反馈的有效性，提升通信性能。

Description

信道状态信息处理方法及装置
本申请要求于2021年11月18日提交国家知识产权局、申请号为202111370355.6、申请名称为“信道状态信息处理方法及装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及通信领域，尤其涉及一种信道状态信息处理方法及装置。
背景技术
在通信系统中，终端设备需要周期性或非周期性地向网络设备上报信道状态信息(channel state information，CSI)。其中，CSI中可以包括至少一个CSI报告，该CSI报告用于指示下行信道的信道状态信息。例如，终端设备可以通过测量网络设备发送的下行参考信号，以获取并上报上述至少一个CSI报告。网络设备可以根据终端设备上报的CSI报告，为终端设备分配相应的下行传输资源。
目前的终端设备通常不会向网络设备上报实际的CSI，而是通过CSI报告中传输的预编码矩阵指示(precoding matrix indicator，PMI)来指示预编码矩阵，以实现网络设备侧的预编码和下行数据传输。但是，PMI指示的预编码矩阵并非实际的CSI对应的预编码矩阵，因此会导致通信性能损失。
发明内容
本申请实施例提供一种信道状态信息处理方法及装置，能够保证CSI反馈的有效性，提升通信性能。
为达到上述目的，本申请采用如下技术方案：
第一方面，本申请提供一种信道状态信息处理方法，所述方法可以应用于第一通信装置，第一通信装置例如是终端设备或网络设备。所述方法包括：首先，第一通信装置确定第一信道状态信息对应的第一图模型，第一信道状态信息为第二通信装置到第一通信装置的信道状态信息；然后，第一通信装置利用第一神经网络处理第一图模型，得到第一信息，所述第一信息用于第二通信装置恢复第一信道状态信息；最后，第一通信装置向第二通信装置发送第一信息。
基于第一方面所述的方法可知，第一通信装置在利用第一神经网络处理第一图模型的过程中，可以保留第一图模型中重要的特征信息，省略第一图模型中不重要的特征信息，利用第一神经网络将第一信道状态信息进行压缩，减少CSI反馈开销的同时，保证CSI反馈的有效性，使得第二通信装置恢复的信道状态信息更接近原始信道，从而提升通信性能。
一种可能的实现中，第一通信装置确定第一信道状态信息对应的第一图模型，包括：第一通信装置根据收发天线配置信息确定第一信道状态信息对应的第一图模型。其中，收发天线配置信息包括第一通信装置的接收天线配置信息和第二通信装置的发射天线配置信息。
一种可能的实现中，第一信息包括辅助信息，辅助信息用于确定第二图模型，第二图 模型用于恢复第一信道状态信息。换言之，辅助信息可以帮助第二通信装置恢复第一信道状态信息，提升第二通信装置所恢复的第一信道状态信息的准确性。
可选地，辅助信息包括第一图模型的部分节点的索引信息。可以理解，第一通信装置利用第一神经网络处理第一图模型的过程中，会省略第一图模型中不重要的特征信息，比如省略第一图模型中的部分节点的特征信息。第一通信装置可以利用辅助信息指示第一图模型中被丢弃的节点，或者利用辅助信息指示第一图模型中被保留的节点，如此可以提升第二通信装置恢复第一信道状态信息的准确性。
进一步地，辅助信息还可以包括部分节点的特征的均值。如此，可以进一步帮助第二通信装置恢复第一信道状态信息，提升第二通信装置所恢复的第一信道状态信息的准确性。
可选地，第一信息包括第二信道状态信息和辅助信息，第二信道状态信息包括处理后的第一信道状态信息。
进一步地，第一神经网络可以包括第一图池化层和压缩层。其中，第一图池化层用于根据第一图模型确定第三图模型和辅助信息，压缩层用于根据第三图模型确定第二信道状态信息。可以理解，第一图池化层的输入可以是不同结构的图模型，也即是说，对于天线配置不同的第一通信装置、第二通信装置(以下统称为收发CSI的两端)，第一通信装置均可以确定第一信道状态信息对应的第一图模型，并将该第一图模型输入同样的第一神经网络中。换言之，第一神经网络的设计与训练不依赖于收发CSI的两端的天线配置。这样，在收发CSI的两端的天线配置发生变化时，可以避免重新设计、训练神经网络，适用于动态的多输入多输出(multiple input multiple output，MIMO)系统或天线配置不同的多个MIMO系统。
其中，第一图池化层可以为采用自注意力机制的图池化层。由于自注意机制的图池化层需要训练的参数较少，并能够充分提取第一图模型的节点的特征信息和图的拓扑结构信息，从而可以提升神经网络的训练效率和处理能力。
进一步地，第一神经网络可以包括第一图卷积层、第一图池化层和压缩层。其中，第一图卷积层用于根据第一图模型确定卷积后的第一图模型，第一图池化层用于根据卷积后的第一图模型确定第三图模型和辅助信息，压缩层用于根据第三图模型确定第二信道状态信息。换言之，第一神经网络的设计与训练不依赖于收发CSI的两端的天线配置。这样，在收发CSI的两端的天线配置发生变化时，可以避免重新设计、训练神经网络，适用于动态的MIMO系统或天线配置不同的多个MIMO系统。
进一步地，压缩层可以为全连接层。
一种可能的实现中，第一信道状态信息包括第二通信装置的发射天线与第一通信装置的接收天线之间的时延-角度域信道状态信息。其中，时延-角度域信道状态信息包括R个接收角度和T个发射角度之间的信道状态信息，R和T均为正整数，且R和T中至少一项大于1。第一图模型包括多个节点，节点的特征包括第i个发射角度与第j个接收角度之间的信道状态信息，1≤i≤T，1≤j≤R。第一图模型还包括至少一条边。其中，每条边与两个节点连接，每条边表示与边相连的两个节点对应两个相邻的接收角度或对应两个相邻的发射角度。
可选地，每个节点的特征包括的信道状态信息在时域上包括C个元素，C小于或等于子载波数。其中，由于信道状态信息在时域上存在稀疏性，因此适当减少每个节点的特征包括的信道状态信息在时域上包括的元素数量，不会影响CSI的准确性。换言之，当C小于子 载波数时，可以在不影响第二通信装置准确恢复第一信道状态信息的基础上，减少通信装置的数据处理量，提升通信装置的处理速率。
一种可能的实现中，第一通信装置的接收天线配置信息包括如下一项或多项：第一通信装置的接收天线数量、接收天线阵面的类型、接收天线单元的排列方式。第二通信装置的发射天线配置信息包括如下一项或多项：第二通信装置的发射天线数量、发射天线阵面的类型、发射天线单元的排列方式。
一种可能的实现中，第一通信装置确定第一信道状态信息，包括：第一通信装置根据第二通信装置的发射天线和第一通信装置的接收天线之间的空间-频率域信道状态信息，确定第一信道状态信息。其中，第一信道状态信息为时延-角度域信道状态信息。
一种可能的实现中，第一神经网络是依据训练数据集确定的，训练数据集包括针对第一神经网络的多个第四信道状态信息，第四信道状态信息为第二通信装置到第一通信装置的信道状态信息。
第二方面，本申请提供一种信道状态信息处理方法，所述方法可以应用于第二通信装置，第二通信装置例如是网络设备或终端设备。所述方法包括：首先，第二通信装置接收第一通信装置发送的第一信息，第一信息用于第二通信装置恢复第一信道状态信息，第一信道状态信息为第二通信装置到所述第一通信装置的信道状态信息。然后，第二通信装置利用第二神经网络处理第一信息，得到第二图模型。最后，第二通信装置根据第二图模型确定第三信道状态信息，第三信道状态信息为恢复的第一信道状态信息。
一种可能的实现中，第二通信装置根据第二图模型确定第三信道状态信息，包括：第二通信装置根据第二图模型和收发天线配置信息确定第三信道状态。其中，收发天线配置信息包括第一通信装置的接收天线配置信息和第二通信装置的发射天线配置信息。
一种可能的实现中，第一信息包括辅助信息，辅助信息用于确定第二图模型。
可选地，辅助信息包括第一图模型的部分节点的索引信息，第一图模型为与第一信道状态信息对应的图模型。
进一步地，辅助信息还可以包括部分节点的特征的均值。
可选地，第一信息包括第二信道状态信息和辅助信息，第二信道状态信息包括处理后的第一信道状态信息。
进一步地，第二神经网络可以包括解压层和第二图卷积层。其中，解压层用于根据第二信道状态信息确定第四图模型，第二图卷积层用于根据辅助信息和第四图模型确定第二图模型。
进一步地，第二图卷积层可以包括多个图卷积层，多个图卷积层之间包括直连通道。这样，可以提高神经网络的训练效率和稳定性。
进一步地，解压层可以为全连接层。
一种可能的实现中，第一信道状态信息包括第二通信装置的发射天线与第一通信装置的接收天线之间的时延-角度域信道状态信息。其中，时延-角度域信道状态信息包括R个接收角度和T个发射角度之间的信道状态信息，R和T均为正整数，且R和T中至少一项大于1。第二图模型包括多个节点，节点的特征包括第i个发射角度与第j个接收角度之间的信道状态信息，1≤i≤T，1≤j≤R。第二图模型还包括至少一条边。其中，每条边与两个节点连接，每条边表示与边相连的两个节点对应两个相邻的接收角度或对应两个相邻的发射 角度。
可选地，每个节点的特征包括的信道状态信息在时域上包括C个元素，C小于或等于子载波数。
一种可能的实现中，第一通信装置的接收天线配置信息包括如下一项或多项：第一通信装置的接收天线数量、接收天线阵面的类型、接收天线单元的排列方式。第二通信装置的发射天线配置信息包括如下一项或多项：第二通信装置的发射天线数量、发射天线阵面的类型、发射天线单元的排列方式。
一种可能的实现中，第二神经网络是依据训练数据集确定的，训练数据集包括针对第二神经网络的多个第四信道状态信息，第四信道状态信息为第二通信装置到第一通信装置的信道状态信息。
此外，第二方面所述的方法的技术效果可以参考第一方面所述的方法的技术效果，此处不再赘述。
第三方面，提供一种第一通信装置。该第一通信装置包括处理模块和收发模块。其中，处理模块，用于确定第一信道状态信息对应的第一图模型，第一信道状态信息为第二通信装置到第一通信装置的信道状态信息。处理模块，还用于利用第一神经网络处理第一图模型，得到第一信息，所述第一信息用于第二通信装置恢复第一信道状态信息。收发模块，用于向第二通信装置发送第一信息。
一种可能的实现中，处理模块，还用于根据收发天线配置信息确定第一信道状态信息对应的第一图模型。其中，收发天线配置信息包括第一通信装置的接收天线配置信息和第二通信装置的发射天线配置信息。
一种可能的实现中，第一信息包括辅助信息，辅助信息用于确定第二图模型，第二图模型用于恢复第一信道状态信息。
可选地，辅助信息包括第一图模型的部分节点的索引信息。
进一步地，辅助信息还可以包括部分节点的特征的均值。
可选地，第一信息包括第二信道状态信息和辅助信息，第二信道状态信息包括处理后的第一信道状态信息。
进一步地，第一神经网络可以包括第一图池化层和压缩层。其中，第一图池化层用于根据第一图模型确定第三图模型和辅助信息，压缩层用于根据第三图模型确定第二信道状态信息。
其中，第一图池化层可以为采用自注意力机制的图池化层。
进一步地，第一神经网络可以包括第一图卷积层、第一图池化层和压缩层。其中，第一图卷积层用于根据第一图模型确定卷积后的第一图模型，第一图池化层用于根据卷积后的第一图模型确定第三图模型和辅助信息，压缩层用于根据第三图模型确定第二信道状态信息。
进一步地，压缩层可以为全连接层。
一种可能的实现中，第一信道状态信息包括第二通信装置的发射天线与第一通信装置的接收天线之间的时延-角度域信道状态信息。其中，时延-角度域信道状态信息包括R个接收角度和T个发射角度之间的信道状态信息，R和T均为正整数，且R和T中至少一项大于1。第一图模型包括多个节点，节点的特征包括第i个发射角度与第j个接收角度之间的信道状 态信息，1≤i≤T，1≤j≤R。第一图模型还包括至少一条边。其中，每条边与两个节点连接，每条边表示与边相连的两个节点对应两个相邻的接收角度或对应两个相邻的发射角度。
可选地，每个节点的特征包括的信道状态信息在时域上包括C个元素，C小于或等于子载波数。
一种可能的实现中，第一通信装置的接收天线配置信息包括如下一项或多项：第一通信装置的接收天线数量、接收天线阵面的类型、接收天线单元的排列方式。第二通信装置的发射天线配置信息包括如下一项或多项：第二通信装置的发射天线数量、发射天线阵面的类型、发射天线单元的排列方式。
一种可能的实现中，处理模块，还用于根据第二通信装置的发射天线和第一通信装置的接收天线之间的空间-频率域信道状态信息，确定第一信道状态信息。其中，第一信道状态信息为时延-角度域信道状态信息。
一种可能的实现中，第一神经网络是依据训练数据集确定的，训练数据集包括针对第一神经网络的多个第四信道状态信息，第四信道状态信息为第二通信装置到第一通信装置的信道状态信息。
可选地，收发模块可以包括接收模块和发送模块。其中，接收模块用于实现第三方面所述的第一通信装置的接收功能，发送模块用于实现第三方面所述的第一通信装置的发送功能。
可选地，第三方面所述的第一通信装置还可以包括存储模块，该存储模块存储有程序或指令。当处理模块执行该程序或指令时，使得该第一通信装置可以执行第一方面所述的方法。
需要说明的是，第三方面所述的第一通信装置可以是终端设备或网络设备，也可以是设置于终端设备或网络设备中的芯片(系统)或其他部件或组件，还可以是包含终端设备或网络设备的通信装置，本申请对此不做限定。
此外，第三方面所述的第一通信装置的技术效果可以参考第一方面所述的方法的技术效果，此处不再赘述。
第四方面，提供一种第二通信装置。该第二通信装置包括：处理模块和收发模块。其中，收发模块，用于接收第一通信装置发送的第一信息，所述第一信息用于第二通信装置恢复第一信道状态信息，第一信道状态信息为第二通信装置到所述第一通信装置的信道状态信息。处理模块，用于利用第二神经网络处理第一信息，得到第二图模型。处理模块，还用于根据第二图模型确定第三信道状态信息，第三信道状态信息为恢复的第一信道状态信息。
一种可能的实现中，处理模块，还用于根据第二图模型和收发天线配置信息确定第三信道状态。其中，收发天线配置信息包括第一通信装置的接收天线配置信息和第二通信装置的发射天线配置信息。
一种可能的实现中，第一信息包括辅助信息，辅助信息用于确定第二图模型。
可选地，辅助信息包括第一图模型的部分节点的索引信息，第一图模型为与第一信道状态信息对应的图模型。
进一步地，辅助信息还可以包括部分节点的特征的均值。
可选地，第一信息包括第二信道状态信息和所述辅助信息，第二信道状态信息包括处 理后的第一信道状态信息。
进一步地，第二神经网络可以包括解压层和第二图卷积层。其中，解压层用于根据第二信道状态信息确定第四图模型，第二图卷积层用于根据辅助信息和第四图模型确定第二图模型。
进一步地，第二图卷积层可以包括多个图卷积层，多个图卷积层之间包括直连通道。
进一步地，解压层可以为全连接层。
一种可能的实现中，第一信道状态信息包括第二通信装置的发射天线与第一通信装置的接收天线之间的时延-角度域信道状态信息。其中，时延-角度域信道状态信息包括R个接收角度和T个发射角度之间的信道状态信息，R和T均为正整数，且R和T中至少一项大于1。第二图模型包括多个节点，节点的特征包括第i个发射角度与第j个接收角度之间的信道状态信息，1≤i≤T，1≤j≤R。第二图模型还包括至少一条边。其中，每条边与两个节点连接，每条边表示与边相连的两个节点对应两个相邻的接收角度或对应两个相邻的发射角度。
可选地，每个节点的特征包括的信道状态信息在时域上包括C个元素，C小于或等于子载波数。
一种可能的实现中，第一通信装置的接收天线配置信息包括如下一项或多项：第一通信装置的接收天线数量、接收天线阵面的类型、接收天线单元的排列方式。第二通信装置的发射天线配置信息包括如下一项或多项：第二通信装置的发射天线数量、发射天线阵面的类型、发射天线单元的排列方式。
一种可能的实现中，第二神经网络是依据训练数据集确定的，训练数据集包括针对第二神经网络的多个第四信道状态信息，第四信道状态信息为第二通信装置到第一通信装置的信道状态信息。
可选地，收发模块可以包括接收模块和发送模块。其中，接收模块用于实现第四方面所述的第二通信装置的接收功能，发送模块用于实现第四方面所述的第二通信装置的发送功能。
可选地，第四方面所述的第二通信装置还可以包括存储模块，该存储模块存储有程序或指令。当处理模块执行该程序或指令时，使得该第二通信装置可以执行第二方面所述的方法。
需要说明的是，第四方面所述的第二通信装置可以是网络设备或终端设备，也可以是设置于网络设备或终端设备中的芯片(系统)或其他部件或组件，还可以是包含网络设备或终端设备的通信装置，本申请对此不做限定。
此外，第四方面所述的第二通信装置的技术效果可以参考第二方面所述的方法的技术效果，此处不再赘述。
第五方面，提供一种通信装置。该通信装置用于执行第一方面至第二方面中任意一种实现方式所述的信道状态信息处理方法。
在本申请中，第五方面所述的通信装置可以为第一通信装置或第二通信装置，也可以是设置于第一通信装置或第二通信装置中的芯片(系统)或其他部件或组件，还可以是包含第一通信装置或第二通信装置的通信装置，本申请对此不做限定。其中，第一通信装置用于执行第一方面中任一种可能的实现方式所述的信道状态信息处理方 法，第二通信装置用于执行第二方面中任一种可能的实现方式所述的信道状态信息处理方法。
应理解，第五方面所述的通信装置包括实现上述第一方面至第二方面中任一方面所述的信道状态信息处理方法相应的模块、单元、或手段(means)，该模块、单元、或手段可以通过硬件实现，软件实现，或者通过硬件执行相应的软件实现。该硬件或软件包括一个或多个用于执行上述信道状态信息处理方法所涉及的功能的模块或单元。
第六方面，提供一种通信装置。该通信装置包括：处理器，该处理器用于执行第一方面至第二方面中任意一种可能的实现方式所述的信道状态信息处理方法。
一种可能的实现中，第六方面所述的通信装置还可以包括收发器。该收发器可以为收发电路或接口电路。该收发器可以用于第六方面所述的通信装置与其他通信装置通信。
一种可能的实现中，第六方面所述的通信装置还可以包括存储器。该存储器可以与处理器集成在一起，也可以分开设置。该存储器可以用于存储第一方面至第二方面中任一方面所述的信道状态信息处理方法所涉及的计算机程序和/或数据。
在本申请中，第六方面所述的通信装置可以为第一通信装置或第二通信装置，也可以是设置于第一通信装置或第二通信装置中的芯片(系统)或其他部件或组件，还可以是包含第一通信装置或第二通信装置的通信装置，本申请对此不做限定。其中，第一通信装置用于执行第一方面中任一种可能的实现方式所述的信道状态信息处理方法，第二通信装置用于执行第二方面中任一种可能的实现方式所述的信道状态信息处理方法。
第七方面，提供一种通信装置。该通信装置包括：处理器，该处理器与存储器耦合，该处理器用于执行存储器中存储的计算机程序，以使得该通信装置执行第一方面至第二方面中任意一种可能的实现方式所述的信道状态信息处理方法。
一种可能的实现中，第七方面所述的通信装置还可以包括收发器。该收发器可以为收发电路或接口电路。该收发器可以用于第七方面所述的通信装置与其他通信装置通信。
在本申请中，第七方面所述的通信装置可以为第一通信装置或第二通信装置，也可以是设置于第一通信装置或第二通信装置中的芯片(系统)或其他部件或组件，还可以是包含第一通信装置或第二通信装置的通信装置，本申请对此不做限定。其中，第一通信装置用于执行第一方面中任一种可能的实现方式所述的信道状态信息处理方法，第二通信装置用于执行第二方面中任一种可能的实现方式所述的信道状态信息处理方法。
第八方面，提供一种通信装置。该通信装置包括：处理器和接口电路。其中，接口电路，用于接收代码指令并传输至所述处理器。处理器用于运行上述代码指令以执行第一方面至第二方面中任意一种实现方式所述的信道状态信息处理方法。
一种可能的实现中，第八方面所述的通信装置还可以包括存储器。该存储器可以与处理器集成在一起，也可以分开设置。该存储器可以用于存储第一方面至第二方面中任一方面所述的信道状态信息处理方法所涉及的计算机程序和/或数据。
在本申请中，第八方面所述的通信装置可以为第一通信装置或第二通信装置，也 可以是设置于第一通信装置或第二通信装置中的芯片(系统)或其他部件或组件，还可以是包含第一通信装置或第二通信装置的通信装置，本申请对此不做限定。其中，第一通信装置用于执行第一方面中任一种可能的实现方式所述的信道状态信息处理方法，第二通信装置用于执行第二方面中任一种可能的实现方式所述的信道状态信息处理方法。
第九方面，提供一种通信装置。该通信装置包括处理器和存储介质，该存储介质存储有指令，该指令被处理器运行时，使得第一方面至第二方面中任意一种可能的实现方式所述的信道状态信息处理方法被实现。
在本申请中，第九方面所述的通信装置可以为第一通信装置或第二通信装置，也可以是设置于第一通信装置或第二通信装置中的芯片(系统)或其他部件或组件，还可以是包含第一通信装置或第二通信装置的通信装置，本申请对此不做限定。其中，第一通信装置用于执行第一方面中任一种可能的实现方式所述的信道状态信息处理方法，第二通信装置用于执行第二方面中任一种可能的实现方式所述的信道状态信息处理方法。
第十方面，提供一种处理器。其中，处理器用于执行第一方面至第二方面中任意一种可能的实现方式所述的信道状态信息处理方法。
第十一方面，提供一种通信系统。该通信系统包括第一通信装置或第二通信装置。其中，第一通信装置用于执行第一方面中任一种可能的实现方式所述的信道状态信息处理方法，第二通信装置用于执行第二方面中任一种可能的实现方式所述的信道状态信息处理方法。
第十二方面，提供一种计算机可读存储介质，该计算机可读存储介质包括计算机程序或指令，当该计算机程序或指令被处理器运行时，使得第一方面至第二方面中任意一种可能的实现方式所述的信道状态信息处理方法被实现。
第十三方面，提供一种计算机程序产品，该计算机程序产品包括指令，当该指令被处理器运行时，使得第一方面至第二方面中任意一种可能的实现方式所述的信道状态信息处理方法被实现。
第十四方面，提供一种芯片，该芯片包括处理逻辑电路和接口电路。其中，处理逻辑电路的数量可以是一个或多个，接口电路的数量可以是多个。
其中，接口电路，用于接收代码指令并传输至所述处理逻辑电路。处理逻辑电路用于运行上述代码指令以执行第一方面至第二方面中任意一种实现方式所述的信道状态信息处理方法。
可选地，该芯片可以包括存储器，该存储器可以与处理逻辑电路集成在一起，也可以分开设置。该存储器可以用于存储第一方面至第二方面中任一方面所述的信道状态信息处理方法所涉及的计算机程序和/或数据。
在本申请中，第十四方面所述的芯片可以位于第一通信装置或第二通信装置，可以位于一个通信系统中的第一通信装置或第二通信装置。其中，芯片位于第一通信装置时用于执行第一方面中任一种可能的实现方式所述的信道状态信息处理方法，芯片位于第二通信装置时用于执行第二方面中任一种可能的实现方式所述的信道状态信息处理方法。
第五方面至第十四方面中的任一种实现方式所带来的技术效果可参见第一方面至第二方面中任一方面对应实现方式所带来的技术效果，此处不再赘述。
本申请在上述各方面提供的实现方式的基础上，还可以进行进一步组合以提供更多实现方式。
附图说明
图1为本申请实施例提供的单阵面的天线配置一种示意图；
图2为本申请实施例提供的多阵面的天线配置一种示意图；
图3为本申请实施例提供的全连接神经网络的一种示意图；
图4为本申请实施例提供的损失函数优化的一种示意图；
图5为本申请实施例提供的误差反向传播的一种示意图；
图6为本申请实施例提供的图神经网络的层级结构的一种示意图；
图7为本申请实施例提供的通信系统的一种架构示意图；
图8为本申请实施例提供的通信装置的一种结构示意图；
图9为本申请实施例提供的信道状态信息处理方法的交互示意图；
图10为本申请实施例提供的一种第一子图模型的示意图；
图11为本申请实施例提供的一种收发天线连接示意图；
图12为本申请实施例提供的第一图模型的示意图一；
图13为本申请实施例提供的第一图模型的示意图二；
图14为本申请实施例提供的神经网络的结构示意图一；
图15为本申请实施例提供的神经网络的结构示意图二；
图16为本申请实施例提供的神经网络的结构示意图三；
图17为本申请实施例提供的神经网络的结构示意图四；
图18为本申请实施例提供的通信装置的结构示意图。
具体实施方式
为了方便理解本申请实施例中的方案，首先给出相关技术的简要介绍。
1、MIMO系统中的预编码
在通信系统中，通常使用MIMO技术增加系统容量，即在发送端和接收端同时使用多根天线进行通信。理论上，使用多根天线并结合空分复用，可以成倍的增加系统容量，提升通信速率。但是，多根天线的使用也带来了干扰增强的问题。因此，需要对信号进行一定的处理以抑制干扰带来的影响。这种通过信号处理抑制干扰的方法可以在接收端实现，也可以在发送端实现。其中，在发送端实现时，发送端可以对待发送信号进行预处理，再经过MIMO信道发送，这种方式被称为预编码。下面是对预编码的介绍。
为了识别信道矩阵(记信道矩阵为H)中有用的通道，需要把多个通道转化成类似于单输入单输出(simple input simple output，SISO)的一对一模式，也就是将多个MIMO交叉通道转换成多个平行的一对一信道。这个过程可以通过信道矩阵奇异值分解(singular value decomposition，SVD)实现。其中，SVD的公式为：H＝UΣV T，U和V为正交矩阵，Σ为对角矩阵，Σ中对角线上的元素为奇异值，上标T表示转置操作。在此基础上，假设s为发射信号，n为噪声，r为接收信号，那么r＝H*s+n可变换为r＝UΣV T*s+n。换言之，如果待发送数据为x，那么可使用s＝Vx处理数据。这样，接收端可以使用Σ -1U T进行解码，以得到 无干扰的多个一对一信道。发送端使用s＝Vx处理数据的过程即为预编码操作，V为预编码矩阵。
一般情况下，需要根据完整的H进行奇异值分解，得到对应的预编码矩阵V。而在实际的通信系统中，由于实际的CSI的数据量非常大，如果终端设备向网络设备上报实际的CSI，则会导致很大的传输开销。在此情况下，标准给出了一系列V矩阵，即码本(codebook)，由收发两端协调，从中选择合适的预编码矩阵。具体来说，目前的终端设备通常不会向网络设备上报实际的CSI，而是终端设备通过反馈PMI，指示一个在标准给定的码本中可以使H容量最大的V，从而实现网络设备侧的预编码处理，这种方式相当于隐式地指示了实际的CSI。其中，PMI指示的V并非实际的CSI对应的预编码矩阵，因此会导致性能损失。
2、天线阵面的类型
在MIMO系统中的硬件设备上，天线阵面的类型可以包括单阵面(single-panel)和多阵面(multi-panel)。下面分别介绍：
单阵面：请参照图1，图1为本申请实施例提供的单阵面的天线配置一种示意图。假设N 1为天线阵面上横向的极化天线对数，N 2为天线阵面上纵向的极化天线对数，那么图1中的①所示的单阵面的天线配置为(N 1,N 2)＝(2,2)，图1中的②所示的单阵面的天线配置为(N 1,N 2)＝(4,1)。
多阵面：请参照图2，图2为本申请实施例提供的多阵面的天线配置一种示意图。假设N g为天线阵面数量，N 1和N 2分别为单个天线阵面上横向和纵向的极化天线对数，那么图2所示的多阵面的天线配置为(N g,N 1,N 2)＝(4,2,2)。
3、全连接神经网络
全连接神经网络又叫多层感知机(multilayer perceptron，MLP)。请参照图3，图3为本申请实施例提供的全连接神经网络的一种示意图，一个MLP可以包含一个输入层(包括图3中的x 1～x 4)，一个输出层(包括图3中的y 1～y 6)，以及多个隐藏层。每个隐藏层可以包含数个节点(图3中的黑色圆圈)，称为神经元。其中，相邻两层的神经元间之间两两相连。下面简单介绍全连接神经网络的实现原理。
以相邻两层的神经元举例，下一层的神经元的输出h为经过激活函数的所有与之相连的上一层神经元x的加权和，以矩阵表示可以参照如下公式：
h＝f(wx+b)
其中，x为神经元，w为权重矩阵，b为偏置向量，f为激活函数。在此基础上，可以得出神经网络的输出递归表达式为：
y＝f n(w nf n-1(…)+b n)
简而言之，可以将神经网络理解为一个从输入数据集合到输出数据集合的映射关系。而通常神经网络都是随机初始化的，也即w和b是随机数，用已有数据从随机的w和b得到这个映射关系的过程被称为神经网络的训练。
训练的具体方式可以包括：采用损失函数(loss function)对神经网络的输出结果进行评价，将误差反向传播，通过梯度下降的方法迭代优化w和b，直到损失函数达到最小值。其中，请参照图4，图4为本申请实施例提供的损失函数优化的一种示意图，梯度下降过程的表达式可以如下：
Figure PCTCN2022130632-appb-000001
其中，θ为待优化参数(如w和b)，L为损失函数，η为学习率，η可以用于控制梯度下降的步长。反向传播的过程可以利用到求偏导的链式法则实现，即前一层参数的梯度可以由后一层参数的梯度递推计算得到。其中，请参照图5，图5为本申请实施例提供的误差反向传播的一种示意图，误差反向传播的公式可以表达为：
Figure PCTCN2022130632-appb-000002
其中，L为损失函数，w ij为节点j(也即图3中的神经元)连接节点i的权重，s i为节点i上的输入加权和。
4、图神经网络(graph neural network，GNN)
GNN是一种专门为图数据(以下统称为图模型)提出的神经网络模型，有独立于图规模的良好特性，也即是同样的GNN可以处理不同结构和/或不同规模的图模型。如图6所示，图6为本申请实施例提供的GNN的层级结构的一种示意图，GNN将图模型作为输入，输出的是图、节点、边或者子图的p维向量表示。在GNN的每一层中可以进行图卷积操作，比如图6中的第一图卷积层，类似于传统的卷积操作，图卷积操作可以认为是对邻点信息的进行聚合操作，公式表达可以如下：
Figure PCTCN2022130632-appb-000003
其中，k表示第k层图卷积，x v表示节点v的特征，
Figure PCTCN2022130632-appb-000004
表示节点v的邻点在上一层(第k-1层)的隐藏层状态信息，Agg (k)(·,·；θ)表示第k层的聚合函数，由所有节点共用，θ是待训练的参数。这样，通过多层图卷积操作，节点能够不断地根据拓扑结构聚合邻点信息，并更新该节点的特征。
基于上述简要介绍，可以看出，目前的终端设备通常不会向网络设备上报实际的CSI，而是通过CSI报告中的PMI来指示预编码矩阵，以实现预编码矩阵的选择。PMI指示的预编码矩阵并非实际的CSI对应的预编码矩阵，因此会导致性能损失。因此，如何传输实际的CSI成为亟需解决的问题。
为了解决上述问题，本申请实施例提供一种技术方案，该技术方案包括通信系统、应用于该通信系统的信道状态信息处理方法和通信装置等。下面将结合附图，对本申请提供的技术方案进行说明。
本申请实施例的技术方案可以应用于无线通信系统，例如：无线通信系统可以为第四代(4th generation，4G)通信系统(例如，长期演进系统(long term evolution，LTE)系统)，第五代(5th generation，5G)通信系统(例如，新空口(new radio，NR)系统)，5G之后演进的移动通信系统(例如，6G通信系统)，及窄带物联网系统(narrow band-internet of things，NB-IoT)等。本申请实施例的技术方案还可以应用于卫星通信系统或者非陆地通信网络(non-terrestrial network，NTN)通信系统中，其中，卫星通信系统或NTN通信系统可以与无线通信系统相融合。本申请实施例的技术方案还可以应用于卫星星间链路通信系统、无线投屏系统、虚拟现实(virtual reality，VR)通信系统、接入回传一体化(integrated access and backhual，IAB)系统、无线保真(wireless fidelity，Wi-Fi)通信系统、光通信系统等，对此不作限定。
本申请将围绕可包括多个设备、组件、模块等的系统来呈现各个方面、实施例或特征。应当理解和明白的是，各个系统可以包括另外的设备、组件、模块等，并且/或者可以并不 包括结合附图讨论的所有设备、组件、模块等。此外，还可以使用这些方案的组合。
另外，在本申请实施例中，“示例地”、“例如”等词用于表示作例子、例证或说明。本申请中被描述为“示例”的任何实施例或设计方案不应被解释为比其它实施例或设计方案更优选或更具优势。确切而言，使用示例的一词旨在以具体方式呈现概念。
本申请实施例描述的网络架构以及业务场景是为了更加清楚的说明本申请实施例的技术方案，并不构成对于本申请实施例提供的技术方案的限定，本领域普通技术人员可知，随着网络架构的演变和新业务场景的出现，本申请实施例提供的技术方案对于类似的技术问题，同样适用。
本申请实施例提供一种通信系统，该通信系统可以适用于第一通信装置和第二通信装置之间的通信。本申请实施例提供的通信系统中可以包括一个或多个第一通信装置、一个或多个第二通信装置，本申请实施例对于通信系统中第一通信装置和第二通信装置各自的个数不作限定。第一通信装置例如是终端设备或网络设备，第二通信装置例如是网络设备或终端设备。本申请实施例以第一通信装置为终端设备，第二通信装置为网络设备举例，说明本申请实施例提供的方案，在此统一说明，下文不再赘述。
作为一种示例，图7为本申请实施例提供的通信系统的一种架构示意图，如图7所示，该通信系统中可以包括网络设备和终端设备，网络设备和终端设备之间可以通过无线的方式连接。网络设备和终端设备之间可以交互数据和/或控制信令等。
可选地，本申请实施例中的网络设备，是一种将终端设备接入到无线网络的设备。网络设备可以为无线接入网中的节点，又可以称为基站，还可以称为无线接入网(radio access network，RAN)节点(或设备)，其中，基站可以是一个分布式天线系统，与某个终端设备通信的可以是基站的一个射频头端。例如，网络设备可以包括LTE系统或演进的LTE系统(LTE-Advanced，LTE-A)中的演进型基站(evolved Node B，eNB或eNodeB，)，如传统的宏基站eNB和异构网络场景下的微基站eNB；或者也可以包括5G NR系统中的下一代节点B(next generation node B，gNB)，或者还可以包括传输接收点(transmitting and receiving point，TRP)、发射点(transmitting point，TP)、家庭基站(例如，home evolved NodeB，或home Node B，HNB)、基带单元(base band unit，BBU)、基带池BBU pool，或Wi-Fi接入点(access point，AP)、移动交换中心以及设备到设备(device-to-device，D2D)、车到万物(vehicle-to-everything，V2X)、机器到机器(machine-to-machine，M2M)通信中承担基站功能的设备等；或者也可以是5G网络中的基站设备或者未来演进的公共陆地移动网络(public land mobile network，PLMN)中的网络设备；或者还可以是可穿戴设备或车载设备等；再或者还可以包括云接入网(cloud radio access network，CloudRAN)系统中的集中式单元(centralized unit，CU)和分布式单元(distributed unit，DU)；又或者可以包括NTN中的网络设备，即可以部署于高空平台或者卫星。在NTN中，网络设备可以作为层1(L1)中继(relay)，或者可以作为基站，或者可以作为DU，或者可以作为IAB节点，本申请实施例并不限定。当然，网络设备也可以为核心网中的节点。
其中，本申请实施例中的终端设备，可以是用于实现无线通信功能的设备，例如终端或者可用于终端中的芯片等。其中，终端可以是5G网络或者未来演进的PLMN中的用户设备(user equipment，UE)、接入终端、终端单元、终端站、移动站、移动台、远方站、远程终端、移动设备、无线通信设备、终端代理或终端装置、工业场景中的各类终端(例如 机器人或装配有无线传输模块的机械臂)等。接入终端可以是蜂窝电话、无绳电话、会话启动协议(session initiation protocol，SIP)电话、无线本地环路(wireless local loop，WLL)站、个人数字处理(personal digital assistant，PDA)、具有无线通信功能的手持设备、计算设备或连接到无线调制解调器的其它处理设备、车载设备或可穿戴设备，VR终端设备、增强现实(augmented reality，AR)终端设备、工业控制(industrial control)中的无线终端、无人驾驶(self driving)中的无线终端、无人机、传感器、执行器、卫星终端、远程医疗(remote medical)中的无线终端、智能电网(smart grid)中的无线终端、运输安全(transportation safety)中的无线终端、智慧城市(smart city)中的无线终端、智慧家庭(smart home)中的无线终端等。或者，终端可以是车联网(vehicle-to-everything，V2X)中的终端(例如车联网设备)、设备到设备(Device to Device)通信中的终端、或者机器到机器(machine to machine，M2M)通信中的终端等。
可选地，本申请实施例中的网络设备、终端设备可以部署在陆地上，包括室内或室外、手持或车载；也可以部署在水面上；还可以部署在空中的飞机、气球和人造卫星上。本申请的实施例对网络设备和终端设备的应用场景不做限定。
本申请实施例并未对本申请实施例提供的方法的执行主体的具体结构特别限定，只要能够通过运行记录有本申请实施例的提供的方法的代码的程序，以根据本申请实施例提供的方法进行通信即可，例如，本申请实施例提供的信道状态信息处理方法的执行主体可以是第一通信装置或第二通信装置，或者，是网络设备、终端设备中能够调用程序并执行程序的功能模块。
换言之，本申请实施例中的第一通信装置或第二通信装置的相关功能可以由一个设备实现，也可以由多个设备共同实现，还可以是由一个设备内的一个或多个功能模块实现，本申请实施例对此不作具体限定。可以理解的是，上述功能既可以是硬件设备中的网络元件，也可以是在专用硬件上运行的软件功能，或者是硬件与软件的结合，或者是平台(例如，云平台)上实例化的虚拟化功能。
例如，本申请实施例中的第一通信装置或第二通信装置的相关功能可以通过图8中的通信装置800来实现。图8所示为本申请实施例提供的通信装置800的一种结构示意图。该通信装置800可以包括一个或多个处理器801，通信线路802，以及至少一个通信接口(图8中仅是示例性的以包括通信接口804，以及一个处理器801为例进行说明)，可选地还可以包括存储器803。
处理器801可以是一个中央处理器(central processing unit，CPU)，微处理器，特定应用集成电路(application-specific integrated circuit，ASIC)，或一个或多个用于控制本申请方案程序执行的集成电路。
通信线路802可包括一通路，用于连接不同组件之间。示例性的，该通信线路802可以为总线，如地址总线、数据总线、控制总线等。
通信接口804可以是收发模块，可以用于与其他设备或通信网络通信。例如，所述收发模块可以是收发器、收发机一类的装置。可选地，所述通信接口804也可以是位于处理器801内的收发电路，用以实现处理器的信号输入和信号输出。
存储器803可以是具有存储功能的装置。例如可以是只读存储器(read-only memory，ROM)或可存储静态信息和指令的其他类型的静态存储设备，随机存取存储器(random  access memory，RAM)或者可存储信息和指令的其他类型的动态存储设备，也可以是电可擦可编程只读存储器(electrically erasable programmable read-only memory，EEPROM)、只读光盘(compact disc read-only memory，CD-ROM)或其他光盘存储、光碟存储(包括压缩光碟、激光碟、光碟、数字通用光碟、蓝光光碟等)、磁盘存储介质或者其他磁存储设备、或者能够用于携带或存储具有指令或数据结构形式的期望的程序代码并能够由计算机存取的任何其他介质，但不限于此。存储器可以是独立存在，通过通信线路802与处理器相连接。存储器也可以和处理器集成在一起。
其中，存储器803用于存储执行本申请方案的计算机执行指令，并由处理器801来控制执行。处理器801用于执行存储器803中存储的计算机执行指令，从而实现本申请实施例中提供的信道状态信息处理方法。
或者，本申请实施例中，也可以是处理器801执行本申请下述实施例提供的信道状态信息处理方法中的处理相关的功能，通信接口804负责与其他设备或通信网络通信，本申请实施例对此不作具体限定。
本申请实施例中的计算机可执行指令也可以称之为应用程序代码，本申请实施例对此不作具体限定。
在具体实现中，作为一种实施例，处理器801可以包括一个或多个CPU，例如图8中的CPU0和CPU1。
在具体实现中，作为一种实施例，通信装置800可以包括多个处理器，例如图8中的处理器801和处理器808。这些处理器中的每一个可以是一个单核(single-CPU)处理器，也可以是一个多核(multi-CPU)处理器。这里的处理器可以指一个或多个设备、电路、和/或用于处理数据(例如计算机程序指令)的处理核。
在具体实现中，作为一种实施例，通信装置800还可以包括输出设备805和输入设备806。输出设备805和处理器801通信，可以以多种方式来显示信息。
以上对本申请提供的通信系统进行了介绍，下面将结合附图对本申请实施例提供的信道状态信息处理方法进行说明。
请参照图9，图9为本申请实施例提供的信道状态信息处理方法的交互示意图，该信道状态信息处理方法可以应用于上述通信系统，可以由上述通信系统中的网络设备或终端设备执行。该信道状态信息处理方法能够保证CSI的有效性，提升通信性能。该方法可以包括S901～S905，下面依次说明。
S901，终端设备确定第一信道状态信息对应的第一图模型。
下面分别对第一信道状态信息和第一图模型进行介绍。
1、对第一信道状态信息的介绍如下：
第一信道状态信息可以包括网络设备和终端设备之间的信道状态信息。具体地，第一信道状态信息表示的是网络设备到终端设备之间的CSI。第一信道状态信息可以包括网络设备的发射天线与终端设备的接收天线之间的时延-角度域信道状态信息。该时延-角度域信道状态信息可以包括N r个接收角度和N t个发射角度之间的信道状态信息，N r和N t均为正整数，且R和T中至少一项大于1。在一些情况下，N r可以记为R，N t可以记为T。另外，时延-角度域信道状态信息可以通过矩阵表示，也即时延-角度域信道状态信息可以为时延-角度域的信道矩阵。
本申请实施例中，在没有特殊说明的情况下，N r可以表示终端设备的接收天线的个数，N t可以表示网络设备的发射天线的个数，在此统一说明，以下不再赘述。
在S901之前，终端设备可以确定第一信道状态信息。下面对“终端设备确定第一信道状态信息”的实施方式以及第一信道状态信息进行详细说明。
可选地，终端设备确定第一信道状态信息，可以包括：终端设备根据网络设备的发射天线和终端设备的接收天线之间的空间-频率域信道状态信息，确定所述第一信道状态信息。其中，空间-频率域信道状态信息可以通过矩阵表示，也即空间-频率域信道状态信息可以为空间-频率域的信道矩阵。
示例性地，终端设备可以接收网络设备发送的参考信号，并依据该参考信号进行CSI估计，得到网络设备的发射天线和终端设备的接收天线之间的空间-频率域信道矩阵(记为
Figure PCTCN2022130632-appb-000005
)。其中，网络设备发送的参考信号可以是信道状态信息-参考信号(channel state information-reference signal，CSI-RS)。然后，终端设备可以对
Figure PCTCN2022130632-appb-000006
进行离散傅里叶变换(discrete fourier transform，DFT)，得到时延-角度域的信道矩阵(记为
Figure PCTCN2022130632-appb-000007
)，
Figure PCTCN2022130632-appb-000008
即为时延-角度域信道状态信息，也即是第一信道状态信息。
其中，假设网络设备的发射天线包括N t个发射天线，终端设备的接收天线包括N r个接收天线，则空间-频率域信道矩阵中包含N t个发射天线和N r个接收天线之间的信道状态信息，对该空间-频率域信道矩阵的空间域进行离散傅里叶变换后相应地可以得到角度域的N t个发射角度和N r个接收角度之间的信道状态信息。基于此，在上述确定
Figure PCTCN2022130632-appb-000009
的过程，可以包括如下几种实施方式(方式1～方式3)。
方式1，针对每个接收角度，确定一个对应的时延-角度域信道矩阵(记为
Figure PCTCN2022130632-appb-000010
)，i为正整数，i≤N r。
Figure PCTCN2022130632-appb-000011
包括N r个
Figure PCTCN2022130632-appb-000012
示例性地，终端设备可以根据参考信号估计N r个接收天线中第i个接收天线与N t个发射天线之间的信道，得到第i个接收天线与N t个发射天线之间的空间-频率域信道矩阵(记为
Figure PCTCN2022130632-appb-000013
)，
Figure PCTCN2022130632-appb-000014
为子载波数，
Figure PCTCN2022130632-appb-000015
表示复数域。然后，终端设备可以对
Figure PCTCN2022130632-appb-000016
进行二维DFT，得到与第i个接收角度对应的时延-角度域的信道矩阵(也即是
Figure PCTCN2022130632-appb-000017
)。可以看出，方式1得到的
Figure PCTCN2022130632-appb-000018
有N r个，每个
Figure PCTCN2022130632-appb-000019
与一个接收角度对应，所有的
Figure PCTCN2022130632-appb-000020
即为时延-角度域的信道矩阵(也即是
Figure PCTCN2022130632-appb-000021
)。其中，在没有特殊说明的情况下，
Figure PCTCN2022130632-appb-000022
可以表示子载波数，在此统一说明，以下不再赘述。
在方式1中，
Figure PCTCN2022130632-appb-000023
可以表示一个接收角度与N t个发射角度之间的信道状态信息，所有的
Figure PCTCN2022130632-appb-000024
可以表示N r个接收角度与N t个发射角度之间的信道状态信息，也即是说，
Figure PCTCN2022130632-appb-000025
可以包括N r个接收角度和N t个发射角度之间的信道状态信息。
方式2，针对每个发射角度，确定一个对应的时延-角度域信道矩阵(记为
Figure PCTCN2022130632-appb-000026
)，j为正整数，j≤N t。
Figure PCTCN2022130632-appb-000027
包括N t个
Figure PCTCN2022130632-appb-000028
示例性地，终端设备可以根据参考信号估计N t个发射天线中第j个发射天线与N r个接收天线之间的信道，得到第j个发射天线与N r个接收天线之间的空间-频率域信道矩阵(记为
Figure PCTCN2022130632-appb-000029
)，
Figure PCTCN2022130632-appb-000030
然后，终端设备可以对
Figure PCTCN2022130632-appb-000031
进行二维DFT，得到与第j个发射角度对应的时延-角度域的信道矩阵(也即是
Figure PCTCN2022130632-appb-000032
)。可以看出，方式2得到的
Figure PCTCN2022130632-appb-000033
有N t个，每个
Figure PCTCN2022130632-appb-000034
与一个发射天线对应，所有的
Figure PCTCN2022130632-appb-000035
即为时延-角度域的信道矩阵(也即是
Figure PCTCN2022130632-appb-000036
)。
在方式2中，
Figure PCTCN2022130632-appb-000037
可以表示一个发射角度与N r个接收角度之间的信道状态信息，所有的
Figure PCTCN2022130632-appb-000038
可以表示N r个接收角度与N t个发射角度之间的信道状态信息，也即是说，
Figure PCTCN2022130632-appb-000039
可以包括N r个接收角度和N t个发射角度之间的信道状态信息。
方式3，针对所有的发射角度和所有的接收角度，确定对应的时延-角度域信道矩阵(记为
Figure PCTCN2022130632-appb-000040
)。
Figure PCTCN2022130632-appb-000041
即为
Figure PCTCN2022130632-appb-000042
示例性地，终端设备可以根据参考信号估计N r个接收天线与N t个发射天线之间的信道，得到N r个接收天线与N t个发射天线之间的空间-频率域信道矩阵(记为
Figure PCTCN2022130632-appb-000043
)，
Figure PCTCN2022130632-appb-000044
然后，终端设备可以对
Figure PCTCN2022130632-appb-000045
进行三维DFT，得到时延-角度域的信道矩阵(也即是
Figure PCTCN2022130632-appb-000046
)。
在方式3中，
Figure PCTCN2022130632-appb-000047
可以表示N r个接收角度与N t个发射角度之间的信道状态信息，也即是说，
Figure PCTCN2022130632-appb-000048
可以包括N r个接收角度和N t个发射角度之间的信道状态信息。
2、对第一图模型的介绍如下：
第一图模型可以包括多个节点，一个节点的特征可以包括N t个发射角度中的第j个发射角度与N r个接收角度中的第i个接收角度之间的信道状态信息，i和j均为正整数，i≤N r，j≤N t。换言之，多个节点的特征可以包括N t个发射角度与N r个接收角度之间的信道状态信息。第一图模型还可以包括至少一条边。其中，每条边可以与两个节点连接，每条边可以表示与边相连的两个节点对应两个相邻的接收角度或对应两个相邻的发射角度。
下面对S901的实施方式以及第一图模型进行详细说明。
在一些可能的实施例中，S901，终端设备确定第一信道状态信息对应的第一图模型，可以包括：终端设备根据收发天线配置信息确定第一信道状态信息对应的第一图模型。
其中，收发天线配置信息可以包括：终端设备的接收天线配置信息和网络设备的发射天线配置信息。可选地，终端设备的接收天线配置信息可以包括如下一项或多项：终端设备的接收天线数量、接收天线阵面的类型、接收天线单元的排列方式。网络设备的发射天线配置信息可以包括如下一项或多项：网络设备的发射天线数量、发射天线阵面的类型、发射天线单元的排列方式。其中，终端设备根据上述收发天线配置信息可以确定如下一项或多项信息：网络设备的N t个发射天线构成N t1列、N t2行天线阵列，终端设备的N r个发射天线构成N r1列、N r2行天线阵列，N t＝N t1×N t2。
在实际应用中，上述收发天线配置信息可以通过协议预定义；也可以由网络设备通过信令指示给终端设备，例如，在S902之前，网络设备还可以向终端设备发送收发天线配置信息，相应地，终端设备接收来自网络设备的收发天线配置信息。本申请实施例对终端设备获取收发天线配置信息的具体实现方式不作限定。
具体地，终端设备根据收发天线配置信息确定第一信道状态信息对应的第一图模型，可以包括如下几种实现方式(方式4～方式6)。
方式4，针对每个接收角度，确定一个对应的第一子图模型，所有的第一子图模型即为第一图模型。
具体来说，以针对N r个接收角度中的第i个接收角度确定第i个接收角度对应的第一子图模型举例：终端设备可以采用上述方式1获取与第i个接收角度对应的
Figure PCTCN2022130632-appb-000049
然后，终端设备根据收发天线配置信息，确定
Figure PCTCN2022130632-appb-000050
对应的第一子图模型。
例如，假设终端设备根据收发天线配置信息确定网络设备的N t个发射天线构成N 1列、N 2行的天线阵列，那么终端设备可以确定
Figure PCTCN2022130632-appb-000051
也即是说，
Figure PCTCN2022130632-appb-000052
在角度域可以包括N 1×N 2个发射角度。其中，N 1×N 2个发射角度中的第j个角度在时域维度上可以包括一 个向量(记为Xj，j为正整数，j≤N t)，Xj包括
Figure PCTCN2022130632-appb-000053
个元素，也即是Xj长度为
Figure PCTCN2022130632-appb-000054
Xj可以表示第j个发射角度，与第i个接收角度之间的信道状态信息，所有的Xj即可表示N t个发射角度与第i个接收角度之间的信道状态信息。
请参照图10中的①，图10为本申请实施例提供的一种第一子图模型的示意图。终端设备可以将这N 1×N 2个角度确定为第一子图模型中的节点，也即是第一子图模型中包括N 1×N 2个节点，每个节点的特征为该节点对应的发射角度在时域维度上包括的向量(即Xj)。终端设备还可以根据
Figure PCTCN2022130632-appb-000055
确定第一子图模型中的边。例如，终端设备可以根据
Figure PCTCN2022130632-appb-000056
在角度域包括的N 1×N 2个角度对应的发射角度，确定第一子图模型中相邻的发送角度对应的两个节点之间存在一条边。换言之，每条边可以表示与边相连的两个节点对应两个相邻的发射角度。
为了便于设备计算，减少设备的数据处理量，可选地，终端设备可以根据
Figure PCTCN2022130632-appb-000057
在角度域包括N 1×N 2个角度之间的相邻关系，确定第一子图模型中相邻节点之间存在边，不相邻节点之间不存在边。换言之，每条边可以表示与边相连的两个节点在第一子图模型中为相邻关系。由于不相邻的角度之间通常相关性较小，因此，第一子图模型中相邻的节点之间存在一条边的实现方式不仅能够减少设备的数据处理量，也能够保证信道状态信息的有效性。
可选地，终端设备可以根据
Figure PCTCN2022130632-appb-000058
在角度域包括的N 1×N 2个角度对应的发射角度，确定第一子图模型中对应同一个接收角度的两个节点之间存在一条边。
其中，Xj为
Figure PCTCN2022130632-appb-000059
维的复数向量，Xj也可以记为
Figure PCTCN2022130632-appb-000060
表示第j个发射角度在时域维度上的所有元素，
Figure PCTCN2022130632-appb-000061
也可以简单表示为H[:,j]。为了便于设备计算，减少设备的数据处理量，可选地，可以将H[:,j]的虚部和实部级联成维度
Figure PCTCN2022130632-appb-000062
的向量，也即是将H[:,j]的虚部和实部分开处理。
在一些可能的实施例中，第一子图模型可以通过一个邻接矩阵和一个节点特征矩阵表示。以上述的终端设备将N 1×N 2个角度确定为第一子图模型中的节点举例，终端设备可以通过N 1×N 2行、N 1×N 2列的邻接矩阵
Figure PCTCN2022130632-appb-000063
表示第一子图模型中节点与节点之间的连接关系。其中，
Figure PCTCN2022130632-appb-000064
A的第i行、第j列元素(记为A i,j)为1可以表示节点i和节点j之间存在边，A i,j＝0可以表示节点i和节点j之间不存在边。终端设备还可以通过节点特征矩阵B(B＝[X1 X2 … Xj … XN t])表示N 1×N 2个节点各自的特征。
请参照图11，图11为本申请实施例提供的一种收发天线连接示意图，网络设备的发射天线包括T1、T2、T3(也即N t＝3)，终端设备的接收天线包括R1、R2(也即N r＝2)。下面结合图11进一步说明上述方式4。
基于图11，终端设备采用上述方式1可以获取
Figure PCTCN2022130632-appb-000065
(i＝1或2)，
Figure PCTCN2022130632-appb-000066
对应R1，
Figure PCTCN2022130632-appb-000067
对应R2。然后，对于每个
Figure PCTCN2022130632-appb-000068
终端设备可以根据收发天线配置信息(比如图11)，确定该
Figure PCTCN2022130632-appb-000069
对应的第一子图模型，具体实现方式可以参照上述方式4中的相关说明。
其中，结合上述方式4中的相关说明，对于
Figure PCTCN2022130632-appb-000070
终端设备可以确定出如图12中的①所示的第一子图模型；对于
Figure PCTCN2022130632-appb-000071
终端设备可以确定出如图12中的②所示的第一子图模型。
可以看出，
Figure PCTCN2022130632-appb-000072
对应的第一子图模型可以包括3个节点(包括节点1～节点3)和2条边(包括边1和边2)。节点1～节点3均对应接收角度1，节点1还对应发射角度1，节点2还对应发 射角度2，节点3还对应发射角度3，边1连接节点1、节点2，边2连接节点2、节点3。
Figure PCTCN2022130632-appb-000073
对应的第一子图模型与
Figure PCTCN2022130632-appb-000074
对应的第一子图模型类似，在此不再对图12中的②进行赘述。
对于上述
Figure PCTCN2022130632-appb-000075
对应的第一子图模型，可以通过如下邻接矩阵A和节点特征矩阵B表示：
Figure PCTCN2022130632-appb-000076
B＝[X1 X2 X3]
方式5，针对每个发射角度，确定一个对应的第二子图模型，所有的第二子图模型即为第一图模型。
具体来说，以针对N t个发射角度中的第j个发射角度确定第j个发射角度对应的第二子图模型举例：终端设备可以采用上述方式2获取与第j个发射角度对应的
Figure PCTCN2022130632-appb-000077
然后，终端设备根据收发天线配置信息，确定
Figure PCTCN2022130632-appb-000078
对应的第二子图模型。
例如，假设终端设备根据收发天线配置信息确定网络设备的N r个接收天线构成N 1列、N 2行的天线阵列，那么终端设备可以确定
Figure PCTCN2022130632-appb-000079
也即是说，
Figure PCTCN2022130632-appb-000080
在角度域可以包括N 1×N 2个角度。其中，N 1×N 2个角度中的第i个角度在时域维度上可以包括一个向量(记为Ui，i为正整数，i≤N r)，Ui包括
Figure PCTCN2022130632-appb-000081
个元素，也即是Ui长度为
Figure PCTCN2022130632-appb-000082
Ui可以表示第i个接收角度，与第j个发射角度之间的信道状态信息，所有的Ui即可表示N r个接收角度与第j个发射角度之间的信道状态信息。
请再参照图10中的①，终端设备可以将这N 1×N 2个角度确定为第二子图模型中的节点，也即是第二子图模型中包括N 1×N 2个节点，每个节点的特征为该节点对应的接收角度在时域维度上包括的向量(即Ui)。终端设备还可以根据
Figure PCTCN2022130632-appb-000083
确定第二子图模型中的边。例如，终端设备可以根据
Figure PCTCN2022130632-appb-000084
在角度域包括的N 1×N 2个角度对应的接收角度，确定第二子图模型中相邻的接收角度对应的两个节点之间存在一条边。换言之，每条边可以表示与边相连的两个节点对应两个相邻的接收角度。
为了便于设备计算，减少设备的数据处理量，可选地，终端设备可以根据
Figure PCTCN2022130632-appb-000085
在角度域包括N 1×N 2个角度之间的相邻关系，确定第二子图模型中相邻节点之间存在边，不相邻节点之间不存在边。换言之，每条边可以表示与边相连的两个节点在第二子图模型中为相邻关系。由于不相邻的角度之间通常相关性较小，因此，第二子图模型中相邻的节点之间存在一条边的实现方式不仅能够减少设备的数据处理量，也能够保证信道状态信息的有效性。
可选地，终端设备可以根据
Figure PCTCN2022130632-appb-000086
在角度域包括的N 1×N 2个角度对应的接收角度，确定第二子图模型中对应同一个发射角度的两个节点之间存在一条边。
其中，Ui为
Figure PCTCN2022130632-appb-000087
维的复数向量，Ui也可以记为
Figure PCTCN2022130632-appb-000088
表示第i个接收角度在时域维度上的所有元素，
Figure PCTCN2022130632-appb-000089
也可以简单表示为H[:,i]。为了便于设备计算，减少设备的数据处理量，可选地，可以将H[:,i]的虚部和实部级联成维度
Figure PCTCN2022130632-appb-000090
的向量，也即是将H[:,i]的虚部和实部分开处理。
在一些可能的实施例中，第二子图模型可以通过一个邻接矩阵和一个节点特征矩阵表示。以上述的终端设备将N 1×N 2个角度确定为第二子图模型中的节点举例，终端设备可以通过N 1×N 2行、N 1×N 2列的邻接矩阵
Figure PCTCN2022130632-appb-000091
表示第二子图模型中节点与节点之间的连接关系。其中，
Figure PCTCN2022130632-appb-000092
A的第i行、第j列元素(记为A i,j)为1可以表示节点i和节点j之间存在边，A i,j＝0可以表示节点i和节点j之间不存在边。终端 设备还可以通过节点特征矩阵B(B＝[U1 U2 … Ui … UN r])表示N 1×N 2个节点各自的特征。
方式5的实施过程的详细说明可以参照上述方式4中结合图11、图12的相关说明，在此不再赘述。
方式6，针对所有的发射角度和所有的接收角度，确定对应的第一图模型。
具体来说，终端设备可以采用上述方式3获取
Figure PCTCN2022130632-appb-000093
然后根据收发天线配置信息，确定
Figure PCTCN2022130632-appb-000094
对应的第一图模型。
例如，假设终端设备根据收发天线配置信息确定：网络设备包括N t个发射天线、终端设备包括N r个接收天线，那么终端设备可以确定
Figure PCTCN2022130632-appb-000095
也即是说，
Figure PCTCN2022130632-appb-000096
在角度域可以包括N t×N r个角度对。其中，如果N t×N r个角度对以矩阵表示，也即N t×N r个角度对包括N r行、N t列角度对，那么N t×N r个角度对中的第i行、第j列角度对(记为G(i，j))在时域维度上可以包括一个向量(记为V(i，j)，i和j均为正整数，i≤N r，j≤N t)，V(i，j)包括
Figure PCTCN2022130632-appb-000097
个元素，也即是V(i，j)长度为
Figure PCTCN2022130632-appb-000098
G(i，j)与第j个发射角度对应，且与第i个接收角度对应。V(i，j)可以表示第j个发射角度，与第i个接收角度之间的信道状态信息，所有的V(i，j)即可表示N t个发射角度与N r个接收角度之间的信道状态信息。
请参照图10中的②，终端设备可以将这N t×N r个角度对确定为第一图模型中的节点，也即是第一图模型中包括N t×N r个节点，每个节点的特征为该节点对应的角度对在时域维度上包括的向量(即V(i，j))。终端设备还可以根据
Figure PCTCN2022130632-appb-000099
确定第一子图模型中的边。例如，终端设备可以根据
Figure PCTCN2022130632-appb-000100
在角度域包括的角度对对应的发射角度和接收角度，确定第一图模型中对应相邻的接收角度或对应相邻的发射角度的两个节点之间存在一条边。换言之，每条边可以表示与边相连的两个节点对应两个相邻的接收角度或对应两个相邻的发射角度。
为了便于设备计算，减少设备的数据处理量，可选地，终端设备可以根据
Figure PCTCN2022130632-appb-000101
在角度域包括的角度之间的相邻关系，确定第一图模型中相邻节点之间存在边，不相邻节点之间不存在边。换言之，每条边可以表示与边相连的两个节点在第一图模型中为相邻关系。由于不相邻的角度之间通常相关性较小，因此，第一图模型中相邻的节点之间存在一条边的实现方式不仅能够减少设备的数据处理量，也能够保证信道状态信息的有效性。
可选地，终端设备可以根据
Figure PCTCN2022130632-appb-000102
在角度域包括的角度对应的发射角度和接收角度，确定第二子图模型中对应同一个发射角度或对应同一个接收角度的两个节点之间存在一条边。
其中，V(i，j)为
Figure PCTCN2022130632-appb-000103
维的复数向量，V(i，j)也可以记为
Figure PCTCN2022130632-appb-000104
表示角度对在时域维度上的所有元素，
Figure PCTCN2022130632-appb-000105
也可以简单表示为H[:,i,j]。为了便于设备计算，减少设备的数据处理量，可选地，可以将H[:,i,j]的虚部和实部级联成维度
Figure PCTCN2022130632-appb-000106
的向量，也即是将H[:,i,j]的虚部和实部分开处理。
在一些可能的实施例中，第一图模型可以通过一个邻接矩阵和一个节点特征矩阵表示。以上述的终端设备将N t×N r个角度对确定为第一图模型中的节点举例，终端设备可以通过N t×N r行、N t×N r列的邻接矩阵
Figure PCTCN2022130632-appb-000107
表示第一图模型中节点与节点之间的连接关系。其中，
Figure PCTCN2022130632-appb-000108
A的第i行、第j列元素(记为A i,j)为1可以表示节点i和节点j之间存在边，A i,j＝0可以表示节点i和节点j之间不存在边。终端 设备还可以通过节点特征矩阵
Figure PCTCN2022130632-appb-000109
表示节点各自的特征。其中，B的第i行、第j列元素为V(i，j)。
下面结合图11进一步说明上述方式6。
基于图11，终端设备采用上述方式3可以获取
Figure PCTCN2022130632-appb-000110
然后根据收发天线配置信息(比如图11)，确定该
Figure PCTCN2022130632-appb-000111
对应的第一图模型，具体实现方式可以参照上述方式6中的相关说明。
其中，结合上述方式6中的相关说明，对于
Figure PCTCN2022130632-appb-000112
终端设备可以确定出如图13所示的第一图模型。可以看出，
Figure PCTCN2022130632-appb-000113
对应的第一图模型可以包括6个节点(包括节点1～节点6)和7条边(包括边1～边7)。各个节点对应的发射角度和接收角度，以及节点之间的连接关系可以参照图13所示，在此不再赘述。其中，为了减少设备的数据处理量，图13中的不相邻的两个节点之间不存在边。
对于上述
Figure PCTCN2022130632-appb-000114
对应的第一图模型，可以通过如下邻接矩阵A和节点特征矩阵B表示：
Figure PCTCN2022130632-appb-000115
在上述方式4～方式6中，每个节点的特征包括的信道状态信息在时域上可以包括
Figure PCTCN2022130632-appb-000116
个元素。在一些可能的实施例中，每个节点的特征包括的信道状态信息在时域上可以包括C个元素，C小于或等于子载波数，也即
Figure PCTCN2022130632-appb-000117
换言之，本申请实施例中，对于每个节点的特征包括的信道状态信息在时域上包括的元素，可以全部保留，也可以舍弃一部分并将剩余部分作为每个节点的特征。
例如，在上述方式4中，对于得到的
Figure PCTCN2022130632-appb-000118
在时域上有
Figure PCTCN2022130632-appb-000119
个元素，可以保留
Figure PCTCN2022130632-appb-000120
在时域上的前N c个元素，得到截断的
Figure PCTCN2022130632-appb-000121
(可以记为Hi，
Figure PCTCN2022130632-appb-000122
)，然后，终端设备可以根据收发天线配置信息，确定Hi对应的第一子图模型，具体方式参照上述方式4，在此不再赘述。这种方式可以记为在时域维度对时延-角度域信道矩阵进行截断，因此
Figure PCTCN2022130632-appb-000123
还可以表示时域维度对时延-角度域信道矩阵在截断前的时域维度的长度，N c可以表示时域维度对时延-角度域信道矩阵在截断后的时域维度的长度，在此统一说明，下文不再赘述。其中，在此情况下，如果将H[:,i,j]的虚部和实部分开处理，那么上述方式4中的Xj的维数为2N c。
其中，由于信道状态信息在时域上存在稀疏性，因此适当减少每个节点的特征包括的信道状态信息在时域上包括的元素数量，不会影响CSI的准确性。换言之，当C小于子载波数时，可以在不影响第二通信装置准确恢复第一信道状态信息的基础上，减少通信装置的数据处理量，提升通信装置的处理速率。
S902，终端设备利用第一神经网络处理第一图模型，得到第一信息。
其中，第一信息用于网络设备恢复第一信道状态信息。例如，第一信息可以包括经过第一神经网络处理后的第一信道状态信息，这样，网络设备可以根据第一信息恢复出第一信道状态信息。
第一神经网络在处理第一图模型的过程中，可以对第一图模型中包含的特征信息进行 压缩。在第一神经网络对第一图模型中包含的特征信息进行压缩的过程中，可以保留其中重要的特征信息，并丢弃(或省略)其中不重要的特征信息。由于第一图模型中包含的特征信息中不重要的特征信息可能会被丢弃，为了提升网络设备根据第一信息恢复第一信道状态信息的准确性，在一些可能的实施例中，第一信息可以包括辅助信息，辅助信息可以用于确定第二图模型，第二图模型可以用于恢复第一信道状态信息。其中，第一信息可以包括第二信道状态信息和的辅助信息，辅助信息用于确定第二图模型是指：辅助信息结合第二信道状态信息可以恢复第二图模型。这样，网络设备可以利用辅助信息确定第二图模型，并根据第二图模型恢复第一信道状态信息。换言之，辅助信息可以帮助第二通信装置恢复第一信道状态信息，提升第二通信装置所恢复的第一信道状态信息的准确性。
可选地，辅助信息可以包括第一图模型的部分节点的索引(index)信息。换言之，辅助信息可以用于指示第一图模型中的部分特征信息(比如被丢弃或减少的特征信息)。这样，终端设备可以利用辅助信息指示第一图模型中因压缩而被丢弃的部分节点的索引，从而能够帮助网络设备恢复第一信道状态信息，提升网络设备所恢复的第一信道状态信息的准确性。
例如，辅助信息可以包括第一图模型中被丢弃的节点的索引信息，或者辅助信息可以包括第一图模型中保留的节点的索引信息。如此，可以进一步帮助网络设备恢复第一信道状态信息，提升网络设备所恢复的第一信道状态信息的准确性。
进一步地，辅助信息还可以包括部分节点的特征的均值。例如，辅助信息还可以包括第一图模型在处理过程中被丢弃的节点的特征的均值。如此，可以进一步帮助网络设备恢复第一信道状态信息，提升网络设备所恢复的第一信道状态信息的准确性。
其中，节点的索引信息也可以被称为节点的序号，节点的标识或节点的位置信息等，在此不予限定。
关于上述辅助信息的具体实现方式可以参照下文中图17所示示例的相关说明，在此不予赘述。
下面介绍S902中，利用第一神经网络处理第一图模型，得到第一信息的实施方式。
在一些可能的实施例中，如图14所示，第一神经网络可以用于根据第一图模型确定第一信息，第一信息可以包括第二信道状态信息和上述的辅助信息，第二信道状态信息可以包括处理后的第一信道状态信息。
示例性地，请参照图15，图15为本申请实施例提供的神经网络的结构示意图二，第一神经网络可以包括第一图池化层和压缩层。
其中，第一图池化层可以用于根据第一图模型确定第三图模型和辅助信息，压缩层可以用于根据第三图模型确定第二信道状态信息。
第一图池化层的输入可以是不同结构和/或不同规模的图模型，也即是说，对于天线配置不同的终端设备、网络设备(以下统称为收发CSI的两端)，终端设备均可以确定第一信道状态信息对应的第一图模型，并将该第一图模型输入同样的第一神经网络中。具体来说，第一图池化层可以保留第一图模型中固定数目的重要的节点，丢弃第一图模型中不重要的节点，使得确定的第三图模型的结构(或者称为尺寸、大小)是固定的，也即第三图模型与输入的第一图模型的结构和/或规模无关。这样，对于不同结构和/或不同规模的第一图模型，均可以用同样的第一图池化层和压缩层进行处理。又由于收发CSI的两端的天 线配置不同会导致第一图模型的结构不同，因此，对于天线配置不同的收发CSI的两端，均可以用同样的第一图池化层和压缩层对第一信道状态信息进行处理。换言之，第一神经网络的设计与训练不依赖于收发CSI的两端的天线配置。这样，在收发CSI的两端的天线配置发生变化时，可以避免重新设计、训练神经网络，适用于动态的MIMO系统和天线配置不同的MIMO系统。
可选地，第一图池化层可以为采用自注意力机制的图池化层。由于自注意机制的图池化层需要训练的参数较少，并能够充分提取第一图模型的节点的特征信息和图的拓扑结构信息，从而“第一图池化层为采用自注意力机制的图池化层”可以提升神经网络的训练效率和处理能力。
在另一些可能的实施例中，请参照图16，图16为本申请实施例提供的神经网络的结构示意图三，第一神经网络可以包括第一图卷积层、第一图池化层和压缩层。
其中，第一图卷积层用于根据第一图模型确定卷积后的第一图模型，第一图池化层用于根据卷积后的第一图模型确定第三图模型和辅助信息，压缩层用于根据第三图模型确定第二信道状态信息。
可以看出，图15所示的第一神经网络与图16所示的第一神经网络的区别在于是否包括第一图卷积层，因此，关于第一图池化层和压缩层的说明可以参照上述图15所示的第一神经网络中的相关说明，在此不再赘述。可选地，第一图卷积层可以包括多个图卷积层。其中，第一图卷积层可以是图卷积网络(graph convolution network，GCN)或图形样本和聚合(graph sample and aggregate，GraphSage)图卷积层等任意类型的图卷积层。
其中，第一图卷积层可以提取并保留第一图模型中重要的节点的特征，省略不重要的节点的特征，因此，在图16所示的第一神经网络中，第一图卷积层能够实现初步的信道信息压缩。
基于与图15所示的第一神经网络相同的理由，图16所示的第一神经网络的设计与训练不依赖于收发CSI的两端的天线配置。这样，在收发CSI的两端的天线配置发生变化时，可以避免重新设计、训练神经网络，适用于动态的MIMO系统和天线配置不同的MIMO系统。
示例性地，上述图15所示的第一神经网络与图16所示的第一神经网络中的压缩层，可以将第三图模型压缩为一定长度的数据，也即是第二信道状态信息。其中，第二信道状态信息中可以包括压缩后的第一信道状态信息。
可选地，上述第一神经网路中的压缩层可以为全连接层。
关于第一神经网络的具体实现方式可以参照下文中图17所示示例的相关说明，在此不予赘述。
S903，终端设备向网络设备发送第一信息。相应地，网络设备接收来自终端设备的第一信息。
其中，第一信息可以承载于上行信道中，比如物理层上行控制信道(physical uplink control channel，PUCCH)、物理上行共享信道(physical uplink shared channel，PUSCH)、其他物理上行信道等，比如未来可能定义的可以用于承载所述第一信息的物理信道，本申请实施例对此不作限定。
S904，网络设备利用第二神经网络处理第一信息，得到第二图模型。
可选地，第一信息中可以包括辅助信息和第二信道状态信息。关于辅助信息和第二信 道状态信息的相关说明可以参照上述S902中的相关说明，在此不再赘述。
可选地，请再参照图14所示，第二神经网络可以用于根据第二信道状态信息和辅助信息，确定第二图模型。
示例性地，请参照图15或图16，第二神经网络可以包括解压层和第二图卷积层。
其中，解压层可以用于根据第二信道状态信息确定第四图模型。示例性地，解压层可以将第二信道状态信息解压为一定长度的数据，用于确定第四图模型。可选地，解压层可以为全连接层。
第二图卷积层可以用于根据辅助信息和第四图模型确定第二图模型。具体地，第二图卷积层用于根据辅助信息和第四图模型确定第二图模型，可以包括：网络设备根据辅助信息恢复第四图模型中缺失的节点，得到第五图模型；然后，将第五图模型输入第二图卷积层，第二图卷积层用于根据第五图模型确定第二图模型。例如，第二图卷积层可以用于对第五图模型进行卷积，得到第二图模型。可以理解，第五图模型可以认为是补全丢弃部分节点的第四图模型。
由于辅助信息还可以包括第一图模型在处理过程中被丢弃的节点的特征的均值，换言之，辅助信息还可以包括第一图模型中被丢弃的节点的均值，第一图模型的说明参照上文，在此不再赘述。
可选地，第二图卷积层可以包括多个图卷积层。
可选地，第二图卷积层包括的多个图卷积层之间可以包括直连通道。这样，可以提高神经网络的训练效率和稳定性。
应理解，基于与图15所示的第一神经网络相同的理由，第二神经网络的设计与训练也不依赖于收发CSI的两端的天线配置。这样，在收发CSI的两端的天线配置发生变化时，可以避免重新设计、训练神经网络，适用于动态的MIMO系统和天线配置不同的MIMO系统。
关于第二神经网络的具体实现方式可以参照下文中图17所示示例的相关说明，在此不予赘述。
S905，网络设备根据第二图模型确定第三信道状态信息。
其中，第三信道状态信息为恢复的第一信道状态信息。换言之，第三信道状态信息包括网络设备和终端设备之间的信道状态信息。第三信道状态信息表示的是网络设备到终端设备之间的CSI。
一种可能的实现中，网络设备根据第二图模型确定第三信道状态信息，可以包括：网络设备根据第二图模型和收发天线配置信息确定第三信道状态。其中，收发天线配置信息的相关说明可以参照上述S901～S902中的相关说明，在此不再赘述。
应理解，S905可以认为是上述S901的逆过程，从而网络设备根据第二图模型和收发天线配置信息确定第三信道状态的实施方式，可以对应参照上述S901，在此不再赘述。
进一步地，本申请实施例还提供的一种可能的第一神经网络、第二神经网络的结构，具体说明如下。
请参照图17，图17为本申请实施例提供的神经网络的结构示意图四。第一神经网络可以包括2层GraphSage图卷积层(包括第1层GraphSage图卷积层、第2层GraphSage图卷积层)、自注意力机制的图池化层和全连接层(记为第一全连接层)。第二神经网络可以包括全连接层(记为第二全连接层)和4层GraphSage图卷积层(包括第3层GraphSage图卷积层～第6 层GraphSage图卷积层)。第一神经网络和第二神经网络中各个网络层之间的级联关系可以参照图17，在此不再赘述。
下面依次介绍图17所示的两个神经网络的具体实现方式。
1，第1层GraphSage图卷积层、第2层GraphSage图卷积层。
终端设备可以将第一图模型输入第1层GraphSage图卷积层，经过第1层GraphSage图卷积层、第2层GraphSage图卷积层卷积后，得到卷积后的第一图模型。其中，第k层GraphSage图卷积层的计算表达式可以如下：
Figure PCTCN2022130632-appb-000124
其中，k＝1或2；
Figure PCTCN2022130632-appb-000125
是第k层GraphSage图卷积层输出的节点i的点嵌入特征向量(也即是节点i在第k层的输出结果)；
Figure PCTCN2022130632-appb-000126
表示第一次输入图卷积层的数据为节点i的特征，例如上述方式4中的Xj；{W 1,W 2}为神经网络参数(也即是待学习的参数)；N(i)表示节点i的邻点集合，N(i)可以通过邻接矩阵A确定；
Figure PCTCN2022130632-appb-000127
表示第k-1层GraphSage图卷积层输出所有节点中属于N(i)的节点的点嵌入特征向量的平均值。
基于上述表达式，通过第1层GraphSage图卷积层、第2层GraphSage图卷积层卷积，可以学习节点之间的关系，每一个节点属性与其他节点属性的相关性，得到第一图模型中每个节点的点嵌入特征向量。值得注意的是，点嵌入特征向量的维度可以小于原始特征的维度(比如上述S901中的2N c)，因此，第1层GraphSage图卷积层、第2层GraphSage图卷积层可以实现初步的信道信息压缩。另外，上述操作中涉及的参数{W 1,W 2}适用于所有的节点，因此该部分的操作独立于输入的第一图模型的结构和规模，也即独立于天线配置。
2，自注意力机制的图池化层
终端设备可以将卷积后的第一图模型输入到自注意力机制的图池化层，经过自注意力机制的图池化层池化后，获取第三图模型。自注意力机制的图池化层的处理过程可以认为是降采样的过程，即只保留图模型中的部分点及它们之间的连接关系，并且能够充分提取第一图模型的节点的特征信息和图的拓扑结构信息。其中，自注意力机制的图池化层的计算表达式可以如下：
score＝GNN(H (2),A)
其中，A为邻接矩阵；H (2)为矩阵，包括第2层GraphSage图卷积层输出的所有节点的点嵌入特征向量；GNN(H (2),A)表示基于自注意力机制的图卷积操作，也即是计算每个节点的分数，分数可以表示节点的重要程度。
3，图模型处理
对于基于自注意力机制的图池化层获取的第三图模型，终端设备可以根据如下公式对第三图模型进行处理，得到第一向量：
idx＝top s(score)
Figure PCTCN2022130632-appb-000128
Figure PCTCN2022130632-appb-000129
其中，idx＝top s(score)可以表示获取分数最高的s个节点的索引；
Figure PCTCN2022130632-appb-000130
可以表示根据idx获取H (2)中分数最高的s个节点对应的点嵌入特征向量；
Figure PCTCN2022130632-appb-000131
可以表示将分数最高的s个节点对应的点嵌入特征向量重新组成向量y(也即第一向量)。s也可以被称为保留维度，嵌入特征向量的长度也可以被称为嵌入维度，在此统一说明，下文不再赘 述。
根据上述“自注意力机制的图池化层”和“第三图模型处理”的计算表达式，可以看出，“自注意力机制的图池化层”和“第三图模型处理”可以利用GNN根据第2层GraphSage图卷积层输出的所有节点的点嵌入特征向量，计算每个节点的分数，保留得分最高的前s个节点，即得到第三图模型，再将这s个节点的点嵌入特征向量拼接为一个第一向量，用于后续全连接层的压缩。可以看出，第三图模型相比于第一图模型丢弃了部分节点。
网络设备还可以根据idx确定被丢弃的节点的索引信息，并根据被丢弃的节点的索引信息确定辅助信息。例如，假设有32个节点，这些节点的索引依次为1～32，s＝20，并且假设经过上述自注意力机制的图池化层的计算表达式处理后，确定的idx包括1～15、20、22、25、30、31，那么终端设备可以根据该idx确定保留的节点的索引包括1～15、20、22、25、30、31，丢弃的节点的索引包括16～19、21、23、24、26～29、32。在此情况下，终端设备可以将索引1～15、20、22、25、30、31作为辅助信息，以指示网络设备第一图模型中保留的节点，或者将索引16～19、21、23、24、26～29、32，以指示网络设备第一图模型中被丢弃的节点。
可选地，上述辅助信息还可以包括所有被丢弃的节点的特征(即点嵌入特征向量)的均值。例如，假设5个节点中被丢弃的节点包括节点1和节点2，那么被丢弃的节点的特征的均值为节点1的特征与节点2的特征的均值。
其中，上述自注意力机制的图池化层的计算表达式中的s可以通过协议预定义，也可以由网络设备通过信令指示给终端设备，对此不作限定。
4，第一全连接层
终端设备可以将上述基于第三图模型获取的第一向量输入第一全连接层，经过第一全连接层压缩后得到第二信道状态信息。
5，第二全连接层
第二全连接层可以对第二信道状态信息进行解压缩处理，得到第二向量。例如，网络设备可以利用第二全连接层对第二信道状态信息进行解压，并基于解压结果获取第二向量。
6，向量处理
对于基于第二全连接层获取的第二向量，网络设备可以将第二向量转换为第四图模型。例如，第二全连接层可以输出长度为2sN c的实数向量，并将其分为s个长度为2N c的子向量，然后将这s个子向量确定为第四图模型。其中，每个子向量为第四图模型中一个节点的特征。
7，填充处理
对于基于第二全连接层获取的第四图模型，网络设备可以根据辅助信息填充第四图模型中被丢弃的节点，得到第五图模型。
可选地，网络设备可以根据辅助信息和收发天线配置信息恢复第四图模型中被丢弃的节点。具体地，网络设备可以根据辅助信息和收发天线配置信息确定第四图模型中被丢弃的节点的位置(即与其他节点的邻接关系)，然后在这些被丢弃的节点的位置填充零向量，从而恢复第四图模型中被丢弃的节点。零向量的维数与第四图模型中保留的节点的特征的维数一致。这种填充方式可以被称为零填充。
其中，如果辅助信息还包括所有被丢弃的节点的特征的均值，那么网络设备可以在这 些被丢弃的节点的位置填充第一向量，从而恢复第四图模型中被丢弃的节点。第一向量由所有被丢弃的节点的特征的均值确定，或者说，第一向量中的所有元素的大小为所有被丢弃的节点的特征的均值。第一向量的维数与第四图模型中保留的节点的特征的维数一致。这种填充方式可以被称为均值填充。
在一些可能的实施例中，网络设备可以在这些被丢弃的节点的位置填充第二向量，从而恢复第四图模型中被丢弃的节点。第二向量为根据相邻的节点的特征估计的向量，比如通过插值法估计第二向量。第二向量的维数与第四图模型中保留的节点的特征的维数一致。这种填充方式可以被称为估计填充。
8，第3层GraphSage图卷积层～第6层GraphSage图卷积层
网络设备可以将第五图模型输入第3层GraphSage图卷积层，经过第3层GraphSage图卷积层～第6层GraphSage图卷积层卷积后，获得第二图模型。
其中，第3层GraphSage图卷积层～第6层GraphSage图卷积层的计算表达式，可以参照上述第1层GraphSage图卷积层、第2层GraphSage图卷积层的计算表达式，在此不再赘述。
其中，第二神经网络中的第3层GraphSage图卷积层的输入与第4层GraphSage图卷积层的输出之间可以存在第一直连通道，也即是说，可以将输入第3层GraphSage图卷积层的数据和第4层GraphSage图卷积层输出的数据进行加和后再输入第5层GraphSage图卷积层。同样地，第5层GraphSage图卷积层的输入与第6层GraphSage图卷积层的输出之间可以存在第二直连通道，实现原理第一直连通道类似，在此不再赘述。
在实际应用中，图17所示的4层GraphSage图卷积层也可以替换为更少的图卷积层(比如2层GraphSage图卷积层)，或者替换为更多的图卷积层(比如6层GraphSage图卷积层)，对此不作限定。经验证，相较于更少的图卷积层，4层GraphSage图卷积层的性能能够满足要求；相较于更多的图卷积层，4层GraphSage图卷积层在性能满足要求的基础上更容易训练。
在一些可能的实施例中，上述S901～S905所示的信道状态信息处理方法可以应用于单阵面，也可以应用于多阵面。当应用于多阵面时，以终端设备举例，对于多阵面中的每个阵面，终端设备可以使用相同的第一神经网络进行信道状态信息的处理(具体过程可以参照S901～S903)，得到每个阵面对应的第一信息，并向网络设备反馈每个阵面对应的第一信息。当然，终端设备也可以使用第一神经网络同时对多阵面中的所有阵面进行信道状态信息的处理，例如，确定多阵面中的所有阵面对应的第一信道状态信息，并确定该第一信道状态信息对应的第一图模型，然后利用S902～S903得到所有阵面对应的第一信息，并向网络设备反馈所有阵面对应的第一信息。值得注意的是，在上述过程中，即使需要多次复用第一神经网络进行信道状态信息的处理，每次使用时第一神经网络都不需要进行调整和重新训练。当应用于多阵面时，网络设备的应用方式与终端设备的应用方式类似，在此不再赘述。
基于上述图9所述的方法，终端设备利用第一神经网络处理第一图模型的过程中，可以保留第一图模型中重要的特征信息，省略第一图模型中不重要的特征信息，利用第一神经网络将第一信道状态信息进行压缩，减少CSI反馈开销的同时，保证CSI反馈的有效性，使得网络设备恢复的信道状态信息更接近原始信道，从而提升通信性能。
本申请实施例中，还提供了一种对第一神经网络、第二神经网络进行的训练方法，具 体如下：获取包含多个训练样本的训练数据集，其中，每个训练样本包括一个第四信道状态信息，第四信道状态信息为网络设备到终端设备的信道状态信息；将多个第四信道状态信息通过第一神经网络和第二神经网络处理，得到多个第五信道状态信息，所述第五信道状态信息为所述第四信道状态信息经过所述第一神经网络和第二神经网络处理后的信道状态信息。基于多个第四信道状态信息和多个第五信道状态信息确定损失函数。确定损失函数对第一神经网络和第二神经网络的参数的梯度，并基于梯度更新各参数，直至达到收敛条件。例如，可以参照上述图4所示的训练过程，确定损失函数对第一神经网络和第二神经网络的参数的梯度，并基于梯度更新各参数，直至达到收敛条件。
其中，将多个第四信道状态信息通过第一神经网络和第二神经网络处理，得到多个第五信道状态信息的实施过程可以包括：将每个第四信道状态信息作为第一信道状态信息并通过上述S901～S902进行处理，得到每个第四信道状态信息对应的第一信息；每个第四信道状态信息对应的第一信息通过S903处理后(也即是经过信道反馈)，再通过上述S904～S905进行处理，得到每个第四信道状态信息对应的第五信道状态信息。
其中，上述训练过程可以以离线或在线的方式进行，训练过程在此不再赘述。
为了验证本申请实施例所提供的技术方案的有效性，本申请还对上述方法实施例进行了仿真，具体的仿真场景以及仿真结果如下：
仿真场景：利用Deep MIMO数据集，分别产生了100,000(记为训练数据集)、30,000(记为验证数据集)和10,000的数据集(记为测试数据集)用于上述神经网络的训练、验证和测试，这些数据集在训练之间均归一化。通信系统的总带宽为0.015625千兆赫兹(giga hertz，GHz)，
Figure PCTCN2022130632-appb-000132
N c＝32。在仿真的过程中，设定嵌入维度为32，保留维度s＝14。
我们测试图17所示的神经网络的效果。其中，在评估第一信道状态信息(也即是CSI)压缩与恢复的性能时，通过归一化均方误差(normalized mean square error，NMSE)和相关度(ρ)两个指标评估，NMSE和ρ分别定义为：
Figure PCTCN2022130632-appb-000133
Figure PCTCN2022130632-appb-000134
其中，E{x}表示求x的期望，x比如是
Figure PCTCN2022130632-appb-000135
或者是
Figure PCTCN2022130632-appb-000136
H表示原本的信道矩阵(也即是实际的信道矩阵)；
Figure PCTCN2022130632-appb-000137
表示经过压缩反馈后恢复的信道矩阵，也即是经过上述S901～S905恢复的第一信道状态信息；h n为H中的第n列向量；
Figure PCTCN2022130632-appb-000138
为
Figure PCTCN2022130632-appb-000139
中的第n列向量。
仿真测试如下：
测试一，利用N t＝32的数据集来训练上述第一神经网络和第二神经网络，并在N t(网络设备的发射天线个数)不同的数据集中进行测试。结果如表1所示，表1为N t变化时图9所示方法的性能。参照表1可以看出，本申请实施例提供的技术方案在天线数量变大时，恢复的准确率只下降了一个百分点左右。因此，本申请实施例提供的技术方案在天线数量变化时有很好的可扩展性。
表1
N t 32 64 128 256 512
NMSE分贝(decibel，dB) -12.28 -12.07 -11.68 -11.25 -10.69
ρ 97.39％ 97.23％ 96.92％ 96.65％ 96.17％
测试二，利用天线配置为(N 1,N 2)＝(32,1)的数据集来训练上述第一神经网络和第二神经网络，并在天线阵列排列方式不同的数据集中进行测试。结果如表2所示，表2为天线阵列排布变化时图9所示方法的性能。参照表2可以看出，即使天线配置改变，相关度仍至少高于94％，因此本申请实施例提供的技术方案在天线阵列排列方式变化时，性能比较稳定。
表2
(N 1,N 2) (32,1) (16,2) (8,4)
NMSE(dB) -12.28 -10.25 -9.69
ρ 97.39％ 95.75％ 94.94％
以上结合图1～图17详细说明了本申请实施例提供的信道状态信息处理方法。以下结合图18详细说明用于执行本申请实施例提供的信道状态信息处理方法的通信装置。
如图18所示，本申请实施例提供了一种通信装置1800。该通信装置1800可以是网络设备或终端设备，也可以是网络设备或终端设备中的装置，或者是能够和网络设备或终端设备匹配使用的装置。一种可能的实现中，该通信装置1800可以包括执行上述方法实施例中网络设备或终端设备执行的方法/操作/步骤/动作所一一对应的模块或单元，该单元可以是硬件电路，也可是软件，也可以是硬件电路结合软件实现。
一种可能的实现中，该通信装置1800可以包括：处理模块1801和收发模块1802。为了便于说明，图18仅示出了该通信装置的主要部件。
在一些实施例中，通信装置1800可适用于图7中所示出的通信系统中，执行图9中所示出的信道状态信息处理方法中终端设备的功能，在此情况下，通信装置1800可以理解为第一通信装置。
其中，处理模块1801，用于确定第一信道状态信息对应的第一图模型，第一信道状态信息为第二通信装置到第一通信装置的信道状态信息。处理模块1801，还用于利用第一神经网络处理第一图模型，得到第一信息，所述第一信息用于第二通信装置恢复第一信道状态信息。收发模块1802，用于向第二通信装置发送第一信息。
一种可能的实现中，处理模块1801，还用于根据收发天线配置信息确定第一信道状态信息对应的第一图模型。其中，收发天线配置信息包括第一通信装置的接收天线配置信息和第二通信装置的发射天线配置信息。
一种可能的实现中，第一信息包括辅助信息，辅助信息用于确定第二图模型，第二图模型用于恢复第一信道状态信息。
可选地，辅助信息包括第一图模型的部分节点的索引信息。
进一步地，辅助信息还可以包括部分节点的特征的均值。
可选地，第一信息包括第二信道状态信息和辅助信息，第二信道状态信息包括处理后的第一信道状态信息。
进一步地，第一神经网络可以包括第一图池化层和压缩层。其中，第一图池化层用于根据第一图模型确定第三图模型和辅助信息，压缩层用于根据第三图模型确定第二信道状 态信息。
其中，第一图池化层可以为采用自注意力机制的图池化层。
进一步地，第一神经网络可以包括第一图卷积层、第一图池化层和压缩层。其中，第一图卷积层用于根据第一图模型确定卷积后的第一图模型，第一图池化层用于根据卷积后的第一图模型确定第三图模型和辅助信息，压缩层用于根据第三图模型确定第二信道状态信息。
进一步地，压缩层可以为全连接层。
一种可能的实现中，第一信道状态信息包括第二通信装置的发射天线与第一通信装置的接收天线之间的时延-角度域信道状态信息。其中，时延-角度域信道状态信息包括R个接收角度和T个发射角度之间的信道状态信息，R和T均为正整数，且R和T中至少一项大于1。第一图模型包括多个节点，节点的特征包括第i个发射角度与第j个接收角度之间的信道状态信息，1≤i≤T，1≤j≤R。第一图模型还包括至少一条边。其中，每条边与两个节点连接，每条边表示与边相连的两个节点对应两个相邻的接收角度或对应两个相邻的发射角度。
可选地，每个节点的特征包括的信道状态信息在时域上包括C个元素，C小于或等于子载波数。
一种可能的实现中，第一通信装置的接收天线配置信息包括如下一项或多项：第一通信装置的接收天线数量、接收天线阵面的类型、接收天线单元的排列方式。第二通信装置的发射天线配置信息包括如下一项或多项：第二通信装置的发射天线数量、发射天线阵面的类型、发射天线单元的排列方式。
一种可能的实现中，处理模块1801，还用于根据第二通信装置的发射天线和第一通信装置的接收天线之间的空间-频率域信道状态信息，确定第一信道状态信息。其中，第一信道状态信息为时延-角度域信道状态信息。
一种可能的实现中，第一神经网络是依据训练数据集确定的，训练数据集包括针对第一神经网络的多个第四信道状态信息，第四信道状态信息为第二通信装置到第一通信装置的信道状态信息。
在另一些实施例中，通信装置1800可适用于图7中所示出的通信系统中，执行图9中所示出的信道状态信息处理方法中网络设备的功能，在此情况下，通信装置1800可以理解为第二通信装置。
其中，收发模块1802，用于接收第一通信装置发送的第一信息，所述第一信息用于第二通信装置恢复第一信道状态信息，第一信道状态信息为第二通信装置到所述第一通信装置的信道状态信息。处理模块1801，用于利用第二神经网络处理第一信息，得到第二图模型。处理模块1801，还用于根据第二图模型确定第三信道状态信息，第三信道状态信息为恢复的第一信道状态信息。
一种可能的实现中，处理模块1801，还用于根据第二图模型和收发天线配置信息确定第三信道状态。其中，收发天线配置信息包括第一通信装置的接收天线配置信息和第二通信装置的发射天线配置信息。
一种可能的实现中，第一信息包括辅助信息，辅助信息用于确定第二图模型。
可选地，辅助信息包括第一图模型的部分节点的索引信息，第一图模型为与第一信道状态信息对应的图模型。
进一步地，辅助信息还可以包括部分节点的特征的均值。
可选地，第一信息包括第二信道状态信息和所述辅助信息，第二信道状态信息包括处理后的第一信道状态信息。
进一步地，第二神经网络可以包括解压层和第二图卷积层。其中，解压层用于根据第二信道状态信息确定第四图模型，第二图卷积层用于根据辅助信息和第四图模型确定第二图模型。
进一步地，第二图卷积层可以包括多个图卷积层，多个图卷积层之间包括直连通道。
进一步地，解压层可以为全连接层。
一种可能的实现中，第一信道状态信息包括第二通信装置的发射天线与第一通信装置的接收天线之间的时延-角度域信道状态信息。其中，时延-角度域信道状态信息包括R个接收角度和T个发射角度之间的信道状态信息，R和T均为正整数，且R和T中至少一项大于1。第二图模型包括多个节点，节点的特征包括第i个发射角度与第j个接收角度之间的信道状态信息，1≤i≤T，1≤j≤R。第二图模型还包括至少一条边。其中，每条边与两个节点连接，每条边表示与边相连的两个节点对应两个相邻的接收角度或对应两个相邻的发射角度。
可选地，每个节点的特征包括的信道状态信息在时域上包括C个元素，C小于或等于子载波数。
一种可能的实现中，第一通信装置的接收天线配置信息包括如下一项或多项：第一通信装置的接收天线数量、接收天线阵面的类型、接收天线单元的排列方式。第二通信装置的发射天线配置信息包括如下一项或多项：第二通信装置的发射天线数量、发射天线阵面的类型、发射天线单元的排列方式。
一种可能的实现中，第二神经网络是依据训练数据集确定的，训练数据集包括针对第二神经网络的多个第四信道状态信息，第四信道状态信息为第二通信装置到第一通信装置的信道状态信息。
可选地，收发模块1802可以包括接收模块和发送模块(图18中未示出)。其中，发送模块用于实现通信装置1800的发送功能，接收模块用于实现通信装置1800的接收功能。
可选地，通信装置1800还可以包括存储模块(图18中未示出)，该存储模块存储有程序或指令。当处理模块1801执行该程序或指令时，使得通信装置1800可以执行图9中所示出的信道状态信息处理方法中网络设备或终端设备的功能。
应理解，通信装置1800中涉及的处理模块1801可以由处理器或处理器相关电路组件实现，可以为处理器或处理单元；收发模块1802可以由收发器或收发器相关电路组件实现，可以为收发器或收发单元。
此外，通信装置1800的技术效果可以参考图9中所示出的信道状态信息处理方法的技术效果，此处不再赘述。
如果本申请实施例提供的通信装置1800是芯片，那么通信装置1800中的收发模块1802可以分别对应芯片输入输出，比如，收发模块1802中的接收模块对应芯片的输入，收发模块1802中的发送模块对应芯片的输出，本申请对此不作限定。
本申请实施例还提供一种芯片系统，包括：处理器，所述处理器与存储器耦合，所述存储器用于存储程序或指令，当所述程序或指令被所述处理器执行时，使得该芯片系统实现上述任一方法实施例中的方法。
可选地，该芯片系统中的处理器可以为一个或多个。该处理器可以通过硬件实现也可以通过软件实现。当通过硬件实现时，该处理器可以是逻辑电路、集成电路等。当通过软件实现时，该处理器可以是一个通用处理器，通过读取存储器中存储的软件代码来实现。
可选地，该芯片系统中的存储器也可以为一个或多个。该存储器可以与处理器集成在一起，也可以和处理器分离设置，本申请并不限定。示例性的，存储器可以是非瞬时性处理器，例如只读存储器ROM，其可以与处理器集成在同一块芯片上，也可以分别设置在不同的芯片上，本申请对存储器的类型，以及存储器与处理器的设置方式不作具体限定。
示例性的，该芯片系统可以是现场可编程门阵列(field programmable gate array，FPGA)，可以是ASIC，还可以是系统芯片(system on chip，SoC)，还可以是CPU，还可以是网络处理器(network processor，NP)，还可以是数字信号处理器(digital signal processor，DSP)，还可以是微控制器(micro controller unit，MCU)，还可以是可编程控制器(programmable logic device，PLD)或其他集成芯片。
本申请实施例提供一种通信系统。该通信系统包括网络设备、终端设备。该网络设备、终端设备结合起来可以执行上述方法实施例，具体执行过程可以参照上述方法实施例，在此不再赘述。
本申请还提供了一种计算机可读存储介质，其上存储有计算机程序，该计算机可读存储介质被计算机执行时实现上述任一方法实施例的功能。
本申请还提供了一种计算机程序产品，该计算机程序产品被计算机执行时实现上述任一方法实施例的功能。
应理解，在本申请实施例中的处理器可以是CPU，该处理器还可以是其他通用处理器、DSP、ASIC、FPGA或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
还应理解，本申请实施例中的存储器可以是易失性存储器或非易失性存储器，或可包括易失性和非易失性存储器两者。其中，非易失性存储器可以是ROM、可编程只读存储器(programmable ROM，PROM)、可擦除可编程只读存储器(erasable PROM，EPROM)，EEPROM或闪存。易失性存储器可以是RAM，其用作外部高速缓存。通过示例性但不是限制性说明，许多形式的RAM可用，例如静态随机存取存储器(static RAM，SRAM)、动态随机存取存储器(DRAM)、同步动态随机存取存储器(synchronous DRAM，SDRAM)、双倍数据速率同步动态随机存取存储器(double data rate SDRAM，DDR SDRAM)、增强型同步动态随机存取存储器(enhanced SDRAM，ESDRAM)、同步连接动态随机存取存储器(synchlink DRAM，SLDRAM)和直接内存总线随机存取存储器(direct rambus RAM，DR RAM)。
上述实施例，可以全部或部分地通过软件、硬件(如电路)、固件或其他任意组合来实现。当使用软件实现时，上述实施例可以全部或部分地以计算机程序产品的形式实现。所述计算机程序产品包括一个或多个计算机指令或计算机程序。在计算机上加载或执行所述计算机指令或计算机程序时，全部或部分地产生按照本申请实施例所述的流程或功能。所述计算机可以为通用计算机、专用计算机、计算机网络、或者其他可编程装置。所述计算机指令可以存储在计算机可读存储介质中，或者从一个计算机可读存储介质向另一个计算机可读存储介质传输，例如，所述计算机指令可以从一个网站站点、计算机、服务器或 数据中心通过有线(例如红外、无线、微波等)方式向另一个网站站点、计算机、服务器或数据中心进行传输。所述计算机可读存储介质可以是计算机能够存取的任何可用介质或者是包含一个或多个可用介质集合的服务器、数据中心等数据存储设备。所述可用介质可以是磁性介质(例如，软盘、硬盘、磁带)、光介质(例如，DVD)、或者半导体介质。半导体介质可以是固态硬盘。
本申请说明书和权利要求书及上述附图中的术语“第一”、“第二”和“第三”等是用于区别不同对象，而不是用于限定特定顺序。
应理解，本文中术语“和/或”，仅仅是一种描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况，其中A，B可以是单数或者复数。另外，本文中字符“/”，一般表示前后关联对象是一种“或”的关系，但也可能表示的是一种“和/或”的关系，具体可参考前后文进行理解。
本申请中，“至少一个”是指一个或者多个，“多个”是指两个或两个以上。“以下至少一项(个)”或其类似表达，是指的这些项中的任意组合，包括单项(个)或复数项(个)的任意组合。例如，a，b，c中的至少一项(个)，可以表示：a，b，c，a-b，a-c，b-c，或a-b-c，其中a，b，c可以是单个，也可以是多个。
应理解，在本申请的各种实施例中，上述各过程的序号的大小并不意味着执行顺序的先后，各过程的执行顺序应以其功能和内在逻辑确定，而不应对本申请实施例的实施过程构成任何限定。
本领域普通技术人员可以意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本申请的范围。
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的系统、装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
在本申请所提供的几个实施例中，应该理解到，所揭露的系统、装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。
所述功能如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。根据这样的理解，本申请的技术方案可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计 算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、ROM、RAM、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述，仅为本申请的具体实施方式，但本申请的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本申请揭露的技术范围内，可轻易想到变化或替换，都应涵盖在本申请的保护范围之内。因此，本申请的保护范围应以所述权利要求的保护范围为准。

Claims (66)

  1. 一种信道状态信息处理方法，其特征在于，包括：
    第一通信装置确定第一信道状态信息对应的第一图模型；所述第一信道状态信息为第二通信装置到所述第一通信装置的信道状态信息；
    所述第一通信装置利用第一神经网络处理所述第一图模型，得到第一信息；所述第一信息用于所述第二通信装置恢复所述第一信道状态信息；
    所述第一通信装置向所述第二通信装置发送所述第一信息。
  2. 根据权利要求1所述的方法，其特征在于，所述第一通信装置确定第一信道状态信息对应的第一图模型，包括：
    所述第一通信装置根据收发天线配置信息确定所述第一信道状态信息对应的第一图模型；所述收发天线配置信息包括所述第一通信装置的接收天线配置信息和所述第二通信装置的发射天线配置信息。
  3. 根据权利要求1或2所述的方法，其特征在于，所述第一信息包括辅助信息，所述辅助信息用于确定第二图模型，所述第二图模型用于恢复所述第一信道状态信息。
  4. 根据权利要求3所述的方法，其特征在于，所述辅助信息包括所述第一图模型的部分节点的索引信息。
  5. 根据权利要求4所述的方法，其特征在于，所述辅助信息还包括所述部分节点的特征的均值。
  6. 根据权利要求3-5中任一项所述的方法，其特征在于，所述第一信息包括第二信道状态信息和所述辅助信息，所述第二信道状态信息包括处理后的所述第一信道状态信息。
  7. 根据权利要求6所述的方法，其特征在于，所述第一神经网络包括第一图池化层和压缩层；
    所述第一图池化层用于根据所述第一图模型确定第三图模型和所述辅助信息；
    所述压缩层用于根据所述第三图模型确定所述第二信道状态信息。
  8. 根据权利要求6所述的方法，其特征在于，所述第一神经网络包括第一图卷积层、第一图池化层和压缩层；
    所述第一图卷积层用于根据所述第一图模型确定卷积后的所述第一图模型；
    所述第一图池化层用于根据所述卷积后的所述第一图模型确定第三图模型和所述辅助信息；
    所述压缩层用于根据所述第三图模型确定所述第二信道状态信息。
  9. 根据权利要求7或8所述的方法，其特征在于，所述第一图池化层为采用自注意力机制的图池化层。
  10. 根据权利要求7-9中任一项所述的方法，其特征在于，所述压缩层为全连接层。
  11. 根据权利要求1-10中任一项所述的方法，其特征在于，所述第一信道状态信息包括所述第二通信装置的发射天线与所述第一通信装置的接收天线之间的时延-角度域信道状态信息，所述时延-角度域信道状态信息包括R个接收角度和T个发射角度之间的信道状态信息，R和T均为正整数，且R和T中至少一项大于1；
    所述第一图模型包括多个节点，所述节点的特征包括第i个发射角度与第j个接收角度之间的信道状态信息，1≤i≤T，1≤j≤R；
    所述第一图模型还包括至少一条边；其中，每条边与两个节点连接，每条边表示与所述边相连的两个节点对应两个相邻的接收角度或对应两个相邻的发射角度。
  12. 根据权利要求11所述的方法，其特征在于，所述多个节点中的每个节点的特征包括的信道状态信息在时域上包括C个元素，C小于或等于子载波数。
  13. 根据权利要求1-12中任一项所述的方法，其特征在于，所述第一通信装置的接收天线配置信息包括如下一项或多项：所述第一通信装置的接收天线数量、接收天线阵面的类型、接收天线单元的排列方式；
    所述第二通信装置的发射天线配置信息包括如下一项或多项：所述第二通信装置的发射天线数量、发射天线阵面的类型、发射天线单元的排列方式。
  14. 根据权利要求1-13中任一项所述的方法，其特征在于，所述方法还包括：
    所述第一通信装置根据所述第二通信装置的发射天线和所述第一通信装置的接收天线之间的空间-频率域信道状态信息，确定所述第一信道状态信息，所述第一信道状态信息为时延-角度域信道状态信息。
  15. 根据权利要求1-14中任一项所述的方法，其特征在于，所述第一神经网络是依据训练数据集确定的，所述训练数据集包括针对所述第一神经网络的多个第四信道状态信息，所述第四信道状态信息为所述第二通信装置到所述第一通信装置的信道状态信息。
  16. 一种信道状态信息处理方法，其特征在于，包括：
    第二通信装置接收第一通信装置发送的第一信息；所述第一信息用于所述第二通信装置恢复第一信道状态信息，所述第一信道状态信息为所述第二通信装置到所述第一通信装置的信道状态信息；
    所述第二通信装置利用第二神经网络处理所述第一信息，得到第二图模型；
    所述第二通信装置根据所述第二图模型确定第三信道状态信息，所述第三信道状态信息为恢复的所述第一信道状态信息。
  17. 根据权利要求16所述的方法，其特征在于，所述第二通信装置根据所述第二图模型确定所述第三信道状态信息，包括：
    所述第二通信装置根据所述第二图模型和收发天线配置信息确定所述第三信道状态信息；所述收发天线配置信息包括所述第一通信装置的接收天线配置信息和所述第二通信装置的发射天线配置信息。
  18. 根据权利要求16或17所述的方法，其特征在于，所述第一信息包括辅助信息，所述辅助信息用于确定所述第二图模型。
  19. 根据权利要求18所述的方法，其特征在于，所述辅助信息包括第一图模型的部分节点的索引信息，所述第一图模型为与所述第一信道状态信息对应的图模型。
  20. 根据权利要求19所述的方法，其特征在于，所述辅助信息还包括所述部分节点的特征的均值。
  21. 根据权利要求18-20中任一项所述的方法，其特征在于，所述第一信息包括第二信道状态信息和所述辅助信息，所述第二信道状态信息包括处理后的所述第一信道状态信息。
  22. 根据权利要求21所述的方法，其特征在于，所述第二神经网络包括解压层和第二图卷积层；
    所述解压层用于根据所述第二信道状态信息确定第四图模型；
    所述第二图卷积层用于根据所述辅助信息和所述第四图模型确定所述第二图模型。
  23. 根据权利要求22所述的方法，其特征在于，所述第二图卷积层包括多个图卷积层，所述多个图卷积层之间包括直连通道。
  24. 根据权利要求22或23所述的方法，其特征在于，所述解压层为全连接层。
  25. 根据权利要求16-24中任一项所述的方法，其特征在于，所述第一信道状态信息包括所述第二通信装置的发射天线与所述第一通信装置的接收天线之间的时延-角度域信道状态信息，所述时延-角度域信道状态信息包括R个接收角度和T个发射角度之间的信道状态信息，R和T均为正整数，且R和T中至少一项大于1；
    所述第二图模型包括多个节点，所述节点的特征包括第i个发射角度与第j个接收角度之间的信道状态信息，1≤i≤T，1≤j≤R；
    所述第二图模型还包括至少一条边；其中，每条边与两个节点连接，每条边表示与所述边相连的两个节点对应两个相邻的接收角度或对应两个相邻的发射角度。
  26. 根据权利要求25所述的方法，其特征在于，所述多个节点中的每个节点的特征包括的信道状态信息在时域上包括C个元素，C小于或等于子载波数。
  27. 根据权利要求16-26中任一项所述的方法，其特征在于，所述第一通信装置的接收天线配置信息包括如下一项或多项：第一通信装置的接收天线数量、接收天线阵面的类型、接收天线单元的排列方式；
    所述第二通信装置的发射天线配置信息包括如下一项或多项：第二通信装置的发射天线数量、发射天线阵面的类型、发射天线单元的排列方式。
  28. 根据权利要求16-27中任一项所述的方法，其特征在于，所述第二神经网络是依据训练数据集确定的，所述训练数据集包括针对所述第二神经网络的多个第四信道状态信息，所述第四信道状态信息为所述第二通信装置到所述第一通信装置的信道状态信息。
  29. 一种通信装置，其特征在于，所述通信装置包括处理模块和收发模块；
    所述处理模块，用于确定第一信道状态信息对应的第一图模型，所述第一信道状态信息为第二通信装置到第一通信装置的信道状态信息；
    所述处理模块，还用于利用第一神经网络处理所述第一图模型，得到第一信息；所述第一信息用于所述第二通信装置恢复所述第一信道状态信息；
    所述收发模块，用于向所述第二通信装置发送所述第一信息。
  30. 根据权利要求29所述的通信装置，其特征在于，所述处理模块，还用于根据收发天线配置信息确定所述第一信道状态信息对应的第一图模型；所述收发天线配置信息包括所述第一通信装置的接收天线配置信息和所述第二通信装置的发射天线配置信息。
  31. 根据权利要求29或30所述的通信装置，其特征在于，所述第一信息包括辅助信息，所述辅助信息用于确定第二图模型，所述第二图模型用于恢复所述第一信道状态信息。
  32. 根据权利要求31所述的通信装置，其特征在于，所述辅助信息包括所述第一图模型的部分节点的索引信息。
  33. 根据权利要求32所述的通信装置，其特征在于，所述辅助信息还包括所述部分节点的特征的均值。
  34. 根据权利要求31-33中任一项所述的通信装置，其特征在于，所述第一信息包括第 二信道状态信息和所述辅助信息，所述第二信道状态信息包括处理后的所述第一信道状态信息。
  35. 根据权利要求34所述的通信装置，其特征在于，所述第一神经网络包括第一图池化层和压缩层；
    所述第一图池化层用于根据所述第一图模型确定第三图模型和所述辅助信息；
    所述压缩层用于根据所述第三图模型确定所述第二信道状态信息。
  36. 根据权利要求34所述的通信装置，其特征在于，所述第一神经网络包括第一图卷积层、第一图池化层和压缩层；
    所述第一图卷积层用于根据所述第一图模型确定卷积后的所述第一图模型；
    所述第一图池化层用于根据所述卷积后的所述第一图模型确定第三图模型和所述辅助信息；
    所述压缩层用于根据所述第三图模型确定所述第二信道状态信息。
  37. 根据权利要求35或36所述的通信装置，其特征在于，所述第一图池化层为采用自注意力机制的图池化层。
  38. 根据权利要求35-37中任一项所述的通信装置，其特征在于，所述压缩层为全连接层。
  39. 根据权利要求29-38中任一项所述的通信装置，其特征在于，所述第一信道状态信息包括所述第二通信装置的发射天线与所述第一通信装置的接收天线之间的时延-角度域信道状态信息，所述时延-角度域信道状态信息包括R个接收角度和T个发射角度之间的信道状态信息，R和T均为正整数，且R和T中至少一项大于1；
    所述第一图模型包括多个节点，所述节点的特征包括第i个发射角度与第j个接收角度之间的信道状态信息，1≤i≤T，1≤j≤R；
    所述第一图模型还包括至少一条边；其中，每条边与两个节点连接，每条边表示与所述边相连的两个节点对应两个相邻的接收角度或对应两个相邻的发射角度。
  40. 根据权利要求39所述的通信装置，其特征在于，所述多个节点中的每个节点的特征包括的信道状态信息在时域上包括C个元素，C小于或等于子载波数。
  41. 根据权利要求29-40中任一项所述的通信装置，其特征在于，所述第一通信装置的接收天线配置信息包括如下一项或多项：所述第一通信装置的接收天线数量、接收天线阵面的类型、接收天线单元的排列方式；
    所述第二通信装置的发射天线配置信息包括如下一项或多项：所述第二通信装置的发射天线数量、发射天线阵面的类型、发射天线单元的排列方式。
  42. 根据权利要求29-41中任一项所述的通信装置，其特征在于，所述处理模块，还用于根据所述第二通信装置的发射天线和所述第一通信装置的接收天线之间的空间-频率域信道状态信息，确定所述第一信道状态信息，所述第一信道状态信息为时延-角度域信道状态信息。
  43. 根据权利要求29-42中任一项所述的通信装置，其特征在于，所述第一神经网络是依据训练数据集确定的，所述训练数据集包括针对所述第一神经网络的多个第四信道状态信息，所述第四信道状态信息为所述第二通信装置到所述第一通信装置的信道状态信息。
  44. 一种通信装置，其特征在于，所述通信装置包括处理模块和收发模块；
    所述收发模块，用于接收第一通信装置发送的第一信息；所述第一信息用于第二通信装置恢复第一信道状态信息，所述第一信道状态信息为所述第二通信装置到所述第一通信装置的信道状态信息；
    所述处理模块，用于利用第二神经网络处理所述第一信息，得到第二图模型；
    所述处理模块，还用于根据所述第二图模型确定第三信道状态信息，所述第三信道状态信息为恢复的所述第一信道状态信息。
  45. 根据权利要求44所述的通信装置，其特征在于，所述处理模块，还用于根据所述第二图模型和收发天线配置信息确定所述第三信道状态；所述收发天线配置信息包括所述第一通信装置的接收天线配置信息和所述第二通信装置的发射天线配置信息。
  46. 根据权利要求44或45所述的通信装置，其特征在于，所述第一信息包括辅助信息，所述辅助信息用于确定所述第二图模型。
  47. 根据权利要求46所述的通信装置，其特征在于，所述辅助信息包括第一图模型的部分节点的索引信息，所述第一图模型为与所述第一信道状态信息对应的图模型。
  48. 根据权利要求47所述的通信装置，其特征在于，所述辅助信息还包括所述部分节点的特征的均值。
  49. 根据权利要求46-48中任一项所述的通信装置，其特征在于，所述第一信息包括第二信道状态信息和所述辅助信息，所述第二信道状态信息包括处理后的所述第一信道状态信息。
  50. 根据权利要求49所述的通信装置，其特征在于，所述第二神经网络包括解压层和第二图卷积层；
    所述解压层用于根据所述第二信道状态信息确定第四图模型；
    所述第二图卷积层用于根据所述辅助信息和所述第四图模型确定第二图模型。
  51. 根据权利要求50所述的通信装置，其特征在于，所述第二图卷积层包括多个图卷积层，所述多个图卷积层之间包括直连通道。
  52. 根据权利要求50或51所述的通信装置，其特征在于，所述解压层为全连接层。
  53. 根据权利要求44-52中任一项所述的通信装置，其特征在于，所述第一信道状态信息包括所述第二通信装置的发射天线与所述第一通信装置的接收天线之间的时延-角度域信道状态信息，所述时延-角度域信道状态信息包括R个接收角度和T个发射角度之间的信道状态信息，R和T均为正整数，且R和T中至少一项大于1；
    所述第二图模型包括多个节点，所述节点的特征包括第i个发射角度与第j个接收角度之间的信道状态信息，1≤i≤T，1≤j≤R；
    所述第二图模型还包括至少一条边；其中，每条边与两个节点连接，每条边表示与所述边相连的两个节点对应两个相邻的接收角度或对应两个相邻的发射角度。
  54. 根据权利要求53所述的通信装置，其特征在于，所述多个节点中的每个节点的特征包括的信道状态信息在时域上包括C个元素，C小于或等于子载波数。
  55. 根据权利要求44-54中任一项所述的通信装置，其特征在于，所述第一通信装置的接收天线配置信息包括如下一项或多项：第一通信装置的接收天线数量、接收天线阵面的类型、接收天线单元的排列方式；
    所述第二通信装置的发射天线配置信息包括如下一项或多项：第二通信装置的发射天线数量、发射天线阵面的类型、发射天线单元的排列方式。
  56. 根据权利要求44-55中任一项所述的通信装置，其特征在于，所述第二神经网络是依据训练数据集确定的，所述训练数据集包括针对所述第二神经网络的多个第四信道状态信息，所述第四信道状态信息为所述第二通信装置到所述第一通信装置的信道状态信息。
  57. 一种通信装置，其特征在于，所述通信装置包括处理器，所述处理器用于执行如权利要求1-28中任一项所述的信道状态信息处理方法。
  58. 根据权利要求57所述的通信装置，其特征在于，所述通信装置还包括收发器，所述收发器用于所述通信装置与其他通信装置通信。
  59. 一种通信装置，其特征在于，包括：与存储器耦合的处理器，所述处理器用于执行所述存储器中的计算机程序或指令，使得权利要求1-28中任一项所述的方法被执行。
  60. 一种通信装置，其特征在于，包括：处理器和接口电路；其中，所述接口电路，用于接收代码指令并传输至所述处理器；
    所述处理器用于运行所述代码指令以执行如权利要求1-28中任一项所述的信道状态信息处理方法。
  61. 一种通信装置，其特征在于，包括：处理器和存储介质；
    所述存储介质存储有指令，所述指令被所述处理器运行时，使得如权利要求1-28中任一项所述的信道状态信息处理方法被执行。
  62. 一种处理器，其特征在于，所述处理器用于执行如权利要求1-28中任一项所述的信道状态信息处理方法。
  63. 一种通信系统，其特征在于，包括：第一通信装置或第二通信装置；所述第一通信装置用于执行如权利要求1-15中任一项所述的信道状态信息处理方法，所述第二通信装置用于执行如权利要求16-28中任一项所述的信道状态信息处理方法。
  64. 一种计算机可读存储介质，其特征在于，包括计算机程序或指令，当所述计算机程序或指令在计算机上运行时，使得如权利要求1-28中任一项所述的方法被执行。
  65. 一种计算机程序产品，其特征在于，包括指令，当所述指令被处理器运行时，使得如权利要求1-28中任一项所述的信道状态信息处理方法被执行。
  66. 一种芯片，其特征在于，包括处理逻辑电路和接口电路；
    所述接口电路，用于接收代码指令并传输至所述处理逻辑电路；
    所述处理逻辑电路用于运行所述代码指令以执行如权利要求1-28中任一项所述的信道状态信息处理方法。
PCT/CN2022/130632 2021-11-18 2022-11-08 信道状态信息处理方法及装置 WO2023088138A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/666,833 US20240322883A1 (en) 2021-11-18 2024-05-17 Channel state information processing method and apparatus

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111370355.6A CN116155330A (zh) 2021-11-18 2021-11-18 信道状态信息处理方法及装置
CN202111370355.6 2021-11-18

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/666,833 Continuation US20240322883A1 (en) 2021-11-18 2024-05-17 Channel state information processing method and apparatus

Publications (1)

Publication Number Publication Date
WO2023088138A1 true WO2023088138A1 (zh) 2023-05-25

Family

ID=86360487

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/130632 WO2023088138A1 (zh) 2021-11-18 2022-11-08 信道状态信息处理方法及装置

Country Status (3)

Country Link
US (1) US20240322883A1 (zh)
CN (1) CN116155330A (zh)
WO (1) WO2023088138A1 (zh)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108833313A (zh) * 2018-07-12 2018-11-16 北京邮电大学 一种基于卷积神经网络的无线信道估计方法及装置
US20210273707A1 (en) * 2020-02-28 2021-09-02 Qualcomm Incorporated Neural network based channel state information feedback
US20210351885A1 (en) * 2019-04-16 2021-11-11 Samsung Electronics Co., Ltd. Method and apparatus for reporting channel state information

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108833313A (zh) * 2018-07-12 2018-11-16 北京邮电大学 一种基于卷积神经网络的无线信道估计方法及装置
US20210351885A1 (en) * 2019-04-16 2021-11-11 Samsung Electronics Co., Ltd. Method and apparatus for reporting channel state information
US20210273707A1 (en) * 2020-02-28 2021-09-02 Qualcomm Incorporated Neural network based channel state information feedback

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
LU, CHAO ET AL.: "MIMO Channel Information Feedback Using Deep Recurrent Network", IEEE COMMUNICATIONS LETTERS, vol. 23, no. 1, 9 November 2019 (2019-11-09), XP011696277, DOI: 10.1109/LCOMM.2018.2882829 *

Also Published As

Publication number Publication date
CN116155330A (zh) 2023-05-23
US20240322883A1 (en) 2024-09-26

Similar Documents

Publication Publication Date Title
WO2021008477A1 (zh) 预编码处理方法和装置
WO2015101109A1 (zh) 一种信道状态信息测量、参考信号的发送方法和装置
CN115053465B (zh) 一种信息传输方法及装置
WO2022267633A1 (zh) 信息传输的方法和装置
WO2022253023A1 (zh) 一种通信方法及装置
WO2022174642A1 (zh) 基于空间划分的数据处理方法和通信装置
WO2020244496A1 (zh) 一种信道测量方法和通信装置
US20240320488A1 (en) Communication method and communication apparatus
WO2022151064A1 (zh) 信息发送方法、信息接收方法、装置、设备及介质
WO2023088138A1 (zh) 信道状态信息处理方法及装置
WO2023126007A1 (zh) 信道信息传输方法及装置
WO2020156095A1 (zh) 译码的方法和译码装置
WO2023006096A1 (zh) 一种通信方法及装置
WO2023115254A1 (zh) 处理数据的方法及装置
WO2023236803A1 (zh) 信道信息的传输方法、通信装置及通信设备
WO2023231881A1 (zh) 一种模型应用方法及装置
WO2023070675A1 (zh) 数据处理的方法及装置
WO2023231934A1 (zh) 一种通信方法及装置
WO2024148566A1 (zh) 数据压缩传输方法、装置、设备以及存储介质
WO2024036631A1 (zh) 一种信息反馈方法、装置、设备及存储介质
WO2024046288A1 (zh) 通信方法和装置
WO2023125996A1 (zh) 一种上行预编码方法及装置
WO2024027424A1 (zh) 一种数据质量度量方法及装置
WO2023279947A1 (zh) 一种通信方法及装置
US20230412227A1 (en) Method and apparatus for transmitting or receiving information for artificial intelligence based channel state information feedback in wireless communication system

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22894675

Country of ref document: EP

Kind code of ref document: A1