WO2024032701A1 - 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
WO2024032701A1
WO2024032701A1 PCT/CN2023/112162 CN2023112162W WO2024032701A1 WO 2024032701 A1 WO2024032701 A1 WO 2024032701A1 CN 2023112162 W CN2023112162 W CN 2023112162W WO 2024032701 A1 WO2024032701 A1 WO 2024032701A1
Authority
WO
WIPO (PCT)
Prior art keywords
state information
channel state
module
subband
input
Prior art date
Legal status (The legal status 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 status listed.)
Ceased
Application number
PCT/CN2023/112162
Other languages
English (en)
French (fr)
Inventor
金立强
黄秋萍
高秋彬
费永强
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Datang Mobile Communications Equipment Co Ltd
Original Assignee
Datang Mobile Communications Equipment Co Ltd
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 Datang Mobile Communications Equipment Co Ltd filed Critical Datang Mobile Communications Equipment Co Ltd
Priority to US19/103,210 priority Critical patent/US20260051989A1/en
Priority to EP23851922.7A priority patent/EP4572171A4/en
Publication of WO2024032701A1 publication Critical patent/WO2024032701A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L5/00—Arrangements affording multiple use of the transmission path
    • H04L5/003—Arrangements for allocating sub-channels of the transmission path
    • H04L5/0048—Allocation of pilot signals, i.e. of signals known to the receiver
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • 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
    • H04B7/0413—MIMO systems
    • H04B7/0452—Multi-user 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
    • 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]
    • 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
    • H04B7/0658—Feedback reduction

Definitions

  • the present disclosure relates to the field of communication technology, and in particular, to a channel state information processing method and device.
  • CSI channel state information
  • MIMO multiple-input multiple-output
  • MIMO multiple-input multiple-output
  • the relevant solution requires training multiple sets of AI models, which will increase the training complexity of the AI model and the storage overhead of the AI model.
  • Embodiments of the present disclosure provide a channel state information processing method and device to solve the problem in the prior art that a large number of AI models need to be deployed for channel state information processing, causing model training and deployment to waste a large amount of resources, and to reduce the cost of AI model training. complexity and save AI model deployment resources.
  • embodiments of the present disclosure provide a channel state information processing method, which is applied to a terminal.
  • the method includes:
  • the encoder model includes an encoder and a channel state information extension enhancement module cascaded with the encoder.
  • the channel state information extension enhancement module includes a downsampling module
  • the down-sampling module is cascaded at the output side of the encoder and is used to compress the feedback overhead of the channel state information to obtain a compressed feedback overhead;
  • the extended enhanced channel state information includes the compressed feedback overhead.
  • the channel state information extension enhancement module includes a pre-conversion module
  • the pre-conversion module is cascaded on the input side of the encoder and used to perform at least one of the following:
  • the extended enhanced channel state information includes channel state information of different MIMO layers and/or different MIMO ports whose probability distribution and dimension are unified.
  • the channel state information extension enhancement module includes a subband filling module
  • the subband filling module is cascaded at the input side of the encoder and is used to fill in the number of subbands of the channel state information to obtain channel state information after subband filling;
  • the extended enhanced channel state information includes the channel state information after subband filling.
  • the downsampling module includes any of the following configurations:
  • the downsampling module is a pooling layer
  • the downsampling module is a fully connected layer whose input dimension is greater than the output dimension
  • the downsampling module is a convolutional downsampling layer.
  • the output dimension of the downsampling module is determined based on the type of the feedback overhead and the predetermined number of quantization bits.
  • the pre-conversion module includes any of the following configurations:
  • the pre-transformation module is an identity transformation layer
  • the pre-transformation module adopts a fully connected layer in the spatial domain and frequency domain respectively;
  • the pre-transformation module uses a 2-dimensional convolution layer in the spatial domain and frequency domain;
  • the pre-transformation module uses a fully connected layer to combine dimensions in the spatial domain and frequency domain;
  • the pre-transformation module combines dimensions in the spatial domain and frequency domain and uses a 1-dimensional convolution layer
  • the pre-transformation module uses a fully connected layer in the spatial domain
  • the pre-transformation module uses a 1-dimensional convolutional layer in the spatial domain.
  • the subband filling module is specifically used to:
  • the sub-bands of the channel state information are processed based on the number of input sub-bands of the encoder model. grouping;
  • the subbands in the target group whose number of subbands of the channel state information is less than the number of input subbands of the encoder model are periodically filled until the number of subbands in the target group is equal to the number of input subbands of the encoder model. Enter the number of subbands;
  • the periodic loop filling includes: copying each subband in sequence according to the arrangement order of each subband in the group.
  • embodiments of the present disclosure also provide a channel state information processing method, which is applied to network equipment.
  • the method includes:
  • the decoder model includes a decoder, and a channel state information restoration module cascaded with the decoder.
  • the expanded enhanced channel state information includes compressed feedback overhead;
  • the channel state information restoration module includes an upsampling module;
  • the upsampling module is cascaded at the input side of the decoder and is used to restore the compressed feedback overhead to obtain restored feedback overhead.
  • the extended enhanced channel state information includes channel state information of different multiple-input multiple-output MIMO layers and/or different MIMO ports with unified probability distribution and dimension;
  • the channel state information restoration module includes a transformation module;
  • the transformation module is cascaded on the output side of the decoder and is used to perform at least one of the following:
  • the probability distribution and dimensionality of the channel state information of different MIMO ports with unified probability distributions and dimensions are restored to obtain original probability distributions and channel state information of different MIMO ports with dimensions.
  • the extended enhanced channel state information includes channel state information after subband filling;
  • the channel state information restoration module includes a subband clipping module;
  • the subband clipping module is cascaded at the output side of the decoder and is used to clip and/or combine the channel state information after subband filling to obtain the channel state information of the original number of subbands.
  • the upsampling module includes any of the following configurations:
  • the upsampling module is a neighboring interpolation layer
  • the upsampling module is a fully connected layer whose input dimension is smaller than the output dimension
  • the upsampling module is a transposed convolutional upsampling layer.
  • the input dimension of the upsampling module is determined based on the type of feedback overhead and the predetermined number of quantization bits.
  • the transformation module includes any of the following configurations:
  • the transformation module is an identity transformation layer
  • the transformation module adopts a fully connected layer in the spatial domain and frequency domain respectively;
  • the transformation module uses a 2-dimensional transposed convolution layer in the spatial domain and frequency domain;
  • the transformation module combines dimensions in the spatial domain and frequency domain and uses a fully connected layer
  • the transformation module combines dimensions in the spatial domain and frequency domain and uses a 1-dimensional convolution layer
  • the transformation module uses a fully connected layer in the spatial domain
  • the transformation module uses a 1-dimensional convolutional layer in the spatial domain.
  • the subband cropping module is specifically used to:
  • embodiments of the present disclosure also provide a terminal, including a memory, a transceiver, and a processor:
  • Memory used to store computer programs
  • transceiver used to send and receive data under the control of the processor
  • processor used to read the computer program in the memory and perform the following operations:
  • the encoder model includes an encoder and a channel state information extension enhancement module cascaded with the encoder.
  • the channel state information extension enhancement module includes a downsampling module
  • the down-sampling module is cascaded at the output side of the encoder and is used to compress the feedback overhead of the channel state information to obtain a compressed feedback overhead;
  • the extended enhanced channel state information includes the compressed feedback overhead.
  • the channel state information extension enhancement module includes a pre-conversion module
  • the pre-conversion module is cascaded on the input side of the encoder and used to perform at least one of the following:
  • the extended enhanced channel state information includes channel state information of different MIMO layers and/or different MIMO ports whose probability distribution and dimension are unified.
  • the channel state information extension enhancement module includes a subband filling module
  • the subband filling module is cascaded at the input side of the encoder and is used to fill in the number of subbands of the channel state information to obtain channel state information after subband filling;
  • the extended enhanced channel state information includes the channel state information after subband filling.
  • the downsampling module includes any of the following configurations:
  • the downsampling module is a pooling layer
  • the downsampling module is a fully connected layer whose input dimension is greater than the output dimension
  • the downsampling module is a convolutional downsampling layer.
  • the output dimension of the downsampling module is based on the type of feedback overhead.
  • the class and the predetermined number of quantization bits are determined.
  • the pre-conversion module includes any of the following configurations:
  • the pre-transformation module is an identity transformation layer
  • the pre-transformation module adopts a fully connected layer in the spatial domain and frequency domain respectively;
  • the pre-transformation module uses a 2-dimensional transposed convolution layer in the spatial domain and frequency domain;
  • the pre-transformation module uses a fully connected layer to combine dimensions in the spatial domain and frequency domain;
  • the pre-transformation module combines dimensions in the spatial domain and frequency domain and uses a 1-dimensional convolution layer
  • the pre-transformation module uses a fully connected layer in the spatial domain
  • the pre-transformation module uses a 1-dimensional convolutional layer in the spatial domain.
  • the subband filling module is specifically used to:
  • the sub-bands of the channel state information are processed based on the number of input sub-bands of the encoder model. grouping;
  • the subbands in the target group whose number of subbands of the channel state information is less than the number of input subbands of the encoder model are periodically filled until the number of subbands in the target group is equal to the number of input subbands of the encoder model. Enter the number of subbands;
  • the periodic loop filling includes: copying each subband in sequence according to the arrangement order of each subband in the group.
  • an embodiment of the present disclosure also provides a network device, including a memory, a transceiver, and a processor:
  • Memory used to store computer programs
  • transceiver used to send and receive data under the control of the processor
  • processor used to read the computer program in the memory and perform the following operations:
  • the decoder model includes a decoder, and a channel state information restoration module cascaded with the decoder.
  • the extended enhanced channel state information includes compressed feedback overhead;
  • the channel state information restoration module includes an upsampling module;
  • the upsampling module is cascaded at the input side of the decoder and is used to restore the compressed feedback overhead to obtain restored feedback overhead.
  • the extended enhanced channel state information includes channel state information of different multiple-input multiple-output MIMO layers and/or different MIMO ports with unified probability distribution and dimension;
  • the channel state information restoration module includes a transformation module;
  • the transformation module is cascaded on the output side of the decoder and is used to perform at least one of the following:
  • the probability distribution and dimensionality of the channel state information of different MIMO ports with unified probability distributions and dimensions are restored to obtain original probability distributions and channel state information of different MIMO ports with dimensions.
  • the extended enhanced channel state information includes channel state information after subband filling;
  • the channel state information restoration module includes a subband clipping module;
  • the subband clipping module is cascaded at the output side of the decoder and is used to clip and/or combine the channel state information after subband filling to obtain the channel state information of the original number of subbands.
  • the upsampling module includes any of the following configurations:
  • the upsampling module is a neighboring interpolation layer
  • the upsampling module is a fully connected layer whose input dimension is smaller than the output dimension
  • the upsampling module is a transposed convolutional upsampling layer.
  • the input dimension of the upsampling module is determined based on the type of feedback overhead and the predetermined number of quantization bits.
  • the transformation module includes any of the following configurations:
  • the transformation module is an identity transformation layer
  • the transformation module adopts a fully connected layer in the spatial domain and frequency domain respectively;
  • the transformation module uses a 2-dimensional transposed convolution layer in the spatial domain and frequency domain;
  • the transformation module combines dimensions in the spatial domain and frequency domain and uses a fully connected layer
  • the transformation module combines dimensions in the spatial domain and frequency domain and uses a 1-dimensional convolution layer
  • the transformation module uses a fully connected layer in the spatial domain
  • the transformation module uses a 1-dimensional convolutional layer in the spatial domain.
  • the subband cropping module is specifically used to:
  • embodiments of the present disclosure also provide a channel state information processing device, which is applied to a terminal.
  • the device includes:
  • An enhancement module configured to input channel state information into the encoder model and obtain extended enhanced channel state information output by the encoder model
  • a sending module configured to send the extended enhanced channel state information to a network device
  • the encoder model includes an encoder and a channel state information extension enhancement module cascaded with the encoder.
  • embodiments of the present disclosure also provide a channel state information processing device, which is applied to network equipment.
  • the device includes:
  • the receiving module is used to receive the extended enhanced channel state information sent by the terminal;
  • a restoration module configured to input the extended enhanced channel state information to a decoder model to obtain restored channel state information output by the decoder model;
  • the decoder model includes a decoder, and a channel state information restoration module cascaded with the decoder.
  • embodiments of the present disclosure also provide a processor-readable storage medium, the processor-readable storage medium stores a computer program, the computer program is used to cause the processor to execute the first aspect as described above. method, or perform the method of the second aspect as described above.
  • the channel state information processing method and device provided by the embodiments of the present disclosure expand and enhance the AI model by cascading the channel state information expansion enhancement module on the input side and/or the output side of the encoder model, which can be implemented under different parameter configurations. Reusing AI models can reduce the complexity of AI model training and save AI model deployment resources.
  • Figure 1 is a schematic structural diagram of an autoencoder in the prior art
  • Figure 2 is a schematic flow chart of an AI-based channel state information compression feedback solution in the prior art
  • Figure 3 is one of the flow diagrams of a channel state information processing method provided by an embodiment of the present disclosure
  • Figure 4 is a schematic structural diagram of a down-sampling module-up-sampling module pair according to an embodiment of the present disclosure
  • Figure 5 is a schematic structural diagram of a pre-conversion module-conversion module pair according to an embodiment of the present disclosure
  • Figure 6 is a schematic structural diagram of a cascaded sub-band filling module and a sub-band clipping module according to an embodiment of the present disclosure
  • Figure 7 is a second schematic flowchart of a channel state information processing method provided by an embodiment of the present disclosure.
  • Figure 8 is one of the structural diagrams of an example of applying the channel state information processing method provided by the embodiment of the present disclosure.
  • Figure 9 is the second structural diagram of an example of applying the channel state information processing method provided by the embodiment of the present disclosure.
  • Figure 10 is the third structural diagram of an example of applying the channel state information processing method provided by the embodiment of the present disclosure.
  • Figure 11 is the fourth structural diagram of an example of applying the channel state information processing method provided by the embodiment of the present disclosure.
  • Figure 12 is the fifth structural diagram of an example of applying the channel state information processing method provided by the embodiment of the present disclosure.
  • Figure 13 is a schematic structural diagram of a terminal provided by an embodiment of the present disclosure.
  • Figure 14 is a schematic structural diagram of a network device provided by an embodiment of the present disclosure.
  • Figure 15 is one of the structural schematic diagrams of a channel state information processing device provided by an embodiment of the present disclosure.
  • Figure 16 is a second structural schematic diagram of a channel state information processing device provided by an embodiment of the present disclosure.
  • the term "and/or” describes the association relationship of associated objects, indicating that there may be Three relationships, for example, A and/or B, can represent: A exists alone, A and B exist simultaneously, and B exists alone.
  • the character "/" generally indicates that the related objects are in an "or” relationship.
  • the term “plurality” refers to two or more than two, and other quantifiers are similar to it.
  • Embodiments of the present disclosure provide a channel state information processing method and device to solve the problem in the prior art that a large number of AI models need to be deployed for channel state information processing, causing model training and deployment to waste a large amount of resources.
  • the method and the device are based on the same application concept. Since the principles of the method and the device to solve the problem are similar, the implementation of the device and the method can be referred to each other, and the repeated details will not be repeated.
  • AI-based neural network models can be used for channel state information feedback.
  • the neural network model can use the encoder module of the Auto Encoder (AE) to compress the channel state information to reduce feedback overhead, and then use the AE decoder The module recovers channel status information.
  • AE Auto Encoder
  • An autoencoder is a neural network with the same input and output. It consists of an encoder and a decoder.
  • the encoder compresses the input data into a latent space representation, and the decoder reconstructs the input based on this representation.
  • the structure of the autoencoder is shown in Figure 1.
  • the input data The purpose is to reconstruct the input data X.
  • the autoencoder solves the mapping relationship between f( ⁇ ) and g( ⁇ ) through backpropagation training to minimize the reconstruction error.
  • the encoder module and decoder module can be constructed using convolutional neural networks, fully connected neural networks, recurrent neural networks, etc. and their combinations.
  • AI-based channel state information compression feedback solutions usually design AI models under given feedback overhead, number of MIMO layers, number of MIMO ports, and number of subbands, as shown in Figure 2.
  • the terminal side sets the channel status
  • the state information V is input to the AI encoder to obtain the compressed codeword C (the information length of the codeword C is much smaller than the length of the input V), then the codeword C is input to the quantizer, and the quantized binary bit stream is input to the inverse quantization
  • the codeword C' with quantization error is obtained by the decoder, and finally C' is input to the decoder module to obtain the restored channel information V'.
  • FIG 3 is one of the schematic flow diagrams of a channel state information processing method provided by an embodiment of the present disclosure.
  • an embodiment of the present disclosure provides a channel state information processing method, the execution subject of which may be a terminal, such as a mobile phone.
  • the method can include:
  • Step 310 Input the channel state information to the encoder model, and obtain the extended and enhanced channel state information output by the encoder model;
  • Step 320 Send the extended enhanced channel state information to the network device
  • the encoder model includes an encoder and a channel state information expansion enhancement module cascaded with the encoder.
  • the AI model that performs CSI compression feedback can be used as the basic AI model, the encoder of the basic AI model is deployed on the terminal to form the encoder model, and the decoder of the basic AI model is deployed on the network device to form the decoder model. ;At the same time, multiple pairs of AI modules are cascaded to the inside of the basic AI model (the output side of the encoder, the input side of the decoder) and/or the outside of the basic AI model (the input side of the encoder, the output side of the decoder) To achieve the expansion and enhancement of the basic AI model, the basic AI model can be reused under different parameter configurations.
  • the terminal can input the channel state information into the encoder model.
  • the input side or output side of the encoder is cascaded with a channel state information expansion enhancement module.
  • the channel state information expansion enhancement module can modify the basic AI model in different ways. Make extended enhancements.
  • extending and enhancing the basic AI model can include compressing the feedback overhead of channel state information, adjusting the probability distribution of channel state information of different MIMO layers and/or different MIMO ports, filling in the number of subbands of channel state information, etc. .
  • the terminal After performing the above processing on the channel state information, the terminal will send the expanded and enhanced channel state information to the network device, so that the decoder model deployed on the network device can process the expanded and enhanced channel state information accordingly, such as compressed feedback.
  • the channel state information processing method provided by the embodiments of the present disclosure expands and enhances the AI model by cascading the channel state information expansion enhancement module on the input side and/or the output side of the encoder model, which can achieve multiplexing under different parameter configurations. AI model, thereby reducing the complexity of AI model training and saving AI model deployment resources.
  • the channel state information extension enhancement module includes a downsampling module
  • the downsampling module is cascaded at the output side of the encoder and is used to compress the feedback overhead of the channel state information to obtain the compressed feedback overhead;
  • Extended enhanced channel state information includes compressed feedback overhead.
  • the network parameter configuration is required to support different feedback overheads.
  • the output of the AI model encoder and the input of the decoder need to support multiple dimensions to achieve different feedback overheads.
  • the output dimension of the encoder (input to the decoder) can be determined, and a basic AI model with a given feedback overhead can be trained.
  • At least a pair of down-sampling modules A-k and up-sampling modules B-k can be cascaded inside the basic AI model (the output side of the encoder and the input side of the decoder) to achieve feedback overhead expansion and enhancement, as shown in Figure 4 Show.
  • the expanded basic AI model has k branches, and each branch corresponds to a feedback overhead configuration.
  • the downsampling module may include any of the following configurations:
  • the downsampling module is the pooling layer
  • the downsampling module is a fully connected layer whose input dimension is larger than the output dimension
  • the downsampling module is a convolutional downsampling layer.
  • the A-k downsampling module and B-k upsampling module should not be designed to be too complex, and any of the following configurations can be used:
  • A-k is a fully connected layer with an input dimension greater than the output dimension
  • B-k is a fully connected layer with an input dimension smaller than the output dimension
  • A-k is a convolutional downsampling layer (a convolutional layer with a sliding step larger than 1)
  • B-k is a transposed convolutional upsampling layer.
  • the basic AI model can be a CSI compression feedback AI model under a given feedback overhead.
  • the Ak downsampling module and the Bk upsampling module further downsample and compress based on the basic AI model to obtain a variety of lower Feedback overhead n payload,k .
  • the basic AI model can also be a set of AI transformation-inverse transformation pairs, without dimensional compression (that is, the output dimension of the encoder of the basic AI model is equal to its input dimension), and the dimensional compression is completely through the A-k downsampling module and B-k upsampling. Module implementation.
  • the channel state information processing method provided by the embodiments of the present disclosure can realize the expansion and enhancement of the AI model for CSI feedback overhead processing by cascading the downsampling module on the output side of the encoder, thereby realizing the multiplexing of the AI model under different feedback overhead configurations. , which can reduce the complexity of AI model training and save AI model deployment resources.
  • the channel state information extension enhancement module includes a pre-transformation module
  • the pre-transformation module is cascaded on the input side of the encoder and used to perform at least one of the following:
  • the expanded enhanced channel state information includes different MIMO layers and/or different dimensions with unified probability distributions and dimensions.
  • the network parameter configuration is required to support different numbers of MIMO layers and ports.
  • the input and output of the basic AI model need to support multiple dimensions.
  • the probability distribution of the basic AI model input under different configurations must also be unified or close to make the basic AI model work at its best. In other words, different Ports
  • the probability distribution of CSI data in different MIMO layers is inconsistent and needs to be transformed into a unified probability distribution.
  • the pre-transformation module includes any of the following configurations:
  • the pre-transformation module is the identity transformation layer
  • the pre-transformation module uses a fully connected layer in the spatial domain and frequency domain respectively;
  • the pre-transformation module uses a 2-dimensional convolutional layer in the spatial and frequency domains
  • the pre-transformation module combines dimensions in the spatial domain and frequency domain and uses a fully connected layer
  • the pre-transformation module combines dimensions in the spatial domain and frequency domain, using a 1-dimensional convolution layer
  • the pre-transformation module uses a fully connected layer in the spatial domain
  • the pre-transformation module uses a 1D convolutional layer in the spatial domain.
  • the A-k pre-transformation/B-k transformation module pair functions as distribution transformation.
  • the A-1/B-1 module is an identity transformation (input equals output)
  • the basic AI model is an AI model suitable for layer 1
  • the CSI probability distribution of k is close to the CSI probability distribution of layer 1
  • B-k transforms the restored CSI of layer k back to the original distribution.
  • the A-k pre-conversion/B-k conversion module pair can include any of the following configurations:
  • Spatial domain-frequency domain two-dimensional linear transformation using a fully connected layer in the spatial domain and frequency domain respectively, or using a 2-dimensional convolution/transposed convolution as a whole;
  • the spatial domain-frequency domain is stretched to 1 dimension, using a fully connected layer or 1-dimensional convolution.
  • the A-k pre-transformation/B-k transformation module pair functions as probability distribution matching and dimension unification, and the CSI inputs of different Port ports are transformed to a unified Port number.
  • the A-1/B-1 module is an identity transformation (input equals output).
  • the A-k pre-conversion/B-k conversion module pair can include any of the following configurations:
  • a fully connected layer is used in the air domain
  • a 1D convolutional layer is used in the spatial domain.
  • the channel state information processing method provided by the embodiment of the present disclosure can realize the expansion and enhancement of the AI model processing of MIMO layers and/or MIMO ports by cascading the pre-conversion module on the input side of the encoder, thereby realizing the processing of different MIMO layers, Reusing AI models under different port configurations can reduce the complexity of AI model training and save AI model deployment resources.
  • the channel state information extension enhancement module includes a subband filling module
  • the subband filling module is cascaded at the input side of the encoder and is used to fill in the number of subbands of the channel state information to obtain the channel state information after subband filling;
  • the extended enhanced channel state information includes channel state information after subband filling.
  • the network parameter configuration is required to support different subband numbers N subband , but the number of subbands N subband, AI input by the basic AI module used for CSI compression feedback is usually given.
  • embodiments of the present disclosure cascade a subband filling module and a subband cropping module outside the basic AI module (the input side of the encoder and the output side of the decoder). As shown in Figure 6.
  • the subband filling module is specifically used to:
  • Periodic loop filling includes: copying each subband in sequence according to the arrangement order of each subband in the group.
  • the subband filling module and the subband clipping module After cascading the subband filling module and the subband clipping module outside the basic AI module (the input side of the encoder and the output side of the decoder), the subband filling module and the subband clipping module use N subband and AI subband as a set of pairs.
  • Input CSI is padded and cropped:
  • CSI can be divided into N subband /N subband,AI groups. Each group is sent to the basic AI network to implement CSI compression feedback, and the subband filling mode Block and subband cropping modules require no additional processing of the input.
  • the subband filling module will include N subband mod N subband,AI subband
  • the members of the group are periodically filled to N subband, and then the AI subband is sent to the basic AI model.
  • the subband clipping module will clip or merge the subbands filled with periodic cycles to obtain the original N subband mod N subband, CSI of the AI subband.
  • This operation can simulate frequency domain correlation to the greatest extent and thereby improve CSI compression feedback performance.
  • the channel state information processing method provided by the embodiment of the present disclosure can realize the expansion and enhancement of the AI model's processing of the number of subbands by cascading the subband filling module on the input side of the encoder, thereby realizing multiplexing under different feedback overhead configurations.
  • AI models can reduce the complexity of AI model training and save AI model deployment resources.
  • FIG. 7 is a second schematic flowchart of a channel state information processing method provided by an embodiment of the present disclosure.
  • an embodiment of the present disclosure provides a channel state information processing method, the execution subject of which may be a network device, such as a base station.
  • the method can include:
  • Step 710 Receive the extended enhanced channel state information sent by the terminal
  • Step 720 Input the extended enhanced channel state information to the decoder model to obtain the restored channel state information output by the decoder model;
  • the decoder model includes a decoder and a channel state information restoration module cascaded with the decoder.
  • the AI model that performs CSI compression feedback can be used as the basic AI model, the encoder of the basic AI model is deployed on the terminal to form the encoder model, and the decoder of the basic AI model is deployed on the network device to form the decoder model. ;At the same time, multiple pairs of AI modules are cascaded to the inside of the basic AI model (the output side of the encoder, the input side of the decoder) and/or the outside of the basic AI model (the input side of the encoder, the output side of the decoder) To achieve the expansion and enhancement of the basic AI model, the basic AI model can be reused under different parameter configurations.
  • the network device may receive the extended enhanced channel state information sent by the terminal.
  • the terminal can input channel state information into the encoder model.
  • the input side or output side of the encoder is cascaded with a channel state information expansion enhancement module.
  • the channel state information expansion enhancement module can modify the AI basic model in different ways. Make extended enhancements.
  • the feedback overhead of the channel state information is compressed, the probability distribution of the channel state information of different MIMO layers and/or different MIMO ports is adjusted, the number of subbands of the channel state information is filled, etc.
  • the terminal After performing the above processing on the channel state information, the terminal will send the extended enhanced channel state information to the network device.
  • the network device After receiving the extended enhanced channel state information, the network device will input the extended enhanced channel state information into the decoder model. , and based on the decoder cascaded channel state information restoration module, the extended enhanced channel state information is restored, such as restoring the compressed feedback overhead, channels of different MIMO layers and/or different MIMO ports with unified probability distributions The state information is restored with probability distribution, and the channel state information after sub-band filling is cropped and/or merged, etc.
  • the channel state information processing method provided by the embodiments of the present disclosure restores the extended and enhanced channel state information by cascading the channel state information restoration module on the input side and/or the output side of the decoder model, which can be implemented under different parameter configurations. Reusing AI models can reduce the complexity of AI model training and save AI model deployment resources.
  • the extended enhanced channel state information includes compressed feedback overhead;
  • the channel state information restoration module includes an upsampling module;
  • the upsampling module is cascaded at the input side of the decoder and is used to restore the compressed feedback overhead to obtain the restored feedback overhead.
  • the network parameter configuration is required to support different feedback overheads.
  • the output of the AI model encoder and the input of the decoder need to support multiple dimensions to achieve different feedback overheads.
  • the output dimension of the encoder (input to the decoder) can be determined, and a basic AI model with a given feedback overhead can be trained.
  • At least a pair of down-sampling modules Ak and up-sampling modules Bk can be cascaded inside the basic AI model (the output side of the encoder and the input side of the decoder) to implement feedback.
  • Overhead expansion is enhanced, as shown in Figure 4.
  • the expanded basic AI model has k branches, and each branch corresponds to a feedback overhead configuration.
  • the upsampling module includes any of the following configurations:
  • the upsampling module is the adjacent interpolation layer
  • the upsampling module is a fully connected layer whose input dimension is smaller than the output dimension
  • the upsampling module is a transposed convolutional upsampling layer.
  • the A-k downsampling module and B-k upsampling module should not be designed to be too complex, and any of the following configurations can be used:
  • A-k is the pooling layer
  • B-k is the adjacent interpolation layer
  • A-k is a fully connected layer with an input dimension greater than the output dimension
  • B-k is a fully connected layer with an input dimension smaller than the output dimension
  • A-k is a convolutional downsampling layer (a convolutional layer with a sliding step larger than 1)
  • B-k is a transposed convolutional upsampling layer.
  • the basic AI model can be a CSI compression feedback AI model under a given feedback overhead.
  • the Ak downsampling module and the Bk upsampling module further downsample and compress based on the basic AI model to obtain a variety of lower Feedback overhead n payload,k .
  • the basic AI model can also be a set of AI transformation-inverse transformation pairs, without dimensional compression (that is, the output dimension of the encoder of the basic AI model is equal to its input dimension), and the dimensional compression is completely through the A-k downsampling module and B-k upsampling. Module implementation.
  • the channel state information processing method provided by the embodiments of the present disclosure can realize the expansion and enhancement of the CSI feedback overhead processing of the AI model by cascading the upsampling module on the input side of the decoder, thereby realizing the multiplexing of the AI model under different feedback overhead configurations. , which can reduce the complexity of AI model training and save AI model deployment resources.
  • the extended enhanced channel state information includes channel state information of different MIMO layers and/or different MIMO ports with uniform probability distribution and dimension;
  • the channel state information restoration module includes a transformation module;
  • the transformation module is cascaded on the output side of the decoder and used to perform at least one of the following:
  • the network parameter configuration is required to support different numbers of MIMO layers and ports.
  • the input and output of the basic AI model need to support multiple dimensions.
  • the probability distribution of the basic AI model input under different configurations must also be unified or close to make the basic AI model work at its best. In other words, the CSI data probability distributions of different Ports and different MIMO layers are inconsistent and need to be transformed into a unified probability distribution.
  • the transformation module includes any of the following configurations:
  • the transformation module is the identity transformation layer
  • the transformation module uses a fully connected layer in the spatial domain and frequency domain respectively;
  • the transformation module uses a 2-dimensional transposed convolution layer in the spatial and frequency domains
  • the transformation module combines dimensions in the spatial domain and frequency domain and uses a fully connected layer
  • the transformation module combines dimensions in the spatial domain and frequency domain, using a 1-dimensional convolutional layer
  • the transformation module uses a fully connected layer in the spatial domain
  • the transformation module uses a 1D convolutional layer in the spatial domain.
  • the A-k pre-transformation/B-k transformation module pair functions as distribution transformation.
  • the A-1/B-1 module is an identity transformation (input equals output)
  • the basic AI model is an AI model suitable for layer 1
  • the CSI probability distribution of k is close to the CSI probability distribution of layer 1
  • B-k transforms the restored CSI of layer k back to the original distribution.
  • the Ak pre-transformation/Bk transformation module pair can include any of the following configurations:
  • Spatial domain-frequency domain two-dimensional linear transformation using a fully connected layer in the spatial domain and frequency domain respectively, or using a 2-dimensional convolution/transposed convolution as a whole;
  • the spatial domain-frequency domain is stretched to 1 dimension, using a fully connected layer or 1-dimensional convolution.
  • the A-k pre-transformation/B-k transformation module pair functions as probability distribution matching and dimension unification, and the CSI inputs of different Port ports are transformed to a unified Port number.
  • the A-1/B-1 module is an identity transformation (input equals output)
  • the basic AI model is an AI model suitable for Port_Max
  • the CSI of (k ⁇ Port_Max) is converted into the CSI of the Port_Max port
  • B-k converts the restored CSI from Port_Max back to the original k port.
  • the A-k pre-conversion/B-k conversion module pair can include any of the following configurations:
  • a fully connected layer is used in the air domain
  • a 1D convolutional layer is used in the spatial domain.
  • the channel state information processing method provided by the embodiments of the present disclosure can realize the expansion and enhancement of the MIMO layer and/or MIMO port processing by the AI model by cascading the pre-conversion module on the output side of the decoder, thereby realizing the processing of different MIMO layers and different MIMO ports.
  • Reusing AI models under port configuration can reduce the complexity of AI model training and save AI model deployment resources.
  • the extended enhanced channel state information includes channel state information after subband filling;
  • the channel state information restoration module includes a subband clipping module;
  • the subband clipping module is cascaded at the output side of the decoder and is used to clip and/or combine the channel state information after subband filling to obtain the channel state information of the original number of subbands.
  • the network parameter configuration is required to support different subband numbers N subband , but the number of subbands N subband, AI input by the basic AI module used for CSI compression feedback is usually given.
  • embodiments of the present disclosure cascade a subband filling module and a subband cropping module outside the basic AI module (the input side of the encoder and the output side of the decoder). As shown in Figure 6.
  • the subband clipping module is specifically used to:
  • subband filling module and subband clipping module fill and clip the input CSI with N subband, AI subband as a group:
  • CSI can be divided into N subband /N subband,AI groups. Each group is sent to the basic AI network to implement CSI compression feedback.
  • the subband filling module and subband clipping module are not Additional processing of input is required.
  • the subband filling module will include N subband mod N subband,AI subband
  • the members of the group are periodically filled to N subband, and then the AI subband is sent to the basic AI model.
  • the subband clipping module will clip or merge the subbands filled with periodic cycles to obtain the original N subband mod N subband, CSI of the AI subband.
  • This operation can simulate frequency domain correlation to the greatest extent and thereby improve CSI compression feedback performance.
  • the channel state information processing method provided by the embodiments of the present disclosure can realize the extension and enhancement of the AI model to the sub-band number processing by cascading the sub-band clipping module on the output side of the decoder, thereby realizing multiplexing under different feedback overhead configurations.
  • AI models can reduce the complexity of AI model training and save AI model deployment resources.
  • EN Block and DE Block are a set of pre-transformation-transformation pairs, using the Transformer model.
  • the input and output dimensions of EN Block and DE Block are 12*64, and the input PMI (precoding vector) is 12 sub-bands 32 Port (the real part and the imaginary part are combined to become 64 dimensions).
  • the trained EN Block transforms the PMI into the implicit domain to facilitate downsampling compression, and the DE Block transforms the data in the implicit domain back to the spatial frequency domain where the PMI is located.
  • the terminal After the terminal receives the network configuration channel (including feedback overhead, the UE can also determine the feedback overhead by itself), it selects the DS-k downsampling branch for CSI compression based on the feedback overhead. Similarly, the network device selects the corresponding US-k upsampling branch to perform CSI recovery based on the feedback overhead configured to the terminal or reported by the terminal.
  • the network configuration channel including feedback overhead, the UE can also determine the feedback overhead by itself
  • the network device selects the corresponding US-k upsampling branch to perform CSI recovery based on the feedback overhead configured to the terminal or reported by the terminal.
  • Example 2 (MIMO layer scalable CSI compression feedback model):
  • the basic AI model is an AE based on the Transformer model, used for PMI compression and decompression of 12 subbands, 32 ports, layer 1, and 240bit feedback overhead.
  • the basic AI model can achieve MIMO multi-layer expansion, among which LPT-1Block , LT-1Block are all identity transformations where the input is equal to the output.
  • the terminal selects the k-th branch for CSI compression, and the network device selects the k-th branch for CSI recovery.
  • Example 3 (MIMO port scalable CSI compression feedback model):
  • the AI basic model is an AE, used for PMI compression and decompression of 12 subbands, 32 ports, layer 1, and 240bit feedback overhead.
  • PPT-k Block/PT-k Block By cascading PPT-k Block/PT-k Block on the outside, the basic AI model realizes MIMO multi-port expansion.
  • PPT-32Block/PT-32Block is the identity transformation where the input is equal to the output.
  • the PMI of AI ⁇ (k ⁇ 2) dimension is transformed into the PMI of N subband, AI ⁇ (32 ⁇ 2) dimension, where k is the actual number of PMI ports.
  • Example 4 (CSI compression feedback model with scalable number of subbands):
  • the basic AI model achieves an expansion of the number of subbands.
  • the last group is less than 12 subbands
  • 8 subbands are cyclically filled in the frequency domain to achieve 12 subband AI inputs (the subband index number after filling is [1,2,3,4,3,2,1,2,3,4,3,2]).
  • the sub-band cropping module divides the last The corresponding loops in the group are filled with PMI and cropped to obtain the original PMI.
  • the basic AI network composed of EN Block and DE Block transforms and inversely transforms the PMI of 12 subbands, 32 ports, and layer 1.
  • Example six (number of subbands + ports + feedback overhead scalable CSI compression feedback model, Layer-Specified AI model):
  • Figure 13 is a schematic structural diagram of a terminal provided by an embodiment of the present disclosure. As shown in Figure 13, the terminal includes a memory 1320, a transceiver 1300, and a processor 1310, wherein:
  • Memory 1320 is used to store computer programs; transceiver 1300 is used to send and receive data under the control of the processor 1310; processor 1310 is used to read the computer program in the memory 1320 and perform the following operations:
  • the encoder model includes an encoder and a channel state information extension enhancement module cascaded with the encoder.
  • the bus architecture may include any number of interconnected buses and bridges, specifically one or more processors represented by processor 1310 and various circuits of the memory represented by memory 1320 are linked together.
  • the bus architecture can also link together various other circuits such as peripherals, voltage regulators, and power management circuits, which are all well known in the art and therefore will not be described further herein.
  • the bus interface provides the interface.
  • the transceiver 1300 may be a plurality of components, including a transmitter and a receiver, providing a unit for communicating with various other devices over transmission media, including wireless channels, wired channels, optical cables, and other transmission media.
  • the user interface 1330 can also be an interface capable of externally connecting internal and external required equipment.
  • the connected equipment includes but is not limited to a keypad, a monitor, a speaker, a microphone, a joystick, etc.
  • the processor 1310 is responsible for managing the bus architecture and general processing, and the memory 1320 can store data used by the processor 1310 when performing operations.
  • the processor 1310 can be a central processing unit (Central Processing Unit, CPU), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field-Programmable Gate Array, FPGA) or a complex programmable Logic device (Complex Programmable Logic Device, CPLD), the processor can also adopt a multi-core architecture.
  • CPU Central Processing Unit
  • ASIC Application Specific Integrated Circuit
  • FPGA Field-Programmable Gate Array
  • CPLD Complex Programmable Logic Device
  • the processor is configured to execute any of the methods provided by the embodiments of the present disclosure according to the obtained executable instructions by calling the computer program stored in the memory.
  • the processor and memory can also be physically separated.
  • the channel state information extension enhancement module includes a downsampling module
  • the down-sampling module is cascaded at the output side of the encoder and is used to compress the feedback overhead of the channel state information to obtain a compressed feedback overhead;
  • the extended enhanced channel state information includes the compressed feedback overhead.
  • the channel state information extension enhancement module includes a pre-conversion module
  • the pre-conversion module is cascaded on the input side of the encoder and used to perform at least one of the following:
  • the extended enhanced channel state information includes channel state information of different MIMO layers and/or different MIMO ports whose probability distribution and dimension are unified.
  • the channel state information extension enhancement module includes a subband filling module
  • the subband filling module is cascaded at the input side of the encoder and is used to fill in the number of subbands of the channel state information to obtain channel state information after subband filling;
  • the extended enhanced channel state information includes the channel state information after subband filling.
  • the downsampling module includes any of the following configurations:
  • the downsampling module is a pooling layer
  • the downsampling module is a fully connected layer whose input dimension is greater than the output dimension
  • the downsampling module is a convolutional downsampling layer.
  • the output dimension of the downsampling module is determined based on the type of the feedback overhead and the predetermined number of quantization bits.
  • the pre-conversion module includes any of the following configurations:
  • the pre-transformation module is an identity transformation layer
  • the pre-transformation module adopts a fully connected layer in the spatial domain and frequency domain respectively;
  • the pre-transformation module uses a 2-dimensional convolution layer in the spatial domain and frequency domain;
  • the pre-transformation module uses a fully connected layer to combine dimensions in the spatial domain and frequency domain;
  • the pre-transformation module combines dimensions in the spatial domain and frequency domain and uses a 1-dimensional convolution layer
  • the pre-transformation module uses a fully connected layer in the spatial domain
  • the pre-transformation module uses a 1-dimensional convolutional layer in the spatial domain.
  • the subband filling module is specifically used to:
  • the sub-bands of the channel state information are processed based on the number of input sub-bands of the encoder model. group;
  • the subbands in the target group whose number of subbands of the channel state information is smaller than the number of input subbands of the encoder model are periodically filled until the number of subbands in the target group is equal to the number of input subbands of the encoder model.
  • the number of input subbands for the encoder model
  • the periodic loop filling includes: copying each subband in sequence according to the arrangement order of each subband in the group.
  • the above-mentioned terminal provided by the embodiment of the present disclosure can implement all the method steps implemented by the above-mentioned method embodiment in which the execution subject is the terminal, and can achieve the same technical effect. No further explanation will be given here. The same parts and beneficial effects as those in the method embodiment will be described in detail.
  • Figure 14 is a schematic structural diagram of a network device provided by an embodiment of the present disclosure. As shown in Figure 14, the network device includes a memory 1420, a transceiver 1400, and a processor 1410, where:
  • Memory 1420 is used to store computer programs; transceiver 1400 is used to send and receive data under the control of the processor 1410; processor 1410 is used to read the computer program in the memory 1420 and perform the following operations:
  • the decoder model includes a decoder, and a channel state information restoration module cascaded with the decoder.
  • the transceiver 1400 is used to receive and transmit data under the control of the processor 1410.
  • the bus architecture may include any number of interconnected buses and bridges, specifically one or more processors represented by processor 1410 and various circuits of the memory represented by memory 1420 are linked together.
  • the bus architecture can also link together various other circuits such as peripherals, voltage regulators, and power management circuits, which are all well known in the art and therefore will not be described further herein.
  • the bus interface provides the interface.
  • the transceiver 1400 may be a plurality of components, including a transmitter and a receiver, providing a unit for communicating with various other devices over transmission media, including wireless channels, wired channels, optical cables, and other transmission media.
  • the processor 1410 is responsible for managing the bus architecture and general processing, and the memory 1420 can store data used by the processor 1410 when performing operations.
  • the processor 1410 may be a central processing unit (CPU), a dedicated processor Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA) or Complex Programmable Logic Device (CPLD), and the processor can also adopt a multi-core architecture.
  • CPU central processing unit
  • ASIC Application Specific Integrated Circuit
  • FPGA Field-Programmable Gate Array
  • CPLD Complex Programmable Logic Device
  • the expanded enhanced channel state information includes compressed feedback overhead;
  • the channel state information restoration module includes an upsampling module;
  • the upsampling module is cascaded at the input side of the decoder and is used to restore the compressed feedback overhead to obtain restored feedback overhead.
  • the extended enhanced channel state information includes channel state information of different multiple-input multiple-output MIMO layers and/or different MIMO ports with unified probability distribution and dimension;
  • the channel state information restoration module includes a transformation module;
  • the transformation module is cascaded on the output side of the decoder and is used to perform at least one of the following:
  • the probability distribution and dimensionality of the channel state information of different MIMO ports with unified probability distributions and dimensions are restored to obtain original probability distributions and channel state information of different MIMO ports with dimensions.
  • the extended enhanced channel state information includes channel state information after subband filling;
  • the channel state information restoration module includes a subband clipping module;
  • the subband clipping module is cascaded at the output side of the decoder and is used to clip and/or combine the channel state information after subband filling to obtain the channel state information of the original number of subbands.
  • the upsampling module includes any of the following configurations:
  • the upsampling module is a neighboring interpolation layer
  • the upsampling module is a fully connected layer whose input dimension is smaller than the output dimension
  • the upsampling module is a transposed convolutional upsampling layer.
  • the input dimension of the upsampling module is determined based on the type of feedback overhead and the predetermined number of quantization bits.
  • the transformation module includes any of the following configurations:
  • the transformation module is an identity transformation layer
  • the transformation module adopts a fully connected layer in the spatial domain and frequency domain respectively;
  • the transformation module uses a 2-dimensional transposed convolution layer in the spatial domain and frequency domain;
  • the transformation module combines dimensions in the spatial domain and frequency domain and uses a fully connected layer
  • the transformation module combines dimensions in the spatial domain and frequency domain and uses a 1-dimensional convolution layer
  • the transformation module uses a fully connected layer in the spatial domain
  • the transformation module uses a 1-dimensional convolutional layer in the spatial domain.
  • the subband cropping module is specifically used to:
  • the above-mentioned network device provided by the embodiment of the present disclosure can implement all the method steps implemented by the above-mentioned method embodiment in which the execution subject is a network device, and can achieve the same technical effect. This implementation will no longer be discussed here. The parts and beneficial effects in the examples that are the same as those in the method embodiments will be described in detail.
  • Figure 15 is one of the structural schematic diagrams of a channel state information processing device provided by an embodiment of the present disclosure. As shown in Figure 15, an embodiment of the present disclosure also provides a channel state information processing device, which is applied to a terminal.
  • the device includes:
  • Enhancement module 1510 configured to input channel state information into the encoder model and obtain extended enhanced channel state information output by the encoder model
  • a sending module 1520 configured to send the extended enhanced channel state information to a network device
  • the encoder model includes an encoder and a channel state information extension enhancement module cascaded with the encoder.
  • the channel state information extension enhancement module includes a downsampling module
  • the down-sampling module is cascaded at the output side of the encoder and is used to compress the feedback overhead of the channel state information to obtain a compressed feedback overhead;
  • the extended enhanced channel state information includes the compressed feedback overhead.
  • the channel state information extension enhancement module includes a pre-conversion module
  • the pre-conversion module is cascaded on the input side of the encoder and used to perform at least one of the following:
  • the extended enhanced channel state information includes channel state information of different MIMO layers and/or different MIMO ports whose probability distribution and dimension are unified.
  • the channel state information extension enhancement module includes a subband filling module
  • the subband filling module is cascaded at the input side of the encoder and is used to fill in the number of subbands of the channel state information to obtain channel state information after subband filling;
  • the extended enhanced channel state information includes the channel state information after subband filling.
  • the downsampling module includes any of the following configurations:
  • the downsampling module is a pooling layer
  • the downsampling module is a fully connected layer whose input dimension is greater than the output dimension
  • the downsampling module is a convolutional downsampling layer.
  • the output dimension of the downsampling module is determined based on the type of the feedback overhead and the predetermined number of quantization bits.
  • the pre-conversion module includes any of the following configurations:
  • the pre-transformation module is an identity transformation layer
  • the pre-transformation module adopts a fully connected layer in the spatial domain and frequency domain respectively;
  • the pre-transformation module uses a 2-dimensional transposed convolution layer in the spatial domain and frequency domain;
  • the pre-transformation module uses a fully connected layer to combine dimensions in the spatial domain and frequency domain;
  • the pre-transformation module combines dimensions in the spatial domain and frequency domain and uses a 1-dimensional convolution layer
  • the pre-transformation module uses a fully connected layer in the spatial domain
  • the pre-transformation module uses a 1-dimensional convolutional layer in the spatial domain.
  • the subband filling module is specifically used to:
  • the sub-bands of the channel state information are processed based on the number of input sub-bands of the encoder model. grouping;
  • the subbands in the target group whose number of subbands of the channel state information is less than the number of input subbands of the encoder model are periodically filled until the number of subbands in the target group is equal to the number of input subbands of the encoder model.
  • the number of input subbands for the encoder model
  • the periodic loop filling includes: copying each subband in sequence according to the arrangement order of each subband in the group.
  • Figure 16 is a second structural schematic diagram of a channel state information processing device provided by an embodiment of the present disclosure. As shown in Figure 16, an embodiment of the present disclosure also provides a channel state information processing device, which is applied to network equipment.
  • the device includes:
  • the receiving module 1610 is used to receive the extended enhanced channel state information sent by the terminal;
  • the restoration module 1620 is configured to input the extended enhanced channel state information to the decoder model to obtain the restored channel state information output by the decoder model;
  • the decoder model includes a decoder, and a channel state information restoration module cascaded with the decoder.
  • the expanded enhanced channel state information includes compressed feedback overhead;
  • the channel state information restoration module includes an upsampling module;
  • the upsampling module is cascaded at the input side of the decoder and is used to restore the compressed feedback overhead to obtain restored feedback overhead.
  • the extended enhanced channel state information includes channel state information of different multiple-input multiple-output MIMO layers and/or different MIMO ports with unified probability distribution and dimension;
  • the channel state information restoration module includes a transformation module;
  • the transformation module is cascaded on the output side of the decoder and is used to perform at least one of the following:
  • the probability distribution and dimensionality of the channel state information of different MIMO ports with unified probability distributions and dimensions are restored to obtain original probability distributions and channel state information of different MIMO ports with dimensions.
  • the extended enhanced channel state information includes channel state information after subband filling;
  • the channel state information restoration module includes a subband clipping module;
  • the subband clipping module is cascaded at the output side of the decoder and is used to clip and/or combine the channel state information after subband filling to obtain the channel state information of the original number of subbands.
  • the upsampling module includes any of the following configurations:
  • the upsampling module is a neighboring interpolation layer
  • the upsampling module is a fully connected layer whose input dimension is smaller than the output dimension
  • the upsampling module is a transposed convolutional upsampling layer.
  • the input dimension of the upsampling module is determined based on the type of feedback overhead and the predetermined number of quantization bits.
  • the transformation module includes any of the following configurations:
  • the transformation module is an identity transformation layer
  • the transformation module adopts a fully connected layer in the spatial domain and frequency domain respectively;
  • the transformation module uses a 2-dimensional transposed convolution layer in the spatial domain and frequency domain;
  • the transformation module combines dimensions in the spatial domain and frequency domain and uses a fully connected layer
  • the transformation module combines dimensions in the spatial domain and frequency domain and uses a 1-dimensional convolution layer
  • the transformation module uses a fully connected layer in the spatial domain
  • the transformation module uses a 1-dimensional convolutional layer in the spatial domain.
  • the subband cropping module is specifically used to:
  • the above-mentioned device provided by the embodiment of the present disclosure can implement all the method steps implemented by the above-mentioned method embodiment in which the execution subject is a network device, and can achieve the same technical effect. This embodiment will no longer be discussed here. The same parts and beneficial effects as those in the method embodiments will be described in detail.
  • each functional unit in various embodiments of the present disclosure may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.
  • the above integrated units are both It can be implemented in the form of hardware or in the form of software functional units.
  • the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a processor-readable storage medium.
  • the technical solution of the present disclosure is essentially or contributes to the existing technology, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to cause a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of the present disclosure.
  • the aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk and other media that can store program code. .
  • embodiments of the present disclosure also provide a processor-readable storage medium.
  • the processor-readable storage medium stores a computer program.
  • the computer program is used to cause the processor to execute the methods provided by the above embodiments. Methods include, for example:
  • the encoder model includes an encoder and a channel state information expansion enhancement module cascaded by the encoder.
  • the decoder model includes a decoder, and a channel state information restoration module cascaded with the decoder.
  • the processor-readable storage medium may be any available media or data storage device that the processor can access, including but not limited to magnetic storage (such as floppy disks, hard disks, tapes, magneto-optical disks (MO), etc.), optical storage (such as CD, DVD, BD, HVD, etc.), and semiconductor memories (such as ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid state drive (SSD)), etc.
  • magnetic storage such as floppy disks, hard disks, tapes, magneto-optical disks (MO), etc.
  • optical storage such as CD, DVD, BD, HVD, etc.
  • semiconductor memories such as ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid state drive (SSD)
  • GSM global system of mobile communication
  • CDMA code division multiple access
  • WCDMA wideband code division multiple access
  • GPRS general packet Wireless service
  • LTE long term evolution
  • FDD frequency division duplex
  • TDD LTE time division duplex
  • UMTS Universal mobile telecommunication system
  • WiMAX microwave access
  • 5G New Radio, NR 5G New Radio
  • EPS Evolved Packet System
  • 5GS 5G system
  • EPS Evolved Packet System
  • 5GS 5G system
  • the terminal device involved in the embodiments of the present disclosure may be a device that provides voice and/or data connectivity to users, a handheld device with a wireless connection function, or other processing devices connected to a wireless modem, etc.
  • the names of terminal equipment may also be different.
  • the terminal equipment may be called User Equipment (UE).
  • UE User Equipment
  • Wireless terminal equipment can communicate with one or more core networks (Core Network, CN) via a Radio Access Network (RAN).
  • RAN Radio Access Network
  • the wireless terminal equipment can be a mobile terminal equipment, such as a mobile phone (also known as a "cellular phone").
  • Wireless terminal equipment may also be called a system, a subscriber unit, a subscriber station, a mobile station, a mobile station, a remote station, or an access point.
  • remote terminal equipment remote terminal equipment
  • access terminal equipment access terminal
  • user terminal user terminal
  • user agent user agent
  • user device user device
  • the network device involved in the embodiment of the present disclosure may be a base station, and the base station may include multiple cells that provide services for terminals.
  • a base station can also be called an access point, or it can be a device in the access network that communicates with wireless terminal equipment through one or more sectors on the air interface, or it can be named by another name.
  • Network equipment can be used to exchange received air frames with Internet Protocol (IP) packets and act as a router between the wireless terminal equipment and the rest of the access network, which can include the Internet. Protocol (IP) communication network.
  • IP Internet Protocol
  • Network devices also coordinate attribute management of the air interface.
  • the network equipment involved in the embodiments of the present disclosure may be a network equipment (Base Transceiver Station, BTS) in the Global System for Mobile communications (GSM) or Code Division Multiple Access (CDMA). ), or it can be a network device (NodeB) in a Wide-band Code Division Multiple Access (WCDMA), or an evolutionary network device in a long term evolution (LTE) system (evolutional Node B, eNB or e-NodeB), 5G base station (gNB) in the 5G network architecture (next generation system), or home evolved base station (Home evolved Node B, HeNB), relay node (relay node) , home base station (femto), pico base station (pico), etc., are not limited in the embodiments of the present disclosure.
  • network equipment may include centralized unit (CU) nodes and distributed unit (DU) nodes.
  • the centralized unit and distributed unit may also be arranged geographically separately.
  • Network equipment and terminal equipment can each use one or more antennas for multi-input multi-output (MIMO) transmission.
  • MIMO transmission can be single-user MIMO (Single User MIMO, SU-MIMO) or multi-user MIMO. (Multiple User MIMO,MU-MIMO).
  • MIMO transmission can be 2D-MIMO, 3D-MIMO, FD-MIMO or massive-MIMO, or it can be diversity transmission, precoding transmission or beamforming transmission, etc.
  • embodiments of the present disclosure may be provided as methods, systems, or computer program products. Accordingly, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment that combines software and hardware aspects. Furthermore, the present disclosure may be embodied in one or more computer-usable storage media (including, but not limited to, disk storage) having computer-usable program code embodied therein. memory and optical storage, etc.)).
  • computer-usable storage media including, but not limited to, disk storage
  • computer-usable program code embodied therein. memory and optical storage, etc.
  • processor-executable instructions may also be stored in a processor-readable memory that causes a computer or other programmable data processing apparatus to operate in a particular manner, such that the generation of instructions stored in the processor-readable memory includes the manufacture of the instruction means product, the instruction device implements the function specified in one process or multiple processes in the flow chart and/or one block or multiple blocks in the block diagram.
  • processor-executable instructions may also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce computer-implemented processing, thereby causing the computer or other programmable device to
  • the instructions that are executed provide steps for implementing the functions specified in a process or processes of the flowchart diagrams and/or a block or blocks of the block diagrams.

Landscapes

  • Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Compression, Expansion, Code Conversion, And Decoders (AREA)
  • Communication Control (AREA)

Abstract

本公开涉及通信领域，提供一种信道状态信息处理方法及装置。所述方法包括：将信道状态信息输入至编码器模型，得到编码器模型输出的扩展增强的信道状态信息；将所述扩展增强的信道状态信息发送至网络设备；其中，所述编码器模型包括编码器，以及与所述编码器级联的信道状态信息扩展增强模块。

Description

信道状态信息处理方法及装置
相关申请的交叉引用
本申请要求于2022年08月12日提交的申请号为202210969110.3，发明名称为“信道状态信息处理方法及装置”的中国专利申请的优先权，其通过引用方式全部并入本文。
技术领域
本公开涉及通信技术领域，尤其涉及一种信道状态信息处理方法及装置。
背景技术
目前，基于人工智能(Artificial Intelligence，AI)的信道状态信息(Channel State Information，CSI)压缩反馈方案通常在给定反馈开销、多输入多输出(Multiple-Input Multiple-Output，MIMO)层数、MIMO端口数，子带数下设计AI模型。
当存在多种网络配置参数时，相关方案需要训练多套AI模型，这会增加AI模型的训练复杂度和AI模型的存储开销。以8种反馈开销，4层MIMO，8种端口为例，网络层和终端需要针对8×4×8＝256种参数配置分别训练和部署一个AI模型，这给模型的训练和存储带来了极大挑战。
发明内容
本公开实施例提供一种信道状态信息处理方法及装置，用以解决现有技术中对于信道状态信息处理需要部署大量AI模型，造成模型训练和部署浪费大量资源的缺陷，实现降低AI模型训练的复杂度并节约AI模型的部署资源的目的。
第一方面，本公开实施例提供一种信道状态信息处理方法，应用于终端，所述方法包括：
将信道状态信息输入至编码器模型，得到所述编码器模型输出的扩展增强的信道状态信息；
将所述扩展增强的信道状态信息发送至网络设备；
其中，所述编码器模型包括编码器，以及与所述编码器级联的信道状态信息扩展增强模块。
在一个实施例中，所述信道状态信息扩展增强模块包括下采样模块；
所述下采样模块级联于所述编码器的输出侧，用于对所述信道状态信息的反馈开销进行压缩，得到压缩后的反馈开销；
所述扩展增强的信道状态信息包括所述压缩后的反馈开销。
在一个实施例中，所述信道状态信息扩展增强模块包括预变换模块；
所述预变换模块级联于所述编码器的输入侧，用于执行以下至少一项：
调整不同多入多输出MIMO层的信道状态信息的概率分布，得到概率分布和维度统一的不同MIMO层的信道状态信息；
调整不同MIMO端口的信道状态信息的概率分布和/或维度，得到概率分布和维度统一的不同MIMO端口的信道状态信息；
所述扩展增强的信道状态信息包括所述概率分布和维度统一的不同MIMO层和/或不同MIMO端口的信道状态信息。
在一个实施例中，所述信道状态信息扩展增强模块包括子带填充模块；
所述子带填充模块级联于所述编码器的输入侧，用于对所述信道状态信息的子带数进行填充，得到子带填充后的信道状态信息；
所述扩展增强的信道状态信息包括所述子带填充后的信道状态信息。
在一个实施例中，所述下采样模块包括以下任一项配置：
所述下采样模块为池化层；
所述下采样模块为输入维度大于输出维度的全连接层；
所述下采样模块为卷积下采样层。
在一个实施例中，所述下采样模块的输出维度是基于所述反馈开销的种类以及预定的量化比特数确定的。
在一个实施例中，所述预变换模块包括以下任一项配置：
所述预变换模块为恒等变换层；
所述预变换模块在空域和频域分别采用一个全连接层；
所述预变换模块在空域和频域采用一个2维卷积层；
所述预变换模块在空域和频域合并维度，采用一个全连接层；
所述预变换模块在空域和频域合并维度，采用一个1维卷积层；
所述预变换模块在空域采用一个全连接层；
所述预变换模块在空域采用一个1维卷积层。
在一个实施例中，所述子带填充模块具体用于：
在所述信道状态信息的子带数与所述编码器模型的输入子带数的余数大于0的情况下，基于所述编码器模型的输入子带数对所述信道状态信息的子带进行分组；
对所述信道状态信息的子带数小于所述编码器模型的输入子带数的目标分组中的子带进行周期循环填充，直到所述目标分组中的子带数量等于所述编码器模型的输入子带数；
所述周期循环填充包括：根据分组中各子带的排列顺序，依次复制各子带。
第二方面，本公开实施例还提供一种信道状态信息处理方法，应用于网络设备，所述方法包括：
接收终端发送的扩展增强的信道状态信息；
将所述扩展增强的信道状态信息输入至解码器模型，得到所述解码器模型输出的还原信道状态信息；
其中，所述解码器模型包括解码器，以及与所述解码器级联的信道状态信息还原模块。
在一个实施例中，所述扩展增强的信道状态信息包括压缩后的反馈开销；所述信道状态信息还原模块包括上采样模块；
所述上采样模块级联于所述解码器的输入侧，用于对所述压缩后的反馈开销进行还原，得到还原的反馈开销。
在一个实施例中，所述扩展增强的信道状态信息包括概率分布和维度统一的不同多输入多输出MIMO层和/或不同MIMO端口的信道状态信息；所述信道状态信息还原模块包括变换模块；
所述变换模块级联于所述解码器的输出侧，用于执行以下至少一项：
对所述概率分布和维度统一的不同MIMO层的信道状态信息进行概率分布还原，得到原始概率分布的不同MIMO层的信道状态信息；
对所述概率分布和维度统一的不同MIMO端口的信道状态信息进行概率分布和维度还原，得到原始概率分布和维度的不同MIMO端口的信道状态信息。
在一个实施例中，所述扩展增强的信道状态信息包括子带填充后的信道状态信息；所述信道状态信息还原模块包括子带裁剪模块；
所述子带裁剪模块级联于所述解码器的输出侧，用于对子带填充后的信道状态信息进行裁剪和/或合并，得到原始子带数量的信道状态信息。
在一个实施例中，所述上采样模块包括以下任一项配置：
所述上采样模块为邻近插值层；
所述上采样模块为输入维度小于输出维度的全连接层；
所述上采样模块为转置卷积上采样层。
在一个实施例中，所述上采样模块的输入维度是基于所述反馈开销的种类以及预定的量化比特数确定的。
在一个实施例中，所述变换模块包括以下任一项配置：
所述变换模块为恒等变换层；
所述变换模块在空域和频域分别采用一个全连接层；
所述变换模块在空域和频域采用一个2维转置卷积层；
所述变换模块在空域和频域合并维度，采用一个全连接层；
所述变换模块在空域和频域合并维度，采用一个1维卷积层；
所述变换模块在空域采用一个全连接层；
所述变换模块在空域采用一个1维卷积层。
在一个实施例中，所述子带裁剪模块具体用于：
对进行过子带周期循环填充的子带分组中的子带进行裁剪和/或合并。
第三方面，本公开实施例还提供一种终端，包括存储器，收发机，处理器：
存储器，用于存储计算机程序；收发机，用于在所述处理器的控制下收发数据；处理器，用于读取所述存储器中的计算机程序并执行以下操作：
将信道状态信息输入至编码器模型，得到所述编码器模型输出的扩展增强的信道状态信息；
将所述扩展增强的信道状态信息发送至网络设备；
其中，所述编码器模型包括编码器，以及与所述编码器级联的信道状态信息扩展增强模块。
在一个实施例中，所述信道状态信息扩展增强模块包括下采样模块；
所述下采样模块级联于所述编码器的输出侧，用于对所述信道状态信息的反馈开销进行压缩，得到压缩后的反馈开销；
所述扩展增强的信道状态信息包括所述压缩后的反馈开销。
在一个实施例中，所述信道状态信息扩展增强模块包括预变换模块；
所述预变换模块级联于所述编码器的输入侧，用于执行以下至少一项：
调整不同多入多输出MIMO层的信道状态信息的概率分布，得到概率分布和维度统一的不同MIMO层的信道状态信息；
调整不同MIMO端口的信道状态信息的概率分布和/或维度，得到概率分布和维度统一的不同MIMO端口的信道状态信息；
所述扩展增强的信道状态信息包括所述概率分布和维度统一的不同MIMO层和/或不同MIMO端口的信道状态信息。
在一个实施例中，所述信道状态信息扩展增强模块包括子带填充模块；
所述子带填充模块级联于所述编码器的输入侧，用于对所述信道状态信息的子带数进行填充，得到子带填充后的信道状态信息；
所述扩展增强的信道状态信息包括所述子带填充后的信道状态信息。
在一个实施例中，所述下采样模块包括以下任一项配置：
所述下采样模块为池化层；
所述下采样模块为输入维度大于输出维度的全连接层；
所述下采样模块为卷积下采样层。
在一个实施例中，所述下采样模块的输出维度是基于所述反馈开销的种 类以及预定的量化比特数确定的。
在一个实施例中，所述预变换模块包括以下任一项配置：
所述预变换模块为恒等变换层；
所述预变换模块在空域和频域分别采用一个全连接层；
所述预变换模块在空域和频域采用一个2维转置卷积层；
所述预变换模块在空域和频域合并维度，采用一个全连接层；
所述预变换模块在空域和频域合并维度，采用一个1维卷积层；
所述预变换模块在空域采用一个全连接层；
所述预变换模块在空域采用一个1维卷积层。
在一个实施例中，所述子带填充模块具体用于：
在所述信道状态信息的子带数与所述编码器模型的输入子带数的余数大于0的情况下，基于所述编码器模型的输入子带数对所述信道状态信息的子带进行分组；
对所述信道状态信息的子带数小于所述编码器模型的输入子带数的目标分组中的子带进行周期循环填充，直到所述目标分组中的子带数量等于所述编码器模型的输入子带数；
所述周期循环填充包括：根据分组中各子带的排列顺序，依次复制各子带。
第四方面，本公开实施例还提供一种网络设备，包括存储器，收发机，处理器：
存储器，用于存储计算机程序；收发机，用于在所述处理器的控制下收发数据；处理器，用于读取所述存储器中的计算机程序并执行以下操作：
接收终端发送的扩展增强的信道状态信息；
将所述扩展增强的信道状态信息输入至解码器模型，得到所述解码器模型输出的还原信道状态信息；
其中，所述解码器模型包括解码器，以及与所述解码器级联的信道状态信息还原模块。
在一个实施例中，所述扩展增强的信道状态信息包括压缩后的反馈开销； 所述信道状态信息还原模块包括上采样模块；
所述上采样模块级联于所述解码器的输入侧，用于对所述压缩后的反馈开销进行还原，得到还原的反馈开销。
在一个实施例中，所述扩展增强的信道状态信息包括概率分布和维度统一的不同多输入多输出MIMO层和/或不同MIMO端口的信道状态信息；所述信道状态信息还原模块包括变换模块；
所述变换模块级联于所述解码器的输出侧，用于执行以下至少一项：
对所述概率分布和维度统一的不同MIMO层的信道状态信息进行概率分布还原，得到原始概率分布的不同MIMO层的信道状态信息；
对所述概率分布和维度统一的不同MIMO端口的信道状态信息进行概率分布和维度还原，得到原始概率分布和维度的不同MIMO端口的信道状态信息。
在一个实施例中，所述扩展增强的信道状态信息包括子带填充后的信道状态信息；所述信道状态信息还原模块包括子带裁剪模块；
所述子带裁剪模块级联于所述解码器的输出侧，用于对子带填充后的信道状态信息进行裁剪和/或合并，得到原始子带数量的信道状态信息。
在一个实施例中，所述上采样模块包括以下任一项配置：
所述上采样模块为邻近插值层；
所述上采样模块为输入维度小于输出维度的全连接层；
所述上采样模块为转置卷积上采样层。
在一个实施例中，所述上采样模块的输入维度是基于所述反馈开销的种类以及预定的量化比特数确定的。
在一个实施例中，所述变换模块包括以下任一项配置：
所述变换模块为恒等变换层；
所述变换模块在空域和频域分别采用一个全连接层；
所述变换模块在空域和频域采用一个2维转置卷积层；
所述变换模块在空域和频域合并维度，采用一个全连接层；
所述变换模块在空域和频域合并维度，采用一个1维卷积层；
所述变换模块在空域采用一个全连接层；
所述变换模块在空域采用一个1维卷积层。
在一个实施例中，所述子带裁剪模块具体用于：
对进行过子带周期循环填充的子带分组中的子带进行裁剪和/或合并。
第五方面，本公开实施例还提供一种信道状态信息处理装置，应用于终端，所述装置包括：
增强模块，用于将信道状态信息输入至编码器模型，得到所述编码器模型输出的扩展增强的信道状态信息；
发送模块，用于将所述扩展增强的信道状态信息发送至网络设备；
其中，所述编码器模型包括编码器，以及与所述编码器级联的信道状态信息扩展增强模块。
第六方面，本公开实施例还提供一种信道状态信息处理装置，应用于网络设备，所述装置包括：
接收模块，用于接收终端发送的扩展增强的信道状态信息；
还原模块，用于将所述扩展增强的信道状态信息输入至解码器模型，得到所述解码器模型输出的还原信道状态信息；
其中，所述解码器模型包括解码器，以及与所述解码器级联的信道状态信息还原模块。
第七方面，本公开实施例还提供一种处理器可读存储介质，所述处理器可读存储介质存储有计算机程序，所述计算机程序用于使所述处理器执行如上所述第一方面的方法，或者执行如上所述第二方面的方法。
本公开实施例提供的信道状态信息处理方法及装置，通过对编码器模型的输入侧和/或输出侧级联信道状态信息扩展增强模块，对AI模型进行扩展增强，可以实现在不同参数配置下复用AI模型，从而能够降低AI模型训练的复杂度并节约AI模型的部署资源。
附图说明
为了更清楚地说明本公开实施例或现有技术中的技术方案，下面将对实 施例或现有技术描述中所需要使用的附图作一简单地介绍，显而易见地，下面描述中的附图是本公开的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1是现有技术中自编码器的结构示意图；
图2是现有技术中基于AI的信道状态信息压缩反馈方案的流程示意图；
图3是本公开实施例提供的信道状态信息处理方法的流程示意图之一；
图4是根据本公开实施例的下采样模块-上采样模块对的结构示意图；
图5是根据本公开实施例的预变换模块-变换模块对的结构示意图；
图6是根据本公开实施例的级联子带填充模块与子带裁剪模块的结构示意图；
图7是本公开实施例提供的信道状态信息处理方法的流程示意图之二；
图8是应用本公开实施例提供的信道状态信息处理方法的实例结构图之一；
图9是应用本公开实施例提供的信道状态信息处理方法的实例结构图之二；
图10是应用本公开实施例提供的信道状态信息处理方法的实例结构图之三；
图11是应用本公开实施例提供的信道状态信息处理方法的实例结构图之四；
图12是应用本公开实施例提供的信道状态信息处理方法的实例结构图之五；
图13是本公开实施例提供的终端的结构示意图；
图14是本公开实施例提供的网络设备的结构示意图；
图15是本公开实施例提供的信道状态信息处理装置的结构示意图之一；
图16是本公开实施例提供的信道状态信息处理装置的结构示意图之二。
具体实施方式
本公开实施例中术语“和/或”，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。字符“/”一般表示前后关联对象是一种“或”的关系。
本公开实施例中术语“多个”是指两个或两个以上，其它量词与之类似。
下面将结合本公开实施例中的附图，对本公开实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本公开一部分实施例，并不是全部的实施例。基于本公开中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本公开保护的范围。
本公开实施例提供了一种信道状态信息处理方法及装置，用以解决现有技术中对于信道状态信息处理需要部署大量AI模型，造成模型训练和部署浪费大量资源的缺陷。
其中，方法和装置是基于同一申请构思的，由于方法和装置解决问题的原理相似，因此装置和方法的实施可以相互参见，重复之处不再赘述。
为便于充分理解本公开的技术方案，现对以下内容进行介绍：
在Massive(大规模)MIMO系统中，基于AI的神经网络模型可用于信道状态信息反馈。通过利用信道状态信息在空域和频域的稀疏性，神经网络模型可采用自编码器(Auto Encoder，AE)的编码器模块实现对信道状态信息的压缩以减少反馈开销，再利用AE的解码器模块恢复出信道状态信息。
自编码器,是一种输入与输出相同的神经网络，由编码器和解码器两部分组成，编码器把输入数据压缩成潜在空间(latent space)表征，解码器根据这一表征重构输入。自编码器的结构如图1所示，输入数据X首先经过编码器被压缩为编码特征Z＝f(X)，然后编码特征Z经过解码器得到目的是重构输入数据X。自编码器通过反向传播的训练方式求解f(·)和g(·)的映射关系，最小化重构误差编码器模块和解码器模块可以采用卷积神经网络、全连接神经网络、递归神经网络等及其组合构造。当自编码器应用于信道状态信息压缩反馈时，编码器输出Z的维度小于其输入X，输出Z的维度越小则压缩率越高，相应地解码器恢复信道状态信息的误差也越大。
目前，基于AI的信道状态信息压缩反馈方案通常在给定反馈开销、MIMO层数、MIMO端口数，子带数下设计AI模型，如图2所示。终端侧将信道状 态信息V输入到AI编码器获得压缩的码字C(码字C的信息长度远小于输入V的长度)，然后将该码字C输入量化器，并把量化后二进制比特流输入到反量化器得到具有量化误差的码字C’，最后把C’输入到解码器模块得到恢复的信道信息V’。当存在多种网络配置超参数时，现有方案需要训练多套AI模型，这会增加AI模型的训练复杂度和AI模型的存储开销。以8种反馈开销，4层MIMO，8种端口为例，网络层和终端需要针对8×4×8＝256种参数配置分别训练和部署一个AI模型，这给模型的训练和存储带来了极大挑战。
图3是本公开实施例提供的信道状态信息处理方法的流程示意图之一。如图3所示，本公开实施例提供一种信道状态信息处理方法，其执行主体可以为终端，例如手机等。该方法可以包括：
步骤310、将信道状态信息输入至编码器模型，得到编码器模型输出的扩展增强的信道状态信息；
步骤320、将扩展增强的信道状态信息发送至网络设备；
其中，编码器模型包括编码器，以及与编码器级联的信道状态信息扩展增强模块。
需要说明的是，可以将进行CSI压缩反馈的AI模型作为基础AI模型，并将基础AI模型的编码器部署在终端形成编码器模型，将基础AI模型的解码器部署在网络设备形成解码器模型；同时，通过对基础AI模型的内侧(编码器的输出侧、解码器的输入侧)和/或基础AI模型的外侧(编码器的输入侧、解码器的输出侧)级联多对AI模块来实现基础AI模型的扩展增强，从而可以在不同参数配置下复用基础AI模型。
在步骤310中，终端可以将信道状态信息输入至编码器模型，编码器的输入侧或输出侧级联有信道状态信息扩展增强模块，该信道状态信息扩展增强模块能够以不同方式对基础AI模型进行扩展增强。
例如，对基础AI模型进行扩展增强可以包括对信道状态信息的反馈开销进行压缩、调整不同MIMO层和/或不同MIMO端口的信道状态信息的概率分布、对信道状态信息的子带数进行填充等。
在对信道状态信息进行上述处理后，终端会将扩展增强的信道状态信息发送至网络设备，以便在网络设备部署的解码器模型对扩展增强的信道状态信息进行对应处理，例如对压缩后的反馈开销进行还原、对概率分布统一的不同MIMO层和/或不同MIMO端口的信道状态信息进行概率分布还原、对子带填充后的信道状态信息进行裁剪和/或合并等。
本公开实施例提供的信道状态信息处理方法，通过对编码器模型的输入侧和/或输出侧级联信道状态信息扩展增强模块，对AI模型进行扩展增强，可以实现在不同参数配置下复用AI模型，从而能够降低AI模型训练的复杂度并节约AI模型的部署资源。
在一个实施例中，信道状态信息扩展增强模块包括下采样模块；
下采样模块级联于编码器的输出侧，用于对信道状态信息的反馈开销进行压缩，得到压缩后的反馈开销；
扩展增强的信道状态信息包括压缩后的反馈开销。
对于CSI压缩反馈方案，网络参数配置要求支持不同的反馈开销，该情况需要AI模型编码器的输出和解码器的输入支持多种维度以实现不同反馈开销。在给定反馈开销和量化阶数的条件下，可以确定编码器的输出(解码器的输入)维度，并且训练一个反馈开销给定的基础AI模型。
为了实现不同反馈开销，可以对基础AI模型的内侧(编码器的输出侧、解码器的输入侧)级联至少一对下采样模块A-k和上采样模块B-k实现反馈开销扩展增强，如图4所示。扩展后的基础AI模型存在k条支路，每条支路对应一种反馈开销配置。
在一个实施例中，下采样模块可以包括以下任一项配置：
下采样模块为池化层；
下采样模块为输入维度大于输出维度的全连接层；
下采样模块为卷积下采样层。
为了降低复杂度，A-k下采样模块以及B-k上采样模块不宜设计得过于复杂，可以采用以下任一项配置：
配置1：A-k是池化层，B-k是邻近插值层；
配置2：A-k是一个输入维度大于输出维度的全连接层，B-k是一个输入维度小于输出维度的全连接层；
配置3：A-k是一个卷积下采样层(一个滑动步长大于1的卷积层)，B-k是一个转置卷积上采样层。
其中，A-k的输出维度(B-k的输入维度)为nk＝npayload,k/q，k＝1,2,3,…；q为量化比特数，npayload,k为网络配置的k种反馈开销。
可选地，基础AI模型可以是一个给定反馈开销下的CSI压缩反馈AI模型，A-k下采样模块以及B-k上采样模块在该基础AI模型的基础上进一步下采样压缩，得到多种更低的反馈开销npayload,k。
基础AI模型还可以是一组AI变换-反变换对，不进行维度压缩(也即，基础AI模型的编码器的输出维度等于其输入维度)，维度压缩完全通过A-k下采样模块以及B-k上采样模块实现。
本公开实施例提供的信道状态信息处理方法，通过在编码器的输出侧级联下采样模块，可以实现AI模型对于CSI反馈开销处理的扩展增强，从而实现在不同反馈开销配置下复用AI模型，能够降低AI模型训练的复杂度并节约AI模型的部署资源。
在一个实施例中，信道状态信息扩展增强模块包括预变换模块；
预变换模块级联于编码器的输入侧，用于执行以下至少一项：
调整不同多入多输出MIMO层的信道状态信息的概率分布，得到概率分布和维度统一的不同MIMO层的信道状态信息；
调整不同MIMO端口的信道状态信息的概率分布和/或维度，得到概率分布和维度统一的不同MIMO端口的信道状态信息扩展增强的信道状态信息包括概率分布和维度统一的不同MIMO层和/或不同MIMO端口的信道状态信息。
对于CSI压缩反馈方案，网络参数配置要求支持不同的MIMO层数、Port端口数。对于不同Port端口数，基础AI模型(编码器+解码器)的输入和输出需要支持多种维度。此外，不同配置下基础AI模型输入的概率分布也必须统一或者接近，才能使得基础AI模型工作在最佳状态。换言之，不同Port端 口、不同MIMO层的CSI数据概率分布不一致，需要变换到一个统一的概率分布上。
如图5所示，通过在基础AI模型外侧(编码器的输入侧、解码器的输出侧)级联A-k预变换/B-k变换模块对，使得不同配置下的输入概率分布对齐和/或输入维度对齐。
在一个实施例中，预变换模块包括以下任一项配置：
预变换模块为恒等变换层；
预变换模块在空域和频域分别采用一个全连接层；
预变换模块在空域和频域采用一个2维卷积层；
预变换模块在空域和频域合并维度，采用一个全连接层；
预变换模块在空域和频域合并维度，采用一个1维卷积层；
预变换模块在空域采用一个全连接层；
预变换模块在空域采用一个1维卷积层。
对于基础AI模型的MIMO层扩展增强，A-k预变换/B-k变换模块对的作用为分布变换。
特殊地，A-1/B-1模块是恒等变换(输入等于输出)，基础AI模型是一个适用于层1的AI模型，A-k(k＝2,3,4…)通过预变换使得层k的CSI概率分布接近层1的CSI概率分布，B-k将恢复的层k的CSI变换回原来的分布。
A-k预变换/B-k变换模块对可以包括以下任一项配置：
空域-频域二维线性变换，在空域和频域分别采用一个全连接层，或者整体采用一个2维卷积/转置卷积；
空域-频域拉伸至1维，采用全连接层或者1维卷积。
对于基础AI模型的Port端口扩展增强，A-k预变换/B-k变换模块对的作用为概率分布匹配和维度统一，不同Port端口的CSI输入被变换到某个统一的Port端口数。
特殊地，A-1/B-1模块是恒等变换(输入等于输出)，基础AI模型是一个适用于Port_Max的AI模型，A-k(k＝2,3,4…)通过预变换使得k端口(k<Port_Max)的CSI变换成Port_Max端口的CSI，B-k将恢复的CSI从 Port_Max变换回原来的k端口。
A-k预变换/B-k变换模块对可以包括以下任一项配置：
在空域采用一个全连接层；
在空域采用1维卷积层。
本公开实施例提供的信道状态信息处理方法，通过在编码器的输入侧级联预变换模块，可以实现AI模型对MIMO层和/或MIMO端口进行处理的扩展增强，从而实现在不同MIMO层、不同端口配置下复用AI模型，能够降低AI模型训练的复杂度并节约AI模型的部署资源。
在一个实施例中，信道状态信息扩展增强模块包括子带填充模块；
子带填充模块级联于编码器的输入侧，用于对信道状态信息的子带数进行填充，得到子带填充后的信道状态信息；
扩展增强的信道状态信息包括子带填充后的信道状态信息。
对于CSI压缩反馈方案，网络参数配置要求支持不同的子带数Nsubband，但是用于CSI压缩反馈的基础AI模块输入的子带数Nsubband,AI通常是给定的。为了实现子带数扩展增强，本公开实施例对基础AI模块外侧(编码器的输入侧、解码器的输出侧)级联子带填充模块和子带裁剪模块。如图6所示。
在一个实施例中，子带填充模块具体用于：
在信道状态信息的子带数与编码器模型的输入子带数的余数大于0的情况下，基于编码器模型的输入子带数对信道状态信息的子带进行分组；
对信道状态信息的子带数小于编码器模型的输入子带数的目标分组中的子带进行周期循环填充，直到目标分组中的子带数量等于编码器模型的输入子带数；
周期循环填充包括：根据分组中各子带的排列顺序，依次复制各子带。
在对基础AI模块外侧(编码器的输入侧、解码器的输出侧)级联子带填充模块和子带裁剪模块后，子带填充模块和子带裁剪模块以Nsubband,AI个子带为一组对输入的CSI进行填充和裁剪：
当0＝Nsubband mod Nsubband,AI时，CSI恰好能被划分为Nsubband/Nsubband,AI组，每一组分别送入基础AI网络实现CSI压缩反馈，子带填充模 块和子带裁剪模块不需要对输入做额外处理。
当0≠Nsubband mod Nsubband,AI时，CSI最后一组子带数为Nsubband mod Nsubband,AI低于Nsubband,AI，子带填充模块会对包括Nsubband mod Nsubband,AI个子带的组内成员周期循环填充至Nsubband,AI个子带后送入基础AI模型。子带裁剪模块会对周期循环填充的子带进行裁剪或合并获得原始的Nsubband mod Nsubband,AI个子带的CSI。
其中，对包括K个子带的分组A＝[A1A2A3,…AK]进行周期循环填充至包括N个子带的分组B是指：
B＝[A1A2A3,…AKAK-1AK-2AK-3,…A1A2A3A4,…AKAK-1…]
该操作可以最大程度模拟频域相关性从而提高CSI压缩反馈性能。
本公开实施例提供的信道状态信息处理方法，通过在编码器的输入侧级联子带填充模块，可以实现AI模型对子带数进行处理的扩展增强，从而实现在不同反馈开销配置下复用AI模型，能够降低AI模型训练的复杂度并节约AI模型的部署资源。
图7是本公开实施例提供的信道状态信息处理方法的流程示意图之二。如图7所示，本公开实施例提供一种信道状态信息处理方法，其执行主体可以为网络设备，例如基站等。该方法可以包括：
步骤710、接收终端发送的扩展增强的信道状态信息；
步骤720、将扩展增强的信道状态信息输入至解码器模型，得到解码器模型输出的还原信道状态信息；
其中，解码器模型包括解码器，以及与解码器级联的信道状态信息还原模块。
需要说明的是，可以将进行CSI压缩反馈的AI模型作为基础AI模型，并将基础AI模型的编码器部署在终端形成编码器模型，将基础AI模型的解码器部署在网络设备形成解码器模型；同时，通过对基础AI模型的内侧(编码器的输出侧、解码器的输入侧)和/或基础AI模型的外侧(编码器的输入侧、解码器的输出侧)级联多对AI模块来实现基础AI模型的扩展增强，从而可以在不同参数配置下复用基础AI模型。
在步骤710中，网络设备可以接收终端发送的扩展增强的信道状态信息。
需要说明的是，终端可以将信道状态信息输入至编码器模型，编码器的输入侧或输出侧级联有信道状态信息扩展增强模块，该信道状态信息扩展增强模块能够以不同方式对AI基础模型进行扩展增强。
例如，对信道状态信息的反馈开销进行压缩、调整不同MIMO层和/或不同MIMO端口的信道状态信息的概率分布、对信道状态信息的子带数进行填充等。
在对信道状态信息进行上述处理后，终端会将扩展增强的信道状态信息发送至网络设备，网络设备接收到扩展增强的信道状态信息后，会将该扩展增强的信道状态信息输入到解码器模型，并基于解码器级联的信道状态信息还原模块对扩展增强的信道状态信息进行还原处理，例如对压缩后的反馈开销进行还原、对概率分布统一的不同MIMO层和/或不同MIMO端口的信道状态信息进行概率分布还原、对子带填充后的信道状态信息进行裁剪和/或合并等。
本公开实施例提供的信道状态信息处理方法，通过对解码器模型的输入侧和/或输出侧级联信道状态信息还原模块，对扩展增强的信道状态信息进行还原，可以实现在不同参数配置下复用AI模型，从而能够降低AI模型训练的复杂度并节约AI模型的部署资源。
在一个实施例中，扩展增强的信道状态信息包括压缩后的反馈开销；信道状态信息还原模块包括上采样模块；
上采样模块级联于解码器的输入侧，用于对压缩后的反馈开销进行还原，得到还原的反馈开销。
对于CSI压缩反馈方案，网络参数配置要求支持不同的反馈开销，该情况需要AI模型编码器的输出和解码器的输入支持多种维度以实现不同反馈开销。在给定反馈开销和量化阶数的条件下，可以确定编码器的输出(解码器的输入)维度，并且训练一个反馈开销给定的基础AI模型。
为了实现不同反馈开销，可以对基础AI模型的内侧(编码器的输出侧、解码器的输入侧)级联至少一对下采样模块A-k和上采样模块B-k实现反馈 开销扩展增强，如图4所示。扩展后的基础AI模型存在k条支路，每条支路对应一种反馈开销配置。
在一个实施例中，上采样模块包括以下任一项配置：
上采样模块为邻近插值层；
上采样模块为输入维度小于输出维度的全连接层；
上采样模块为转置卷积上采样层。
为了降低复杂度，A-k下采样模块以及B-k上采样模块不宜设计得过于复杂，可以采用以下任一项配置：
配置1：A-k是池化层，B-k是邻近插值层；
配置2：A-k是一个输入维度大于输出维度的全连接层，B-k是一个输入维度小于输出维度的全连接层；
配置3：A-k是一个卷积下采样层(一个滑动步长大于1的卷积层)，B-k是一个转置卷积上采样层。
其中，A-k的输出维度(B-k的输入维度)为nk＝npayload,k/q，k＝1,2,3,…；q为量化比特数，npayload,k为网络配置的k种反馈开销。
可选地，基础AI模型可以是一个给定反馈开销下的CSI压缩反馈AI模型，A-k下采样模块以及B-k上采样模块在该基础AI模型的基础上进一步下采样压缩，得到多种更低的反馈开销npayload,k。
基础AI模型还可以是一组AI变换-反变换对，不进行维度压缩(也即，基础AI模型的编码器的输出维度等于其输入维度)，维度压缩完全通过A-k下采样模块以及B-k上采样模块实现。
本公开实施例提供的信道状态信息处理方法，通过在解码器的输入侧级联上采样模块，可以实现AI模型对CSI反馈开销处理的扩展增强，从而实现在不同反馈开销配置下复用AI模型，能够降低AI模型训练的复杂度并节约AI模型的部署资源。
在一个实施例中，扩展增强的信道状态信息包括概率分布和维度统一的不同MIMO层和/或不同MIMO端口的信道状态信息；信道状态信息还原模块包括变换模块；
变换模块级联于解码器的输出侧，用于执行以下至少一项：
对概率分布和维度统一的不同MIMO层的信道状态信息进行概率分布还原，得到原始概率分布的不同MIMO层的信道状态信息；
对概率分布和维度统一的不同MIMO端口的信道状态信息进行概率分布和维度还原，得到原始概率分布和维度的不同MIMO端口的信道状态信息。
对于CSI压缩反馈方案，网络参数配置要求支持不同的MIMO层数、Port端口数。对于不同Port端口数，基础AI模型(编码器+解码器)的输入和输出需要支持多种维度。此外，不同配置下基础AI模型输入的概率分布也必须统一或者接近，才能使得基础AI模型工作在最佳状态。换言之，不同Port端口、不同MIMO层的CSI数据概率分布不一致，需要变换到一个统一的概率分布上。
如图5所示，通过在基础AI模型外侧(编码器的输入侧、解码器的输出侧)级联A-k预变换/B-k变换模块对，使得不同配置下的输入概率分布对齐和/或输入维度对齐。
在一个实施例中，变换模块包括以下任一项配置：
变换模块为恒等变换层；
变换模块在空域和频域分别采用一个全连接层；
变换模块在空域和频域采用一个2维转置卷积层；
变换模块在空域和频域合并维度，采用一个全连接层；
变换模块在空域和频域合并维度，采用一个1维卷积层；
变换模块在空域采用一个全连接层；
变换模块在空域采用一个1维卷积层。
对于基础AI模型的MIMO层扩展增强，A-k预变换/B-k变换模块对的作用为分布变换。
特殊地，A-1/B-1模块是恒等变换(输入等于输出)，基础AI模型是一个适用于层1的AI模型，A-k(k＝2,3,4…)通过预变换使得层k的CSI概率分布接近层1的CSI概率分布，B-k将恢复的层k的CSI变换回原来的分布。
A-k预变换/B-k变换模块对可以包括以下任一项配置：
空域-频域二维线性变换，在空域和频域分别采用一个全连接层，或者整体采用一个2维卷积/转置卷积；
空域-频域拉伸至1维，采用全连接层或者1维卷积。
对于基础AI模型的Port端口扩展增强，A-k预变换/B-k变换模块对的作用为概率分布匹配和维度统一，不同Port端口的CSI输入被变换到某个统一的Port端口数。
特殊地，A-1/B-1模块是恒等变换(输入等于输出)，基础AI模型是一个适用于Port_Max的AI模型，A-k(k＝2,3,4…)通过预变换使得k端口(k<Port_Max)的CSI变换成Port_Max端口的CSI，B-k将恢复的CSI从Port_Max变换回原来的k端口。
A-k预变换/B-k变换模块对可以包括以下任一项配置：
在空域采用一个全连接层；
在空域采用1维卷积层。
本公开实施例提供的信道状态信息处理方法，通过在解码器的输出侧级联预变换模块，可以实现AI模型对MIMO层和/或MIMO端口处理的扩展增强，从而实现在不同MIMO层、不同端口配置下复用AI模型，能够降低AI模型训练的复杂度并节约AI模型的部署资源。
在一个实施例中，扩展增强的信道状态信息包括子带填充后的信道状态信息；信道状态信息还原模块包括子带裁剪模块；
子带裁剪模块级联于解码器的输出侧，用于对子带填充后的信道状态信息进行裁剪和/或合并，得到原始子带数量的信道状态信息。
对于CSI压缩反馈方案，网络参数配置要求支持不同的子带数Nsubband，但是用于CSI压缩反馈的基础AI模块输入的子带数Nsubband,AI通常是给定的。为了实现子带数扩展增强，本公开实施例对基础AI模块外侧(编码器的输入侧、解码器的输出侧)级联子带填充模块和子带裁剪模块。如图6所示。
在一个实施例中，子带裁剪模块具体用于：
对进行过子带周期循环填充的子带分组中的子带进行裁剪和/或合并。
在对基础AI模块外侧(编码器的输入侧、解码器的输出侧)级联子带填 充模块和子带裁剪模块后，子带填充模块和子带裁剪模块以Nsubband,AI个子带为一组对输入的CSI进行填充和裁剪：
当0＝Nsubband mod Nsubband,AI时，CSI恰好能被划分为Nsubband/Nsubband,AI组，每一组分别送入基础AI网络实现CSI压缩反馈，子带填充模块和子带裁剪模块不需要对输入做额外处理。
当0≠Nsubband mod Nsubband,AI时，CSI最后一组子带数为Nsubband mod Nsubband,AI低于Nsubband,AI，子带填充模块会对包括Nsubband mod Nsubband,AI个子带的组内成员周期循环填充至Nsubband,AI个子带后送入基础AI模型。子带裁剪模块会对周期循环填充的子带进行裁剪或合并获得原始的Nsubband mod Nsubband,AI个子带的CSI。
其中，对包括K个子带的分组A＝[A1A2A3,…AK]进行周期循环填充至包括N个子带的分组B是指：
B＝[A1A2A3,…AKAK-1AK-2AK-3,…A1A2A3A4,…AKAK-1…]
该操作可以最大程度模拟频域相关性从而提高CSI压缩反馈性能。
本公开实施例提供的信道状态信息处理方法，通过在解码器的输出侧级联子带裁剪模块，可以实AI模型对现子带数处理的扩展增强，从而实现在不同反馈开销配置下复用AI模型，能够降低AI模型训练的复杂度并节约AI模型的部署资源。
下面介绍应用本公开实施例提供的信道状态信息处理方法的若干实例：
实例一(开销可扩展的CSI压缩反馈模型)：
如图8所示，EN Block和DE Block是一组预变换-变换对，采用Transformer模型，EN Block和DE Block的输入输出维度为12*64，输入PMI(预编码向量)为12个子带32端口(实部和虚部合并，变为64维度)。经过训练的EN Block将PMI变换到隐式域利于下采样压缩，DE Block将隐式域上的数据变换回PMI所在的空频域。DS-k Block/US-k Block为全连接层实现的上-下采样变换对，并且US-k Block内部级联2比特均匀量化器、DS-k Block内部级联了反量化器，k表示量化后比特数取值为k＝20,40,60,…,320。通过内部级联多对DS-k Block/US-k Block，基础AI模型(EN Block+DE Block) 实现了反馈开销扩展增强。
终端接收到网络配置信道(包括反馈开销，UE也可以自行确定反馈开销)后，根据反馈开销选择DS-k下采样分支进行CSI压缩。类似地，网络设备基于配置给终端的或者终端上报的反馈开销选择对应的US-k上采样分支进行CSI恢复。
实例二(MIMO层可扩展的CSI压缩反馈模型)：
如图9所示，基础AI模型是一个基于Transformer模型的AE，用于12子带、32端口，层1，240bit反馈开销的PMI压缩和解压缩。通过在基础AI模型外侧(编码器的输入侧、解码器的输出侧)级联LPT-k Block-LT-k Block预变换-变换对，基础AI模型可以实现MIMO多层扩展，其中LPT-1Block、LT-1Block均为输入等于输出的恒等变换，LPT-k Block/LT-k Block(k＝2,3,4)是空域-频域的2维线性变换，通过2个全连接层实现。对于第k层的PMI，终端选择第k条支路进行CSI压缩，网络设备则选择第k条支路进行CSI恢复。
实例三(MIMO端口可扩展的CSI压缩反馈模型)：
如图10所示，AI基础模型是一个AE，用于12子带、32端口、层1、240bit反馈开销的PMI压缩和解压。通过外侧级联PPT-k Block/PT-k Block，基础AI模型实现MIMO多端口的扩展。PPT-32Block/PT-32Block为输入等于输出的恒等变换，PPT-k Block/PT-k Block(k＝24,16,12)为全连接层实现的空域线性变换,将输入为Nsubband,AI×(k×2)维度的PMI变换为Nsubband,AI×(32×2)维度的PMI，其中k为实际PMI端口数。
实例四(子带数可扩展的CSI压缩反馈模型)：
AI基础模型是一个基于Transformer模型的AE，用于Nsubband,AI＝12子带、Nport＝32端口、层1、240bit反馈开销的PMI压缩和解压。通过在基础AI模型外侧(编码器的输入侧、解码器的输出侧)级联子带填充模块和子带裁剪模块，基础AI模型实现了子带数的扩展。子带填充模块对于当前输入的Nsubband＝52个子带PMI以12个子带为一组划分成5组，对于最后一组少于12个子带的情况，在频域上循环填充8个子带以实现12个子带AI输入(填充后子带索引号为[1,2,3,4,3,2,1,2,3,4,3,2])。子带裁剪模块将最后一个分 组中对应的循环填充PMI进行裁剪获得原始PMI。
实例五(子带数+端口+层+反馈开销可扩展的CSI压缩反馈模型)：
如图11所示，EN Block和DE Block构成的基础AI网络对12子带、32端口、层1的PMI进行变换和反变换。先通过基础AI网络内侧(编码器的输出侧、解码器的输入侧)级联DS-k/US-k实现反馈开销扩展增强，通过全连接层实现；随后，基础AI网络外侧(编码器的输入侧、解码器的输出侧)级联LPT-k/LT-k实现了4层MIMO扩展，LPT-1/LT-1为恒等变换，LPT-k/LT-k(k＝2,3,4)通过全连接层实现空域和频域二维线性变换实现；在基础AI网络外侧进一步级联PPT-k/PT-k实现{32,24,16,12}端口扩展,PPT-32/PT-32为恒等变换，PPT-k/PT-k(k＝24,16,12)通过全连接层空域线性变换实现；最后，在基础AI模型外侧级联子带周期循环填充和子带反填充实现子带数扩展增强。
实例六(子带数+端口+反馈开销可扩展的CSI压缩反馈模型，Layer-Specified AI模型)：
如图12所示，EN Block和DE Block构成的基础AI模型对12子带、32端口、层1的PMI进行变换和反变换，先通过基础AI网络内侧(编码器的输出侧、解码器的输入侧)级联DS-k/US-k实现反馈开销扩展增强，通过全连接层实现；在基础AI模型外侧(编码器的输入侧、解码器的输出侧)进一步级联PPT-k/PT-k实现{32,24,16,12}端口扩展,PPT-32/PT-32为恒等变换，PPT-k/PT-k(k＝24,16,12)通过全连接层实现空域线性变换实现；最后，在基础AI模型外侧级联子带填充模块和子带裁剪模块实现子带数扩展增强。
图13是本公开实施例提供的一种终端的结构示意图，如图13所示，所述终端包括存储器1320，收发机1300，处理器1310，其中：
存储器1320，用于存储计算机程序；收发机1300，用于在所述处理器1310的控制下收发数据；处理器1310，用于读取所述存储器1320中的计算机程序并执行以下操作：
将信道状态信息输入至编码器模型，得到编码器模型输出的扩展增强的信道状态信息；
将所述扩展增强的信道状态信息发送至网络设备；
其中，所述编码器模型包括编码器，以及与所述编码器级联的信道状态信息扩展增强模块。
其中，在图13中，总线架构可以包括任意数量的互联的总线和桥，具体由处理器1310代表的一个或多个处理器和存储器1320代表的存储器的各种电路链接在一起。总线架构还可以将诸如外围设备、稳压器和功率管理电路等之类的各种其他电路链接在一起，这些都是本领域所公知的，因此，本文不再对其进行进一步描述。总线接口提供接口。收发机1300可以是多个元件，即包括发送机和接收机，提供用于在传输介质上与各种其他装置通信的单元，这些传输介质包括无线信道、有线信道、光缆等传输介质。针对不同的用户设备，用户接口1330还可以是能够外接内接需要设备的接口，连接的设备包括但不限于小键盘、显示器、扬声器、麦克风、操纵杆等。
处理器1310负责管理总线架构和通常的处理，存储器1320可以存储处理器1310在执行操作时所使用的数据。
可选的，处理器1310可以是中央处理器(Central Processing Unit,CPU)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列(Field－Programmable Gate Array，FPGA)或复杂可编程逻辑器件(Complex Programmable Logic Device，CPLD)，处理器也可以采用多核架构。
处理器通过调用存储器存储的计算机程序，用于按照获得的可执行指令执行本公开实施例提供的任一所述方法。处理器与存储器也可以物理上分开布置。
在一个实施例中，所述信道状态信息扩展增强模块包括下采样模块；
所述下采样模块级联于所述编码器的输出侧，用于对所述信道状态信息的反馈开销进行压缩，得到压缩后的反馈开销；
所述扩展增强的信道状态信息包括所述压缩后的反馈开销。
在一个实施例中，所述信道状态信息扩展增强模块包括预变换模块；
所述预变换模块级联于所述编码器的输入侧，用于执行以下至少一项：
调整不同多入多输出MIMO层的信道状态信息的概率分布，得到概率分布和维度统一的不同MIMO层的信道状态信息；
调整不同MIMO端口的信道状态信息的概率分布和/或维度，得到概率分布和维度统一的不同MIMO端口的信道状态信息；
所述扩展增强的信道状态信息包括所述概率分布和维度统一的不同MIMO层和/或不同MIMO端口的信道状态信息。
在一个实施例中，所述信道状态信息扩展增强模块包括子带填充模块；
所述子带填充模块级联于所述编码器的输入侧，用于对所述信道状态信息的子带数进行填充，得到子带填充后的信道状态信息；
所述扩展增强的信道状态信息包括所述子带填充后的信道状态信息。
在一个实施例中，所述下采样模块包括以下任一项配置：
所述下采样模块为池化层；
所述下采样模块为输入维度大于输出维度的全连接层；
所述下采样模块为卷积下采样层。
在一个实施例中，所述下采样模块的输出维度是基于所述反馈开销的种类以及预定的量化比特数确定的。
在一个实施例中，所述预变换模块包括以下任一项配置：
所述预变换模块为恒等变换层；
所述预变换模块在空域和频域分别采用一个全连接层；
所述预变换模块在空域和频域采用一个2维卷积层；
所述预变换模块在空域和频域合并维度，采用一个全连接层；
所述预变换模块在空域和频域合并维度，采用一个1维卷积层；
所述预变换模块在空域采用一个全连接层；
所述预变换模块在空域采用一个1维卷积层。
在一个实施例中，所述子带填充模块具体用于：
在所述信道状态信息的子带数与所述编码器模型的输入子带数的余数大于0的情况下，基于所述编码器模型的输入子带数对所述信道状态信息的子带进行分组；
对所述信道状态信息的子带数小于所述编码器模型的输入子带数的目标分组中的子带进行周期循环填充，直到所述目标分组中的子带数量等于所述 编码器模型的输入子带数；
所述周期循环填充包括：根据分组中各子带的排列顺序，依次复制各子带。
在此需要说明的是，本公开实施例提供的上述终端，能够实现上述执行主体为终端的方法实施例所实现的所有方法步骤，且能够达到相同的技术效果，在此不再对本实施例中与方法实施例相同的部分及有益效果进行具体赘述。
图14是本公开实施例提供的网络设备的结构示意图，如图14所示，所述网络设备包括存储器1420，收发机1400，处理器1410，其中：
存储器1420，用于存储计算机程序；收发机1400，用于在所述处理器1410的控制下收发数据；处理器1410，用于读取所述存储器1420中的计算机程序并执行以下操作：
接收终端发送的扩展增强的信道状态信息；
将所述扩展增强的信道状态信息输入至解码器模型，得到解码器模型输出的还原信道状态信息；
其中，所述解码器模型包括解码器，以及与所述解码器级联的信道状态信息还原模块。
具体地，收发机1400，用于在处理器1410的控制下接收和发送数据。
其中，在图14中，总线架构可以包括任意数量的互联的总线和桥，具体由处理器1410代表的一个或多个处理器和存储器1420代表的存储器的各种电路链接在一起。总线架构还可以将诸如外围设备、稳压器和功率管理电路等之类的各种其他电路链接在一起，这些都是本领域所公知的，因此，本文不再对其进行进一步描述。总线接口提供接口。收发机1400可以是多个元件，即包括发送机和接收机，提供用于在传输介质上与各种其他装置通信的单元，这些传输介质包括无线信道、有线信道、光缆等传输介质。处理器1410负责管理总线架构和通常的处理，存储器1420可以存储处理器1410在执行操作时所使用的数据。
处理器1410可以是中央处理器(Central Processing Unit,CPU)、专用集 成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列(Field－Programmable Gate Array，FPGA)或复杂可编程逻辑器件(Complex Programmable Logic Device，CPLD)，处理器也可以采用多核架构。
在一个实施例中，所述扩展增强的信道状态信息包括压缩后的反馈开销；所述信道状态信息还原模块包括上采样模块；
所述上采样模块级联于所述解码器的输入侧，用于对所述压缩后的反馈开销进行还原，得到还原的反馈开销。
在一个实施例中，所述扩展增强的信道状态信息包括概率分布和维度统一的不同多输入多输出MIMO层和/或不同MIMO端口的信道状态信息；所述信道状态信息还原模块包括变换模块；
所述变换模块级联于所述解码器的输出侧，用于执行以下至少一项：
对所述概率分布和维度统一的不同MIMO层的信道状态信息进行概率分布还原，得到原始概率分布的不同MIMO层的信道状态信息；
对所述概率分布和维度统一的不同MIMO端口的信道状态信息进行概率分布和维度还原，得到原始概率分布和维度的不同MIMO端口的信道状态信息。
在一个实施例中，所述扩展增强的信道状态信息包括子带填充后的信道状态信息；所述信道状态信息还原模块包括子带裁剪模块；
所述子带裁剪模块级联于所述解码器的输出侧，用于对子带填充后的信道状态信息进行裁剪和/或合并，得到原始子带数量的信道状态信息。
在一个实施例中，所述上采样模块包括以下任一项配置：
所述上采样模块为邻近插值层；
所述上采样模块为输入维度小于输出维度的全连接层；
所述上采样模块为转置卷积上采样层。
在一个实施例中，所述上采样模块的输入维度是基于所述反馈开销的种类以及预定的量化比特数确定的。
在一个实施例中，所述变换模块包括以下任一项配置：
所述变换模块为恒等变换层；
所述变换模块在空域和频域分别采用一个全连接层；
所述变换模块在空域和频域采用一个2维转置卷积层；
所述变换模块在空域和频域合并维度，采用一个全连接层；
所述变换模块在空域和频域合并维度，采用一个1维卷积层；
所述变换模块在空域采用一个全连接层；
所述变换模块在空域采用一个1维卷积层。
在一个实施例中，所述子带裁剪模块具体用于：
对进行过子带周期循环填充的子带分组中的子带进行裁剪和/或合并。
在此需要说明的是，本公开实施例提供的上述网络设备，能够实现上述执行主体为网络设备的方法实施例所实现的所有方法步骤，且能够达到相同的技术效果，在此不再对本实施例中与方法实施例相同的部分及有益效果进行具体赘述。
图15是本公开实施例提供的信道状态信息处理装置的结构示意图之一。如图15所示，本公开实施例还提供一种信道状态信息处理装置，应用于终端，所述装置包括：
增强模块1510，用于将信道状态信息输入至编码器模型，得到编码器模型输出的扩展增强的信道状态信息；
发送模块1520，用于将所述扩展增强的信道状态信息发送至网络设备；
其中，所述编码器模型包括编码器，以及与所述编码器级联的信道状态信息扩展增强模块。
在一个实施例中，所述信道状态信息扩展增强模块包括下采样模块；
所述下采样模块级联于所述编码器的输出侧，用于对所述信道状态信息的反馈开销进行压缩，得到压缩后的反馈开销；
所述扩展增强的信道状态信息包括所述压缩后的反馈开销。
在一个实施例中，所述信道状态信息扩展增强模块包括预变换模块；
所述预变换模块级联于所述编码器的输入侧，用于执行以下至少一项：
调整不同多入多输出MIMO层的信道状态信息的概率分布，得到概率分布和维度统一的不同MIMO层的信道状态信息；
调整不同MIMO端口的信道状态信息的概率分布和/或维度，得到概率分布和维度统一的不同MIMO端口的信道状态信息；
所述扩展增强的信道状态信息包括所述概率分布和维度统一的不同MIMO层和/或不同MIMO端口的信道状态信息。
在一个实施例中，所述信道状态信息扩展增强模块包括子带填充模块；
所述子带填充模块级联于所述编码器的输入侧，用于对所述信道状态信息的子带数进行填充，得到子带填充后的信道状态信息；
所述扩展增强的信道状态信息包括所述子带填充后的信道状态信息。
在一个实施例中，所述下采样模块包括以下任一项配置：
所述下采样模块为池化层；
所述下采样模块为输入维度大于输出维度的全连接层；
所述下采样模块为卷积下采样层。
在一个实施例中，所述下采样模块的输出维度是基于所述反馈开销的种类以及预定的量化比特数确定的。
在一个实施例中，所述预变换模块包括以下任一项配置：
所述预变换模块为恒等变换层；
所述预变换模块在空域和频域分别采用一个全连接层；
所述预变换模块在空域和频域采用一个2维转置卷积层；
所述预变换模块在空域和频域合并维度，采用一个全连接层；
所述预变换模块在空域和频域合并维度，采用一个1维卷积层；
所述预变换模块在空域采用一个全连接层；
所述预变换模块在空域采用一个1维卷积层。
在一个实施例中，所述子带填充模块具体用于：
在所述信道状态信息的子带数与所述编码器模型的输入子带数的余数大于0的情况下，基于所述编码器模型的输入子带数对所述信道状态信息的子带进行分组；
对所述信道状态信息的子带数小于所述编码器模型的输入子带数的目标分组中的子带进行周期循环填充，直到所述目标分组中的子带数量等于所述 编码器模型的输入子带数；
所述周期循环填充包括：根据分组中各子带的排列顺序，依次复制各子带。
在此需要说明的是，本公开实施例提供的上述装置，能够实现上述执行主体为终端的方法实施例所实现的所有方法步骤，且能够达到相同的技术效果，在此不再对本实施例中与方法实施例相同的部分及有益效果进行具体赘述。
图16是本公开实施例提供的信道状态信息处理装置的结构示意图之二。如图16所示，本公开实施例还提供一种信道状态信息处理装置，应用于网络设备，所述装置包括：
接收模块1610，用于接收终端发送的扩展增强的信道状态信息；
还原模块1620，用于将所述扩展增强的信道状态信息输入至解码器模型，得到解码器模型输出的还原信道状态信息；
其中，所述解码器模型包括解码器，以及与所述解码器级联的信道状态信息还原模块。
在一个实施例中，所述扩展增强的信道状态信息包括压缩后的反馈开销；所述信道状态信息还原模块包括上采样模块；
所述上采样模块级联于所述解码器的输入侧，用于对所述压缩后的反馈开销进行还原，得到还原的反馈开销。
在一个实施例中，所述扩展增强的信道状态信息包括概率分布和维度统一的不同多输入多输出MIMO层和/或不同MIMO端口的信道状态信息；所述信道状态信息还原模块包括变换模块；
所述变换模块级联于所述解码器的输出侧，用于执行以下至少一项：
对所述概率分布和维度统一的不同MIMO层的信道状态信息进行概率分布还原，得到原始概率分布的不同MIMO层的信道状态信息；
对所述概率分布和维度统一的不同MIMO端口的信道状态信息进行概率分布和维度还原，得到原始概率分布和维度的不同MIMO端口的信道状态信息。
在一个实施例中，所述扩展增强的信道状态信息包括子带填充后的信道状态信息；所述信道状态信息还原模块包括子带裁剪模块；
所述子带裁剪模块级联于所述解码器的输出侧，用于对子带填充后的信道状态信息进行裁剪和/或合并，得到原始子带数量的信道状态信息。
在一个实施例中，所述上采样模块包括以下任一项配置：
所述上采样模块为邻近插值层；
所述上采样模块为输入维度小于输出维度的全连接层；
所述上采样模块为转置卷积上采样层。
在一个实施例中，所述上采样模块的输入维度是基于所述反馈开销的种类以及预定的量化比特数确定的。
在一个实施例中，所述变换模块包括以下任一项配置：
所述变换模块为恒等变换层；
所述变换模块在空域和频域分别采用一个全连接层；
所述变换模块在空域和频域采用一个2维转置卷积层；
所述变换模块在空域和频域合并维度，采用一个全连接层；
所述变换模块在空域和频域合并维度，采用一个1维卷积层；
所述变换模块在空域采用一个全连接层；
所述变换模块在空域采用一个1维卷积层。
在一个实施例中，所述子带裁剪模块具体用于：
对进行过子带周期循环填充的子带分组中的子带进行裁剪和/或合并。
在此需要说明的是，本公开实施例提供的上述装置，能够实现上述执行主体为网络设备的方法实施例所实现的所有方法步骤，且能够达到相同的技术效果，在此不再对本实施例中与方法实施例相同的部分及有益效果进行具体赘述。
需要说明的是，本公开实施例中对单元的划分是示意性的，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。另外，在本公开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。上述集成的单元既 可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
所述集成的单元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，可以存储在一个处理器可读取存储介质中。基于这样的理解，本公开的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分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)或处理器(processor)执行本公开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、只读存储器(Read-Only Memory，ROM)、随机存取存储器(Random Access Memory，RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
另一方面，本公开实施例还提供一种处理器可读存储介质，所述处理器可读存储介质存储有计算机程序，所述计算机程序用于使所述处理器执行上述各实施例提供的方法，例如包括：
将信道状态信息输入至编码器模型，得到编码器模型输出的扩展增强的信道状态信息；
将所述扩展增强的信道状态信息发送至网络设备；
其中，所述编码器模型包括编码器，以及所述编码器级联的信道状态信息扩展增强模块。或者，
接收终端发送的扩展增强的信道状态信息；
将所述扩展增强的信道状态信息输入至解码器模型，得到解码器模型输出的还原信道状态信息；
其中，所述解码器模型包括解码器，以及与所述解码器级联的信道状态信息还原模块。
所述处理器可读存储介质可以是处理器能够存取的任何可用介质或数据存储设备，包括但不限于磁性存储器(例如软盘、硬盘、磁带、磁光盘(MO)等)、光学存储器(例如CD、DVD、BD、HVD等)、以及半导体存储器(例如ROM、EPROM、EEPROM、非易失性存储器(NAND FLASH)、固态硬盘(SSD))等。
本公开实施例提供的技术方案可以适用于多种系统，尤其是5G系统。例如适用的系统可以是全球移动通讯(global system of mobile communication，GSM)系统、码分多址(code division multiple access，CDMA)系统、宽带码分多址(Wideband Code Division Multiple Access，WCDMA)通用分组无线业务(general packet radio service，GPRS)系统、长期演进(long term evolution，LTE)系统、LTE频分双工(frequency division duplex，FDD)系统、LTE时分双工(time division duplex，TDD)系统、高级长期演进(long term evolution advanced，LTE-A)系统、通用移动系统(universal mobile telecommunication system，UMTS)、全球互联微波接入(worldwide interoperability for microwave access，WiMAX)系统、5G新空口(New Radio,NR)系统等。这多种系统中均包括终端设备和网络设备。系统中还可以包括核心网部分，例如演进的分组系统(Evloved Packet System,EPS)、5G系统(5GS)等。
本公开实施例涉及的终端设备，可以是指向用户提供语音和/或数据连通性的设备，具有无线连接功能的手持式设备、或连接到无线调制解调器的其他处理设备等。在不同的系统中，终端设备的名称可能也不相同，例如在5G系统中，终端设备可以称为用户设备(User Equipment，UE)。无线终端设备可以经无线接入网(Radio Access Network,RAN)与一个或多个核心网(Core Network,CN)进行通信，无线终端设备可以是移动终端设备，如移动电话(或称为“蜂窝”电话)和具有移动终端设备的计算机，例如，可以是便携式、袖珍式、手持式、计算机内置的或者车载的移动装置，它们与无线接入网交换语言和/或数据。例如，个人通信业务(Personal Communication Service，PCS)电话、无绳电话、会话发起协议(Session Initiated Protocol，SIP)话机、无线本地环路(Wireless Local Loop，WLL)站、个人数字助理(Personal Digital Assistant，PDA)等设备。无线终端设备也可以称为系统、订户单元(subscriber unit)、订户站(subscriber station)，移动站(mobile station)、移动台(mobile)、远程站(remote station)、接入点(access point)、远程终端设备(remote terminal)、接入终端设备(access terminal)、用户终端设备(user terminal)、用户代理(user agent)、用户装置(user device)，本公开实施例中并不限定。
本公开实施例涉及的网络设备，可以是基站，该基站可以包括多个为终端提供服务的小区。根据具体应用场合不同，基站又可以称为接入点，或者可以是接入网中在空中接口上通过一个或多个扇区与无线终端设备通信的设备，或者其它名称。网络设备可用于将收到的空中帧与网际协议(Internet Protocol，IP)分组进行相互更换，作为无线终端设备与接入网的其余部分之间的路由器，其中接入网的其余部分可包括网际协议(IP)通信网络。网络设备还可协调对空中接口的属性管理。例如，本公开实施例涉及的网络设备可以是全球移动通信系统(Global System for Mobile communications，GSM)或码分多址接入(Code Division Multiple Access，CDMA)中的网络设备(Base Transceiver Station，BTS)，也可以是带宽码分多址接入(Wide-band Code Division Multiple Access，WCDMA)中的网络设备(NodeB)，还可以是长期演进(long term evolution，LTE)系统中的演进型网络设备(evolutional Node B，eNB或e-NodeB)、5G网络架构(next generation system)中的5G基站(gNB)，也可以是家庭演进基站(Home evolved Node B，HeNB)、中继节点(relay node)、家庭基站(femto)、微微基站(pico)等，本公开实施例中并不限定。在一些网络结构中，网络设备可以包括集中单元(centralized unit，CU)节点和分布单元(distributed unit，DU)节点，集中单元和分布单元也可以地理上分开布置。
网络设备与终端设备之间可以各自使用一或多根天线进行多输入多输出(Multi Input Multi Output,MIMO)传输，MIMO传输可以是单用户MIMO(Single User MIMO,SU-MIMO)或多用户MIMO(Multiple User MIMO,MU-MIMO)。根据根天线组合的形态和数量，MIMO传输可以是2D-MIMO、3D-MIMO、FD-MIMO或massive-MIMO，也可以是分集传输或预编码传输或波束赋形传输等。
本领域内的技术人员应明白，本公开的实施例可提供为方法、系统、或计算机程序产品。因此，本公开可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本公开可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存 储器和光学存储器等)上实施的计算机程序产品的形式。
本公开是参照根据本公开实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机可执行指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机可执行指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些处理器可执行指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的处理器可读存储器中，使得存储在该处理器可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些处理器可执行指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
显然，本领域的技术人员可以对本公开进行各种改动和变型而不脱离本公开的精神和范围。这样，倘若本公开的这些修改和变型属于本公开权利要求及其等同技术的范围之内，则本公开也意图包含这些改动和变型在内。

Claims (49)

  1. 一种信道状态信息处理方法，应用于终端，所述方法包括：
    将信道状态信息输入至编码器模型，得到所述编码器模型输出的扩展增强的信道状态信息；
    将所述扩展增强的信道状态信息发送至网络设备；
    其中，所述编码器模型包括编码器，以及与所述编码器级联的信道状态信息扩展增强模块。
  2. 根据权利要求1所述的信道状态信息处理方法，其中，所述信道状态信息扩展增强模块包括下采样模块；
    所述下采样模块级联于所述编码器的输出侧，用于对所述信道状态信息的反馈开销进行压缩，得到压缩后的反馈开销；
    所述扩展增强的信道状态信息包括所述压缩后的反馈开销。
  3. 根据权利要求1所述的信道状态信息处理方法，其中，所述信道状态信息扩展增强模块包括预变换模块；
    所述预变换模块级联于所述编码器的输入侧，用于执行以下至少一项：
    调整不同多入多输出MIMO层的信道状态信息的概率分布，得到概率分布和维度统一的不同MIMO层的信道状态信息；
    调整不同MIMO端口的信道状态信息的概率分布和/或维度，得到概率分布和维度统一的不同MIMO端口的信道状态信息；
    所述扩展增强的信道状态信息包括所述概率分布和维度统一的不同MIMO层和/或不同MIMO端口的信道状态信息。
  4. 根据权利要求1所述的信道状态信息处理方法，其中，所述信道状态信息扩展增强模块包括子带填充模块；
    所述子带填充模块级联于所述编码器的输入侧，用于对所述信道状态信息的子带数进行填充，得到子带填充后的信道状态信息；
    所述扩展增强的信道状态信息包括所述子带填充后的信道状态信息。
  5. 根据权利要求2所述的信道状态信息处理方法，其中，所述下采样模块包括以下任一项配置：
    所述下采样模块为池化层；
    所述下采样模块为输入维度大于输出维度的全连接层；
    所述下采样模块为卷积下采样层。
  6. 根据权利要求5所述的信道状态信息处理方法，其中，所述下采样模块的输出维度是基于所述反馈开销的种类以及预定的量化比特数确定的。
  7. 根据权利要求3所述的信道状态信息处理方法，其中，所述预变换模块包括以下任一项配置：
    所述预变换模块为恒等变换层；
    所述预变换模块在空域和频域分别采用一个全连接层；
    所述预变换模块在空域和频域采用一个2维卷积层；
    所述预变换模块在空域和频域合并维度，采用一个全连接层；
    所述预变换模块在空域和频域合并维度，采用一个1维卷积层；
    所述预变换模块在空域采用一个全连接层；
    所述预变换模块在空域采用一个1维卷积层。
  8. 根据权利要求4所述的信道状态信息处理方法，其中，所述子带填充模块具体用于：
    在所述信道状态信息的子带数与所述编码器模型的输入子带数的余数大于0的情况下，基于所述编码器模型的输入子带数对所述信道状态信息的子带进行分组；
    对所述信道状态信息的子带数小于所述编码器模型的输入子带数的目标分组中的子带进行周期循环填充，直到所述目标分组中的子带数量等于所述编码器模型的输入子带数；
    所述周期循环填充包括：根据分组中各子带的排列顺序，依次复制各子带。
  9. 一种信道状态信息处理方法，应用于网络设备，所述方法包括：
    接收终端发送的扩展增强的信道状态信息；
    将所述扩展增强的信道状态信息输入至解码器模型，得到所述解码器模型输出的还原信道状态信息；
    其中，所述解码器模型包括解码器，以及与所述解码器级联的信道状态信息还原模块。
  10. 根据权利要求9所述的信道状态信息处理方法，其中，所述扩展增强的信道状态信息包括压缩后的反馈开销；所述信道状态信息还原模块包括上采样模块；
    所述上采样模块级联于所述解码器的输入侧，用于对所述压缩后的反馈开销进行还原，得到还原的反馈开销。
  11. 根据权利要求9所述的信道状态信息处理方法，其中，所述扩展增强的信道状态信息包括概率分布和维度统一的不同多输入多输出MIMO层和/或不同MIMO端口的信道状态信息；所述信道状态信息还原模块包括变换模块；
    所述变换模块级联于所述解码器的输出侧，用于执行以下至少一项：
    对所述概率分布和维度统一的不同MIMO层的信道状态信息进行概率分布还原，得到原始概率分布的不同MIMO层的信道状态信息；
    对所述概率分布和维度统一的不同MIMO端口的信道状态信息进行概率分布和维度还原，得到原始概率分布和维度的不同MIMO端口的信道状态信息。
  12. 根据权利要求9所述的信道状态信息处理方法，其中，所述扩展增强的信道状态信息包括子带填充后的信道状态信息；所述信道状态信息还原模块包括子带裁剪模块；
    所述子带裁剪模块级联于所述解码器的输出侧，用于对子带填充后的信道状态信息进行裁剪和/或合并，得到原始子带数量的信道状态信息。
  13. 根据权利要求10所述的信道状态信息处理方法，其中，所述上采样模块包括以下任一项配置：
    所述上采样模块为邻近插值层；
    所述上采样模块为输入维度小于输出维度的全连接层；
    所述上采样模块为转置卷积上采样层。
  14. 根据权利要求13所述的信道状态信息处理方法，其中，所述上采样 模块的输入维度是基于所述反馈开销的种类以及预定的量化比特数确定的。
  15. 根据权利要求11所述的信道状态信息处理方法，其中，所述变换模块包括以下任一项配置：
    所述变换模块为恒等变换层；
    所述变换模块在空域和频域分别采用一个全连接层；
    所述变换模块在空域和频域采用一个2维转置卷积层；
    所述变换模块在空域和频域合并维度，采用一个全连接层；
    所述变换模块在空域和频域合并维度，采用一个1维卷积层；
    所述变换模块在空域采用一个全连接层；
    所述变换模块在空域采用一个1维卷积层。
  16. 根据权利要求12所述的信道状态信息处理方法，其中，所述子带裁剪模块具体用于：
    对进行过子带周期循环填充的子带分组中的子带进行裁剪和/或合并。
  17. 一种终端，包括存储器，收发机，处理器：
    存储器，用于存储计算机程序；收发机，用于在所述处理器的控制下收发数据；处理器，用于读取所述存储器中的计算机程序并执行以下操作：
    将信道状态信息输入至编码器模型，得到所述编码器模型输出的扩展增强的信道状态信息；
    将所述扩展增强的信道状态信息发送至网络设备；
    其中，所述编码器模型包括编码器，以及与所述编码器级联的信道状态信息扩展增强模块。
  18. 根据权利要求17所述的终端，其中，所述信道状态信息扩展增强模块包括下采样模块；
    所述下采样模块级联于所述编码器的输出侧，用于对所述信道状态信息的反馈开销进行压缩，得到压缩后的反馈开销；
    所述扩展增强的信道状态信息包括所述压缩后的反馈开销。
  19. 根据权利要求17所述的终端，其中，所述信道状态信息扩展增强模块包括预变换模块；
    所述预变换模块级联于所述编码器的输入侧，用于执行以下至少一项：
    调整不同多入多输出MIMO层的信道状态信息的概率分布，得到概率分布和维度统一的不同MIMO层的信道状态信息；
    调整不同MIMO端口的信道状态信息的概率分布和/或维度，得到概率分布和维度统一的不同MIMO端口的信道状态信息；
    所述扩展增强的信道状态信息包括所述概率分布和维度统一的不同MIMO层和/或不同MIMO端口的信道状态信息。
  20. 根据权利要求17所述的终端，其中，所述信道状态信息扩展增强模块包括子带填充模块；
    所述子带填充模块级联于所述编码器的输入侧，用于对所述信道状态信息的子带数进行填充，得到子带填充后的信道状态信息；
    所述扩展增强的信道状态信息包括所述子带填充后的信道状态信息。
  21. 根据权利要求18所述的终端，其中，所述下采样模块包括以下任一项配置：
    所述下采样模块为池化层；
    所述下采样模块为输入维度大于输出维度的全连接层；
    所述下采样模块为卷积下采样层。
  22. 根据权利要求21所述的终端，其中，所述下采样模块的输出维度是基于所述反馈开销的种类以及预定的量化比特数确定的。
  23. 根据权利要求19所述的终端，其中，所述预变换模块包括以下任一项配置：
    所述预变换模块为恒等变换层；
    所述预变换模块在空域和频域分别采用一个全连接层；
    所述预变换模块在空域和频域采用一个2维卷积层；
    所述预变换模块在空域和频域合并维度，采用一个全连接层；
    所述预变换模块在空域和频域合并维度，采用一个1维卷积层；
    所述预变换模块在空域采用一个全连接层；
    所述预变换模块在空域采用一个1维卷积层。
  24. 根据权利要求20所述的终端，其中，所述子带填充模块具体用于：
    在所述信道状态信息的子带数与所述编码器模型的输入子带数的余数大于0的情况下，基于所述编码器模型的输入子带数对所述信道状态信息的子带进行分组；
    对所述信道状态信息的子带数小于所述编码器模型的输入子带数的目标分组中的子带进行周期循环填充，直到所述目标分组中的子带数量等于所述编码器模型的输入子带数；
    所述周期循环填充包括：根据分组中各子带的排列顺序，依次复制各子带。
  25. 一种网络设备，包括存储器，收发机，处理器：
    存储器，用于存储计算机程序；收发机，用于在所述处理器的控制下收发数据；处理器，用于读取所述存储器中的计算机程序并执行以下操作：
    接收终端发送的扩展增强的信道状态信息；
    将所述扩展增强的信道状态信息输入至解码器模型，得到所述解码器模型输出的还原信道状态信息；
    其中，所述解码器模型包括解码器，以及与所述解码器级联的信道状态信息还原模块。
  26. 根据权利要求25所述的网络设备，其中，所述扩展增强的信道状态信息包括压缩后的反馈开销；所述信道状态信息还原模块包括上采样模块；
    所述上采样模块级联于所述解码器的输入侧，用于对所述压缩后的反馈开销进行还原，得到还原的反馈开销。
  27. 根据权利要求25所述的网络设备，其中，所述扩展增强的信道状态信息包括概率分布和维度统一的不同多输入多输出MIMO层和/或不同MIMO端口的信道状态信息；所述信道状态信息还原模块包括变换模块；
    所述变换模块级联于所述解码器的输出侧，用于执行以下至少一项：
    对所述概率分布和维度统一的不同MIMO层的信道状态信息进行概率分布还原，得到原始概率分布的不同MIMO层的信道状态信息；
    对所述概率分布和维度统一的不同MIMO端口的信道状态信息进行概率 分布和维度还原，得到原始概率分布和维度的不同MIMO端口的信道状态信息。
  28. 根据权利要求25所述的网络设备，其中，所述扩展增强的信道状态信息包括子带填充后的信道状态信息；所述信道状态信息还原模块包括子带裁剪模块；
    所述子带裁剪模块级联于所述解码器的输出侧，用于对子带填充后的信道状态信息进行裁剪和/或合并，得到原始子带数量的信道状态信息。
  29. 根据权利要求26所述的网络设备，其中，所述上采样模块包括以下任一项配置：
    所述上采样模块为邻近插值层；
    所述上采样模块为输入维度小于输出维度的全连接层；
    所述上采样模块为转置卷积上采样层。
  30. 根据权利要求29所述的网络设备，其中，所述上采样模块的输入维度是基于所述反馈开销的种类以及预定的量化比特数确定的。
  31. 根据权利要求27所述的网络设备，其中，所述变换模块包括以下任一项配置：
    所述变换模块为恒等变换层；
    所述变换模块在空域和频域分别采用一个全连接层；
    所述变换模块在空域和频域采用一个2维转置卷积层；
    所述变换模块在空域和频域合并维度，采用一个全连接层；
    所述变换模块在空域和频域合并维度，采用一个1维卷积层；
    所述变换模块在空域采用一个全连接层；
    所述变换模块在空域采用一个1维卷积层。
  32. 根据权利要求28所述的信道状态信息处理方法，其中，所述子带裁剪模块具体用于：
    对进行过子带周期循环填充的子带分组中的子带进行裁剪和/或合并。
  33. 一种信道状态信息处理装置，所述装置包括：
    增强模块，用于将信道状态信息输入至编码器模型，得到所述编码器模 型输出的扩展增强的信道状态信息；
    发送模块，用于将所述扩展增强的信道状态信息发送至网络设备；
    其中，所述编码器模型包括编码器，以及与所述编码器级联的信道状态信息扩展增强模块。
  34. 根据权利要求33所述的信道状态信息处理装置，其中，所述信道状态信息扩展增强模块包括下采样模块；
    所述下采样模块级联于所述编码器的输出侧，用于对所述信道状态信息的反馈开销进行压缩，得到压缩后的反馈开销；
    所述扩展增强的信道状态信息包括所述压缩后的反馈开销。
  35. 根据权利要求33所述的信道状态信息处理装置，其中，所述信道状态信息扩展增强模块包括预变换模块；
    所述预变换模块级联于所述编码器的输入侧，用于执行以下至少一项：
    调整不同多入多输出MIMO层的信道状态信息的概率分布，得到概率分布和维度统一的不同MIMO层的信道状态信息；
    调整不同MIMO端口的信道状态信息的概率分布和/或维度，得到概率分布和维度统一的不同MIMO端口的信道状态信息；
    所述扩展增强的信道状态信息包括所述概率分布和维度统一的不同MIMO层和/或不同MIMO端口的信道状态信息。
  36. 根据权利要求33所述的信道状态信息处理装置，其中，所述信道状态信息扩展增强模块包括子带填充模块；
    所述子带填充模块级联于所述编码器的输入侧，用于对所述信道状态信息的子带数进行填充，得到子带填充后的信道状态信息；
    所述扩展增强的信道状态信息包括所述子带填充后的信道状态信息。
  37. 根据权利要求34所述的信道状态信息处理装置，其中，所述下采样模块包括以下任一项配置：
    所述下采样模块为池化层；
    所述下采样模块为输入维度大于输出维度的全连接层；
    所述下采样模块为卷积下采样层。
  38. 根据权利要求37所述的信道状态信息处理装置，其中，所述下采样模块的输出维度是基于所述反馈开销的种类以及预定的量化比特数确定的。
  39. 根据权利要求35所述的信道状态信息处理装置，其中，所述预变换模块包括以下任一项配置：
    所述预变换模块为恒等变换层；
    所述预变换模块在空域和频域分别采用一个全连接层；
    所述预变换模块在空域和频域采用一个2维卷积层；
    所述预变换模块在空域和频域合并维度，采用一个全连接层；
    所述预变换模块在空域和频域合并维度，采用一个1维卷积层；
    所述预变换模块在空域采用一个全连接层；
    所述预变换模块在空域采用一个1维卷积层。
  40. 根据权利要求36所述的信道状态信息处理装置，其中，所述子带填充模块具体用于：
    在所述信道状态信息的子带数与所述编码器模型的输入子带数的余数大于0的情况下，基于所述编码器模型的输入子带数对所述信道状态信息的子带进行分组；
    对所述信道状态信息的子带数小于所述编码器模型的输入子带数的目标分组中的子带进行周期循环填充，直到所述目标分组中的子带数量等于所述编码器模型的输入子带数；
    所述周期循环填充包括：根据分组中各子带的排列顺序，依次复制各子带。
  41. 一种信道状态信息处理装置，所述装置包括：
    接收模块，用于接收终端发送的扩展增强的信道状态信息；
    还原模块，用于将所述扩展增强的信道状态信息输入至解码器模型，得到所述解码器模型输出的还原信道状态信息；
    其中，所述解码器模型包括解码器，以及与所述解码器级联的信道状态信息还原模块。
  42. 根据权利要求41所述的信道状态信息处理装置，其中，所述扩展增 强的信道状态信息包括压缩后的反馈开销；所述信道状态信息还原模块包括上采样模块；
    所述上采样模块级联于所述解码器的输入侧，用于对所述压缩后的反馈开销进行还原，得到还原的反馈开销。
  43. 根据权利要求41所述的信道状态信息处理装置，其中，所述扩展增强的信道状态信息包括概率分布和维度统一的不同多输入多输出MIMO层和/或不同MIMO端口的信道状态信息；所述信道状态信息还原模块包括变换模块；
    所述变换模块级联于所述解码器的输出侧，用于执行以下至少一项：
    对所述概率分布和维度统一的不同MIMO层的信道状态信息进行概率分布还原，得到原始概率分布的不同MIMO层的信道状态信息；
    对所述概率分布和维度统一的不同MIMO端口的信道状态信息进行概率分布和维度还原，得到原始概率分布和维度的不同MIMO端口的信道状态信息。
  44. 根据权利要求41所述的信道状态信息处理装置，其中，所述扩展增强的信道状态信息包括子带填充后的信道状态信息；所述信道状态信息还原模块包括子带裁剪模块；
    所述子带裁剪模块级联于所述解码器的输出侧，用于对子带填充后的信道状态信息进行裁剪和/或合并，得到原始子带数量的信道状态信息。
  45. 根据权利要求42所述的信道状态信息处理装置，其中，所述上采样模块包括以下任一项配置：
    所述上采样模块为邻近插值层；
    所述上采样模块为输入维度小于输出维度的全连接层；
    所述上采样模块为转置卷积上采样层。
  46. 根据权利要求45所述的信道状态信息处理装置，其中，所述上采样模块的输入维度是基于所述反馈开销的种类以及预定的量化比特数确定的。
  47. 根据权利要求43所述的信道状态信息处理装置，其中，所述变换模块包括以下任一项配置：
    所述变换模块为恒等变换层；
    所述变换模块在空域和频域分别采用一个全连接层；
    所述变换模块在空域和频域采用一个2维转置卷积层；
    所述变换模块在空域和频域合并维度，采用一个全连接层；
    所述变换模块在空域和频域合并维度，采用一个1维卷积层；
    所述变换模块在空域采用一个全连接层；
    所述变换模块在空域采用一个1维卷积层。
  48. 根据权利要求44所述的信道状态信息处理装置，其中，所述子带裁剪模块具体用于：
    对进行过子带周期循环填充的子带分组中的子带进行裁剪和/或合并。
  49. 一种处理器可读存储介质，所述处理器可读存储介质存储有计算机程序，所述计算机程序用于使所述处理器执行权利要求1至8任一项所述的方法，或者执行权利要求9至16任一项所述的方法。
PCT/CN2023/112162 2022-08-12 2023-08-10 信道状态信息处理方法及装置 Ceased WO2024032701A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US19/103,210 US20260051989A1 (en) 2022-08-12 2023-08-10 Channel state information processing method and apparatus
EP23851922.7A EP4572171A4 (en) 2022-08-12 2023-08-10 METHOD AND APPARATUS FOR PROCESSING CHANNEL STATE INFORMATION

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210969110.3 2022-08-12
CN202210969110.3A CN117674926A (zh) 2022-08-12 2022-08-12 信道状态信息处理方法及装置

Publications (1)

Publication Number Publication Date
WO2024032701A1 true WO2024032701A1 (zh) 2024-02-15

Family

ID=89850947

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/112162 Ceased WO2024032701A1 (zh) 2022-08-12 2023-08-10 信道状态信息处理方法及装置

Country Status (5)

Country Link
US (1) US20260051989A1 (zh)
EP (1) EP4572171A4 (zh)
CN (1) CN117674926A (zh)
TW (1) TWI856775B (zh)
WO (1) WO2024032701A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2025247079A1 (zh) * 2024-05-28 2025-12-04 华为技术有限公司 通信方法和装置

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US12615189B2 (en) * 2024-09-18 2026-04-28 Interdigital Patent Holdings, Inc. Systems and methods for statistics based and topology based interoperable AI/ML model monitoring

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109672464A (zh) * 2018-12-13 2019-04-23 西安电子科技大学 基于fcfnn的大规模mimo信道状态信息反馈方法
US20210319286A1 (en) * 2018-08-15 2021-10-14 Imperial College Of Science, Technology And Medicine Joint source channel coding for noisy channels using neural networks
CN114172765A (zh) * 2021-12-03 2022-03-11 中国信息通信研究院 一种无线通信人工智能信道估计方法和设备
CN114501353A (zh) * 2020-10-23 2022-05-13 维沃移动通信有限公司 通信信息的发送、接收方法及通信设备
CN114884549A (zh) * 2022-05-06 2022-08-09 烟台中科网络技术研究所 基于深度学习的大规模mimo信道状态信息反馈方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114070374B (zh) * 2018-11-02 2024-04-16 中兴通讯股份有限公司 一种csi反馈及接收方法、装置、存储介质
CN110350958B (zh) * 2019-06-13 2021-03-16 东南大学 一种基于神经网络的大规模mimo的csi多倍率压缩反馈方法
WO2021108940A1 (en) * 2019-12-01 2021-06-10 Nokia Shanghai Bell Co., Ltd. Channel state information feedback
US11936452B2 (en) * 2020-02-28 2024-03-19 Qualcomm Incorporated Neural network based channel state information feedback
CN113660058B (zh) * 2021-07-05 2024-01-23 广州大学 一种信道状态的信息反馈方法、装置、电子设备及介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210319286A1 (en) * 2018-08-15 2021-10-14 Imperial College Of Science, Technology And Medicine Joint source channel coding for noisy channels using neural networks
CN109672464A (zh) * 2018-12-13 2019-04-23 西安电子科技大学 基于fcfnn的大规模mimo信道状态信息反馈方法
CN114501353A (zh) * 2020-10-23 2022-05-13 维沃移动通信有限公司 通信信息的发送、接收方法及通信设备
CN114172765A (zh) * 2021-12-03 2022-03-11 中国信息通信研究院 一种无线通信人工智能信道估计方法和设备
CN114884549A (zh) * 2022-05-06 2022-08-09 烟台中科网络技术研究所 基于深度学习的大规模mimo信道状态信息反馈方法

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
LIQIANG JIN; QIUPING HUANG; QIUBIN GAO; YONGQIANG FEI; SHAOHUI SUN: "Scalable Framework For Deep Learning based CSI Feedback", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 18 October 2022 (2022-10-18), 201 Olin Library Cornell University Ithaca, NY 14853, XP091347113 *
MIN ZHU, CATT: "Evaluation on AI/ML for CSI feedback enhancement", 3GPP DRAFT; R1-2300671; TYPE DISCUSSION; FS_NR_AIML_AIR, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. 3GPP RAN 1, no. Athens, GR; 20230227 - 20230303, 17 February 2023 (2023-02-17), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France, XP052247817 *
See also references of EP4572171A4 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2025247079A1 (zh) * 2024-05-28 2025-12-04 华为技术有限公司 通信方法和装置

Also Published As

Publication number Publication date
EP4572171A4 (en) 2026-01-07
US20260051989A1 (en) 2026-02-19
CN117674926A (zh) 2024-03-08
TW202408195A (zh) 2024-02-16
TWI856775B (zh) 2024-09-21
EP4572171A1 (en) 2025-06-18

Similar Documents

Publication Publication Date Title
US20250070910A1 (en) Information processing method and apparatus, terminal, and network device
CN110350958B (zh) 一种基于神经网络的大规模mimo的csi多倍率压缩反馈方法
TWI856775B (zh) 通道狀態資訊處理方法及裝置
US20240030980A1 (en) Data transmission method and apparatus
CN118120156A (zh) 信道信息反馈模型的训练方法、装置、设备及存储介质
WO2023202385A1 (zh) 信息传输方法、装置及存储介质
JP7743634B2 (ja) ニューラルネットワーク操作方法、装置及び記憶媒体
JP7772831B2 (ja) コードブック指示方法、装置及び記憶媒体
WO2023174108A1 (zh) 一种通信方法及装置
US20250240075A1 (en) Channel information feedback method, transmitting end device, and receiving end device
WO2024207498A1 (en) Channel state information feedback
Aguerri et al. In-network compression for multiterminal cascade MIMO systems
WO2024055910A1 (zh) 图样映射优化方法及装置
Ibrahim et al. Low Complexity Beam Domain Processing for Autoencoder Based CSI Compression
KR20240121644A (ko) 통신 시스템에서 채널 상태 정보 피드백 방법 및 장치
WO2025213896A1 (zh) 信道状态信息压缩方法、重构方法和电子设备
CN116366401A (zh) 神经网络操作方法、装置及存储介质
WO2025082347A1 (zh) 一种被用于无线通信的节点中的方法和装置
WO2025035429A1 (zh) 信息传输方法及装置
WO2026031920A1 (zh) 信道状态传输、获取方法、装置、终端及网络设备
JP2026506085A (ja) 漸進的成長オートエンコーダ
CN116264474A (zh) 波束赋形方法、网络设备、装置及存储介质
CN121485737A (zh) 一种面向6g的基于深度学习轻量化csi反馈方法
CN120639127A (zh) 生成码本的方法、装置及设备

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: 23851922

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 2023851922

Country of ref document: EP

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2023851922

Country of ref document: EP

Effective date: 20250312

WWP Wipo information: published in national office

Ref document number: 2023851922

Country of ref document: EP