WO2022179402A1 - 信息编码的控制方法及相关装置 - Google Patents

信息编码的控制方法及相关装置 Download PDF

Info

Publication number
WO2022179402A1
WO2022179402A1 PCT/CN2022/075987 CN2022075987W WO2022179402A1 WO 2022179402 A1 WO2022179402 A1 WO 2022179402A1 CN 2022075987 W CN2022075987 W CN 2022075987W WO 2022179402 A1 WO2022179402 A1 WO 2022179402A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
encoder
terminal
network device
encoders
Prior art date
Application number
PCT/CN2022/075987
Other languages
English (en)
French (fr)
Inventor
王四海
李雪茹
秦城
杨锐
Original Assignee
华为技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Priority to US18/547,449 priority Critical patent/US20240146582A1/en
Priority to EP22758767.2A priority patent/EP4283894A1/en
Publication of WO2022179402A1 publication Critical patent/WO2022179402A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L1/00—Arrangements for detecting or preventing errors in the information received
    • H04L1/0001—Systems modifying transmission characteristics according to link quality, e.g. power backoff
    • H04L1/0009—Systems modifying transmission characteristics according to link quality, e.g. power backoff by adapting the channel coding
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/0202—Channel estimation
    • H04L25/024—Channel estimation channel estimation algorithms
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
    • G10L19/04—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using predictive techniques
    • G10L19/16—Vocoder architecture
    • H—ELECTRICITY
    • H03—ELECTRONIC CIRCUITRY
    • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
    • H03M7/00—Conversion of a code where information is represented by a given sequence or number of digits to a code where the same, similar or subset of information is represented by a different sequence or number of digits
    • H03M7/30—Compression; Expansion; Suppression of unnecessary data, e.g. redundancy reduction
    • H03M7/3068—Precoding preceding compression, e.g. Burrows-Wheeler transformation
    • H03M7/3079—Context modeling
    • H—ELECTRICITY
    • H03—ELECTRONIC CIRCUITRY
    • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
    • H03M7/00—Conversion of a code where information is represented by a given sequence or number of digits to a code where the same, similar or subset of information is represented by a different sequence or number of digits
    • H03M7/30—Compression; Expansion; Suppression of unnecessary data, e.g. redundancy reduction
    • H03M7/60—General implementation details not specific to a particular type of compression
    • H03M7/6064—Selection of Compressor
    • H03M7/6076—Selection between compressors of the same type
    • H—ELECTRICITY
    • H03—ELECTRONIC CIRCUITRY
    • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
    • H03M7/00—Conversion of a code where information is represented by a given sequence or number of digits to a code where the same, similar or subset of information is represented by a different sequence or number of digits
    • H03M7/30—Compression; Expansion; Suppression of unnecessary data, e.g. redundancy reduction
    • H03M7/60—General implementation details not specific to a particular type of compression
    • H03M7/6064—Selection of Compressor
    • H03M7/6082—Selection strategies
    • 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
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L1/00—Arrangements for detecting or preventing errors in the information received
    • H04L1/0001—Systems modifying transmission characteristics according to link quality, e.g. power backoff
    • H04L1/0023—Systems modifying transmission characteristics according to link quality, e.g. power backoff characterised by the signalling
    • H04L1/0026—Transmission of channel quality indication
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L1/00—Arrangements for detecting or preventing errors in the information received
    • H04L1/0001—Systems modifying transmission characteristics according to link quality, e.g. power backoff
    • H04L1/0023—Systems modifying transmission characteristics according to link quality, e.g. power backoff characterised by the signalling
    • H04L1/0028—Formatting
    • H04L1/0029—Reduction of the amount of signalling, e.g. retention of useful signalling or differential signalling
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/16—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks using machine learning or artificial intelligence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/096—Transfer learning
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L1/00—Arrangements for detecting or preventing errors in the information received
    • H04L1/0001—Systems modifying transmission characteristics according to link quality, e.g. power backoff
    • H04L1/0015—Systems modifying transmission characteristics according to link quality, e.g. power backoff characterised by the adaptation strategy
    • H04L1/0017—Systems modifying transmission characteristics according to link quality, e.g. power backoff characterised by the adaptation strategy where the mode-switching is based on Quality of Service requirement
    • H04L1/0018—Systems modifying transmission characteristics according to link quality, e.g. power backoff characterised by the adaptation strategy where the mode-switching is based on Quality of Service requirement based on latency requirement

Definitions

  • the present application relates to the field of communication technologies, and in particular, to an information encoding control method and a related device.
  • the two devices may be a sending device (eg, a base station) that sends a signal, and a receiving device (eg, a terminal) that receives a signal.
  • the receiving device can feed back channel state information (CSI) to the sending device, and the sending device can precode the wireless signal to be transmitted in the multi-antenna system based on the CSI.
  • CSI channel state information
  • the precoded wireless signal can resist channel distortion. Increase channel capacity.
  • the above-mentioned measurement information, status information, etc. are usually relatively large, so encoding or compression is often required before transmission, thereby saving air interface resources and transmission overhead.
  • the base station can send an artificial intelligence (AI) model (AI encoder for short) for information encoding to the terminal, and the terminal can use the AI encoder to encode the information to be transmitted (such as the above measurement information, status information), and encode the information.
  • AI artificial intelligence
  • the encoded information is fed back to the base station, and the base station obtains the information reported by the terminal (for example, the above-mentioned measurement information and status information) by decoding the AI decoder corresponding to the AI encoder.
  • AI artificial intelligence
  • AI encoder for short
  • the encoded information is fed back to the base station, and the base station obtains the information reported by the terminal (for example, the above-mentioned measurement information and status information) by decoding the AI decoder corresponding to the AI encoder.
  • the information to be transmitted and the data used to train the AI encoder and the corresponding AI decoder are not identically distributed, then the information to be transmitted will pass through the AI Distortion may occur after encoding and AI decoding, which in turn affects communication system performance.
  • the embodiments of the present application disclose an information encoding control method and a related device, which can quickly and efficiently select an appropriate information encoding scheme, avoid distortion of the information to be transmitted after encoding and decoding, thereby avoiding affecting the performance of the communication system, and improving transmission quality.
  • an embodiment of the present application provides a method for controlling information encoding, which is applied to a terminal.
  • the method includes: receiving first configuration information, where the first configuration information is used to configure N groups of parameters of N AI encoders, and N is a positive integer greater than 1; the first indication information is sent to the network device, the first indication information is used to indicate that the first encoder is used for encoding the first information, and the first encoder is determined according to the N groups of parameters and the first information.
  • An encoder is an encoder of the N AI encoders, or the first encoder is a second encoder different from the N AI encoders.
  • the N groups of parameters of the N AI encoders may be replaced with: N groups of parameters of the N AI decoders, or may be replaced with N groups of parameters of the N AI codecs.
  • the N AI encoders correspond to N AI decoders respectively, and an AI encoder includes an AI encoder and a corresponding AI decoder.
  • the second encoder is an encoder based on a traditional encoding scheme such as a codebook, or another AI encoder with better generalization.
  • the first encoder used for encoding the first information is determined based on N groups of parameters, and the first encoder is an AI encoder among the N AI encoders or a second AI encoder other than the N AI encoders.
  • Encoder it can be understood that the first encoder used for encoding the first information is judged to be suitable for encoding the first information, so as to avoid passing the AI codec (or AI encoder and The problem of information distortion caused by the encoding and decoding of the first information by an AI decoder) and uncontrollable performance deterioration of the communication system.
  • the method further includes: sending second information, where the second information is determined based on encoding of the first information by the first encoder.
  • the first encoder is determined according to the relationship between the first judgment parameter and the first judgment threshold, the first judgment parameter is determined according to the N groups of parameters and the first information, and the first judgment parameter is Cosine similarity CS, probability density function PDF, probability mass function PMF or Euclidean distance.
  • the first judgment parameter is determined according to the N groups of parameters, the first information, and the first preprocessing, and the first preprocessing includes at least one of the following: translation, scaling, Fourier transform, compressed sensing transform, truncation, AI model processing, and corresponding processing parameters.
  • the N groups of parameters are statistical information of training data sets of N AI encoders, and the statistical information includes a mean value and/or a distribution parameter of a mathematical distribution.
  • the first information is channel state information CSI or uplink data.
  • the first information is CSI
  • the method further includes: receiving a CSI reference signal CSI-RS, and the first information is determined according to a measurement result of the CSI-RS.
  • the method further includes: sending second indication information to the network device, where the second indication information is used to indicate the first encoder, or used to indicate the first encoder among the N AI encoders It is suitable for encoding the first information, or is used to indicate that the N AI encoders are not suitable for encoding the first information; receiving second configuration information, the second configuration information is used to configure the first encoder, and the second configuration information is based on the second configuration information.
  • the instructions are determined.
  • sending the second indication information is before sending the second information.
  • the first encoder used by the terminal may be obtained by requesting the network device in real time, and the terminal may not need to store the first encoder in advance, thereby reducing the storage pressure of the terminal.
  • the method further includes: receiving third configuration information, where the third configuration information is used to configure the N AI encoders, and/or the first encoder among the N AI encoders, and /or, a second encoder.
  • the third configuration information is received before the second information is sent.
  • an embodiment of the present application provides another information encoding control method, which is applied to a terminal.
  • the method includes: receiving first configuration information, where the first configuration information is used to configure an AI decider, and the AI decider is used to determine N Among the AI encoders, the AI encoders to which the first information is applicable, and/or for determining that the N AI encoders are not suitable for encoding the first information, where N is a positive integer greater than 1; sending the first indication information to the network device,
  • the first indication information is used to indicate that the first encoder is used for encoding the first information, the first encoder is determined according to the AI decider and the first information, and the first encoder is an encoder among N AI encoders, or,
  • the first encoder is a second encoder different from the N AI encoders.
  • the AI decider is used to determine the AI encoder to which the first information of the N AI encoders is applicable, and/or is used to determine that the N AI encoders are not suitable for encoding the first information, which may be replaced by: the AI decider is used to determine The AI codec to which the first information among the N AI codecs is applicable, and/or used to determine that the N AI codecs are not applicable to the first information codec. It can also be replaced with: the AI decider is configured to determine the AI decoder to which the first information is applicable among the N AI decoders, and/or to determine that the N AI decoders are not suitable for decoding the first information.
  • the N AI encoders correspond to N AI decoders respectively, and an AI encoder includes an AI encoder and a corresponding AI decoder.
  • the second encoder is an encoder based on a traditional encoding scheme such as a codebook, or another AI encoder with better generalization.
  • the first encoder used for encoding the first information is determined based on the AI decider, and the first encoder is an AI encoder among the N AI encoders or a second AI encoder other than the N AI encoders.
  • Encoder it can be understood that the first encoder used for encoding the first information is judged to be suitable for encoding the first information, so as to avoid passing the AI codec (or AI encoder and The problem of information distortion caused by the encoding and decoding of the first information by an AI decoder) and uncontrollable performance deterioration of the communication system.
  • the method further includes: sending second information, where the second information is determined based on encoding of the first information by the first encoder.
  • the first encoder is determined according to the output of the AI decider, and the output of the AI decider is obtained according to the first information and the AI decider.
  • the output of the AI decider is obtained according to the first input information and the AI decider, and the first input information is obtained after the first information or the first information is subjected to a second preprocessing, and the second preprocessing includes at least the following: One item: translation, scaling, Fourier transform, compressed sensing transform, truncation, AI model processing, and corresponding processing parameters.
  • the output of the AI decider is obtained by using the AI decider to process the first input information.
  • the output of the AI decider is an output obtained by using the first input information as the input of the AI decider.
  • the method further includes: the first information is channel state information CSI or uplink data.
  • the first information is CSI
  • the method further includes: receiving a CSI reference signal CSI-RS, and the first information is determined according to a measurement result of the CSI-RS.
  • the method further includes: sending second indication information to the network device, where the second indication information is used to indicate the first encoder, or used to indicate the first encoder among the N AI encoders It is suitable for encoding the first information, or is used to indicate that the N AI encoders are not suitable for encoding the first information; receiving second configuration information, the second configuration information is used to configure the first encoder, and the second configuration information is based on the second configuration information.
  • the instructions are determined.
  • sending the second indication information is before sending the second information.
  • the first encoder used by the terminal may be obtained by requesting the network device in real time, and the terminal may not need to store the first encoder in advance, thereby reducing the storage pressure of the terminal.
  • the method further includes: receiving third configuration information, where the third configuration information is used to configure the N AI encoders, and/or the first encoder among the N AI encoders, and /or, a second encoder.
  • the third configuration information is received before the second information is sent.
  • the embodiments of the present application provide another method for controlling information encoding, which is applied to a terminal.
  • the method includes: receiving first configuration information, where the first configuration information is used to configure parameters of the first AI encoder;
  • the device sends first indication information, the first indication information is used to instruct the first encoder to encode the first information, the first encoder is determined according to the parameters of the first AI encoder and the first information, and the first encoder is the first encoder.
  • An AI encoder, or the first encoder is a second encoder different from the first AI encoder.
  • the N groups of parameters of the first AI encoder can be replaced with: N groups of parameters of the first AI decoder, or N groups of parameters of the first AI encoder and decoder, wherein the first AI encoder corresponds to the first AI encoder and decoder.
  • AI decoder, the first AI codec includes a first AI encoder and a first AI decoder.
  • the second encoder is an encoder based on a traditional encoding scheme such as a codebook, or another AI encoder with better generalization.
  • the first encoder used for encoding the first information is determined based on the parameters of the first AI encoder, and the first encoder is the first AI encoder or the second encoder, which can be understood as being used for the first AI encoder.
  • the first encoder for encoding information is determined to be suitable for encoding the first information, so as to avoid encoding the first information by an AI codec (or an AI encoder and an AI decoder) that is not applicable to the first information And the information distortion caused by decoding, and the problem of uncontrollable performance deterioration of the communication system.
  • the method further includes: sending second information, where the second information is determined based on encoding of the first information by the first encoder.
  • the first encoder is determined according to the relationship between the first judgment parameter and the first judgment threshold, the first judgment parameter is determined according to the parameters of the first AI encoder and the first information, and the first judgment parameter is determined according to the parameters of the first AI encoder and the first information.
  • a judgment parameter is cosine similarity CS, probability density function PDF, probability mass function PMF or Euclidean distance.
  • the first judgment parameter is determined according to parameters of the first AI encoder, first information, and first preprocessing, where the first preprocessing includes at least one of the following: translation, scaling, Fourier transform, compressed sensing Transformation, truncation, AI model processing, and corresponding processing parameters.
  • the parameter of the first AI encoder is statistical information of the training data set of the first AI encoder, and the statistical information includes a mean value and/or a distribution parameter of a mathematical distribution.
  • the first information is channel state information CSI or uplink data.
  • the first information is CSI
  • the method further includes: receiving a CSI reference signal CSI-RS, and the first information is determined according to a measurement result of the CSI-RS.
  • the method further includes: sending second indication information to the network device, where the second indication information is used to indicate the first encoder, or used to indicate that the first AI encoder is suitable for encoding the first information , or used to indicate that the first AI encoder is not suitable for encoding the first information; receiving second configuration information, the second configuration information is used to configure the first encoder, and the second configuration information is determined according to the second indication information.
  • sending the second indication information is before sending the second information.
  • the first encoder used by the terminal may be obtained by requesting the network device in real time, and the terminal may not need to store the first encoder in advance, thereby reducing the storage pressure of the terminal.
  • the method further includes: receiving third configuration information, where the third configuration information is used to configure the first AI encoder, and/or the second encoder.
  • the third configuration information is received before the second information is sent.
  • an embodiment of the present application provides another information encoding control method, which is applied to a terminal.
  • the method includes: receiving first configuration information, where the first configuration information is used to configure an AI determiner, and the AI determiner is used to determine the first configuration information.
  • An AI encoder is suitable for encoding the first information, and/or used to determine that the first AI encoder is not suitable for encoding the first information; sending first indication information to the network device, where the first indication information is used to instruct the first encoder to use
  • the first encoder is determined according to the AI decider and the first information, the first encoder is the first AI encoder, or the first encoder is a second AI encoder different from the first AI encoder. Encoder.
  • the AI determiner is used to determine that the first AI encoder is suitable for encoding the first information, and/or is used to determine that the first AI encoder is not suitable for encoding the first information, which may be replaced by: the AI determiner is configured to determine the first AI The codec is suitable for the first information codec, and/or used to determine that the first AI codec is not suitable for the first information codec. It can also be replaced with: the AI decider is used to determine that the first AI decoder is suitable for decoding the first information, and/or is used to determine that the first AI decoder is not suitable for decoding the first information.
  • the first AI encoder corresponds to the first AI decoder, and the first AI encoder includes a first AI encoder and a first AI decoder.
  • the second encoder is an encoder based on a traditional encoding scheme such as a codebook, or another AI encoder with better generalization.
  • the first encoder used for encoding the first information is determined based on the AI decider, and the first encoder is the first AI encoder or the second encoder, which can be understood as being used for encoding the first information
  • the first encoder is judged to be suitable for encoding the first information, so as to avoid encoding and decoding the first information by an AI codec (or an AI encoder and an AI decoder) to which the first information is not applicable. information distortion and uncontrollable deterioration of the performance of the communication system.
  • the first encoder is determined according to the output of the AI decider, and the output of the AI decider is obtained according to the first information and the AI decider.
  • the output of the AI decider is obtained according to the first input information and the AI decider, and the first input information is obtained after the first information or the first information is subjected to a second preprocessing, and the second preprocessing includes at least the following: One item: translation, scaling, Fourier transform, compressed sensing transform, truncation, AI model processing, and corresponding processing parameters.
  • the output of the AI decider is obtained by using the AI decider to process the first input information.
  • the output of the AI decider is an output obtained by using the first input information as the input of the AI decider.
  • the method further includes: the first information is channel state information CSI or uplink data.
  • the first information is CSI
  • the method further includes: receiving a CSI reference signal CSI-RS, and the first information is determined according to a measurement result of the CSI-RS.
  • the method further includes: sending second indication information to the network device, where the second indication information is used to indicate the first encoder, or used to indicate that the first AI encoder is suitable for encoding the first information , or used to indicate that the first AI encoder is not suitable for encoding the first information; receiving second configuration information, the second configuration information is used to configure the first encoder, and the second configuration information is determined according to the second indication information.
  • sending the second indication information is before sending the second information.
  • the first encoder used by the terminal may be obtained by requesting the network device in real time, and the terminal may not need to store the first encoder in advance, thereby reducing the storage pressure of the terminal.
  • the method further includes: receiving third configuration information, where the third configuration information is used to configure the first AI encoder, and/or the second encoder.
  • the third configuration information is received before the second information is sent.
  • the embodiments of the present application provide another method for controlling information encoding, which is applied to a terminal.
  • the method includes: receiving first configuration information, where the first configuration information is used to configure N groups of parameters of N AI encoders or AI decider, the AI decider is used to determine the AI encoder to which the first information of the N AI encoders is applicable, and/or to determine that the N AI encoders are not suitable for encoding the first information; send the first request to the network device information, the first request information is used to indicate the first AI encoder among the N AI encoders, and the first request information is determined according to the first configuration information and the first information; send the first indication information to the network device, the first The indication information is used to indicate the second encoder, and the second encoder is different from the N AI encoders.
  • the second encoder is an encoder based on a traditional encoding scheme such as a codebook, or another AI encoder with better generalization.
  • the method further includes: receiving second configuration information sent by the network device, where the second configuration information is used to configure the first AI encoder, and the second configuration information is determined according to the first request information.
  • the terminal can use the second encoder for encoding, thereby reducing the transmission delay and ensuring information
  • the process of encoding feedback is not interrupted.
  • the method further includes: sending second information, where the second information is determined based on the first information and the second encoder.
  • the first request information is determined according to the relationship between the first judgment parameter and the first judgment threshold
  • the first judgment parameter is determined according to the N groups of parameters and the first information
  • the first judgment parameter is Cosine similarity CS, probability density function PDF, probability mass function PMF or Euclidean distance.
  • the first request information is determined according to the output of the AI decider, and the output of the AI decider is obtained according to the first information and the AI decider.
  • the method may further include: sending fourth information and third indication information, where the third indication information is used to indicate the first AI encoder, and the fourth information is based on the first AI encoder and the third indication information. Three information is determined.
  • the third information is channel state information CSI or uplink data.
  • the terminal may send information to the network device to request the configuration of the first AI encoder, so that the fourth information applicable to the first AI encoder can be directly encoded subsequently, There is no need to request again, reducing transmission overhead and transmission delay.
  • the first information is channel state information CSI or uplink data.
  • the embodiments of the present application provide another method for controlling information encoding, which is applied to a terminal.
  • the method includes: receiving first configuration information, where the first configuration information is used to configure N groups of parameters of N AI encoders or AI decider, the AI decider is used to determine the AI encoder to which the first information is applicable among the N AI encoders, and/or to determine that the N AI encoders are not suitable for encoding the first information; send a fourth indication to the network device information, the fourth indication information is used to indicate whether the first information is applicable to the first judgment result and/or the first judgment parameter of the N AI encoders, and the fourth indication information is determined according to the first configuration information and the first information; Receive fifth indication information, where the fifth indication information is used to instruct the first encoder to encode the first information, the fifth indication information is determined according to the fourth indication information, and the first encoder is an encoder among the N AI encoders , or, the first encoder is a second encoder different from the N AI encoders.
  • the second encoder is an encoder based on a traditional encoding scheme such as a codebook, or another AI encoder with better generalization.
  • the first encoder used for encoding the first information is determined based on the first configuration information, and the first encoder is an AI encoder among the N AI encoders or an AI encoder other than the N AI encoders.
  • Second encoder it can be understood that the first encoder used for encoding the first information is judged to be suitable for encoding the first information, so as to avoid passing the AI codec (or AI encoder that is not applicable to the first information) and AI decoder) to encode and decode the first information, resulting in information distortion and uncontrollable performance degradation of the communication system.
  • the first judgment result is used to indicate that the first information is applicable to the first AI encoder among the N AI encoders, or is used to indicate that the N AI encoders are not applicable to the first information.
  • the first judgment parameter is determined according to N groups of parameters and the first information, and the first judgment parameter is cosine similarity CS, probability density function PDF, probability mass function PMF or Euclidean distance.
  • the first judgment parameter is the output of the AI judger, and the output of the AI judger is obtained according to the first information and the AI judger.
  • the first information is channel state information CSI or uplink data.
  • an embodiment of the present application provides another information encoding control method, which is applied to a terminal.
  • the method includes: receiving first configuration information, where the first configuration information is used to configure N groups of parameters of N AI encoders or AI decider, the AI decider is used to determine the AI encoder to which the first information of the N AI encoders is applicable, and/or to determine that the N AI encoders are not suitable for encoding the first information; send the first information to the network device , the first information has not been encoded or has undergone high-fidelity encoding, the first information is used as the training data set of the first AI encoder to train the first AI encoder, and the first information is used as the training data set of the first AI encoder is determined according to the first configuration information and the first information.
  • the terminal can filter the information to be transmitted according to the first configuration information, so as to send the data with high value for training the AI codec to the network device, so as to prevent the terminal from sending data with low value for training the AI codec to the network device.
  • Network equipment due to the poor performance of AI codecs caused by unbalanced training data, also avoids wasting uplink bandwidth.
  • the first information is used as the training data set of the first AI encoder, and is determined according to the relationship between the first judgment parameter and the first judgment threshold, and the first judgment parameter is based on N groups of parameters Determined from the first information, the first judgment parameter is cosine similarity CS, probability density function PDF, probability mass function PMF or Euclidean distance.
  • the first information is used as the training data set of the first AI encoder, and is determined according to the output of the AI decider, and the output of the AI decider is obtained according to the first information and the AI decider of.
  • the first information is channel state information CSI or uplink data.
  • an embodiment of the present application provides another information encoding control method, which is applied to a network device.
  • the method includes: sending first configuration information to a terminal; the first configuration information is used to configure N of N AI encoders. Group parameter or AI decider, the AI decider is used to determine the AI encoder to which the first information among the N AI encoders is applicable, and/or is used to determine that the N AI encoders are not suitable for encoding the first information; receiving the first indication information and second information, the second information is determined by the terminal based on the first information and the first encoder, the first indication information is used to indicate that the first encoder is used for encoding the first information, and the first encoder is based on the first configuration information As determined from the first information, the first encoder is an encoder among the N AI encoders, or the first encoder is a second encoder different from the N AI encoders.
  • the method further includes: decoding the second information using a first decoder corresponding to the first encoder.
  • the first encoder is determined according to the relationship between the first judgment parameter and the first judgment threshold, the first judgment parameter is determined according to the N groups of parameters and the first information, and the first judgment parameter is Cosine similarity CS, probability density function PDF, probability mass function PMF or Euclidean distance.
  • the first encoder is determined according to the output of the AI decider, and the output of the AI decider is obtained according to the first information and the AI decider.
  • the first information is channel state information CSI or uplink data.
  • an embodiment of the present application provides another information encoding control method, which is applied to a network device.
  • the method includes: sending first configuration information to a terminal.
  • the first configuration information is used to configure N groups of N AI encoders Parameter or AI decider, the AI decider is used to determine the AI encoder to which the first information is applicable among the N AI encoders, and/or to determine that the N AI encoders are not suitable for encoding the first information; receive the first request information , the first request information is used to indicate the first AI encoder among the N AI encoders, and the first request information is determined according to the first configuration information and the first information; after receiving the first indication information and the second information, the first The indication information is used to indicate that the second encoder is used for encoding the first information, the second encoder is different from the N AI encoders, and the second information is determined based on the first information and the second encoder.
  • the second encoder is an encoder based on a traditional encoding scheme such as a codebook, or another AI encoder with better generalization.
  • the method further includes: sending second configuration information, where the second configuration information is used to configure the first AI encoder, and the second configuration information is determined according to the first request information.
  • the method further includes: decoding the second information using a second decoder corresponding to the second encoder.
  • the first request information is determined according to the relationship between the first judgment parameter and the first judgment threshold
  • the first judgment parameter is determined according to the N groups of parameters and the first information
  • the first judgment parameter is Cosine similarity CS, probability density function PDF, probability mass function PMF or Euclidean distance.
  • the first request information is determined according to the output of the AI decider, and the output of the AI decider is obtained according to the first information and the AI decider.
  • the method may further include: receiving fourth information and third indication information, where the third indication information is used to indicate the first AI encoder, and the fourth information is based on the first AI encoder and the third indication information. Three information is determined.
  • the fourth information is channel state information CSI or uplink data.
  • the first information is channel state information CSI or uplink data.
  • an embodiment of the present application provides another information encoding control method, which is applied to a network device.
  • the method includes: sending first configuration information to a terminal, where the first configuration information is used to configure N of N AI encoders Group parameter or AI decider, the AI decider is used to determine the AI encoder to which the first information among the N AI encoders is applicable, and/or is used to determine that the N AI encoders are not suitable for encoding the first information; receiving a fourth indication information, the fourth indication information is used to indicate whether the first information is applicable to the first judgment result and/or the first judgment parameter of the N AI encoders, and the fourth indication information is determined according to the first configuration information and the first information; Send fifth indication information, where the fifth indication information is used to instruct the first encoder to encode the first information, the fifth indication information is determined according to the fourth indication information, and the first encoder is an encoder among the N AI encoders , or, the first encoder is a second encoder different from the N AI encoders.
  • the second encoder is an encoder based on a traditional encoding scheme such as a codebook, or another AI encoder with better generalization.
  • the first judgment result is used to indicate that the first information is applicable to the first AI encoder among the N AI encoders, or is used to indicate that the N AI encoders are not applicable to the first information.
  • the first judgment parameter is determined according to N groups of parameters and the first information, and the first judgment parameter is cosine similarity CS, probability density function PDF, probability mass function PMF or Euclidean distance.
  • the first judgment parameter is the output of the AI judger, and the output of the AI judger is obtained according to the first information and the AI judger.
  • the first information is channel state information CSI or uplink data.
  • an embodiment of the present application provides another method for controlling information encoding, which is applied to a network device.
  • the method includes: sending first configuration information to a terminal, where the first configuration information is used to configure N AI encoders. N groups of parameters or AI decider, the AI decider is used to determine the AI encoder to which the first information among the N AI encoders is applicable, and/or is used to determine that the N AI encoders are not suitable for encoding the first information; receiving the first information information, the first information has not been encoded or has undergone high-fidelity encoding, the first information is used as the training data set of the first AI encoder to train the first AI encoder, and the first information is used as the training data of the first AI encoder The set is determined according to the first configuration information and the first information.
  • the first information is used as the training data set of the first AI encoder, and is determined according to the relationship between the first judgment parameter and the first judgment threshold, and the first judgment parameter is based on N groups of parameters Determined from the first information, the first judgment parameter is cosine similarity CS, probability density function PDF, probability mass function PMF or Euclidean distance.
  • the first information is used as the training data set of the first AI encoder, and is determined according to the output of the AI decider, and the output of the AI decider is obtained according to the first information and the AI decider of.
  • the first information is channel state information CSI or uplink data.
  • an embodiment of the present application provides a terminal, including a transceiver, a processor, and a memory; the above-mentioned memory is used to store computer program codes, and the above-mentioned computer program codes include computer instructions, and the above-mentioned processor invokes the above-mentioned computer instructions to make
  • the above-mentioned user equipment executes the first aspect to the seventh aspect and the information encoding control method provided by any one of the implementation manners of the first aspect to the seventh aspect of the embodiments of the present application.
  • an embodiment of the present application provides a network device, including a transceiver, a processor, and a memory; the memory is used to store computer program code, and the computer program code includes computer instructions, and the processor calls the computer instructions to
  • the above-mentioned user equipment is made to execute the eighth aspect, the eleventh aspect, and the information encoding control method provided by any one of the implementation manners of the eighth aspect and the eleventh aspect of the embodiments of the present application.
  • an embodiment of the present application provides another terminal, which is configured to execute the method executed by the terminal in any embodiment of the present application.
  • an embodiment of the present application provides a network device for executing the method performed by the network device in any embodiment of the present application.
  • an embodiment of the present application provides a computer storage medium, where the computer storage medium stores a computer program, and when the computer program is executed by an electronic device, is used to execute the first to eleventh aspects of the embodiment of the present application , and the information encoding control method provided by any one of the implementation manners of the first aspect to the eleventh aspect.
  • an embodiment of the present application provides a computer program product that, when the computer program product runs on an electronic device, enables the electronic device to execute the first to eleventh aspects of the embodiments of the present application, and the first The information encoding control method provided by any one of the implementation manners of the aspect to the eleventh aspect.
  • an embodiment of the present application provides an electronic device, where the electronic device includes executing the method or apparatus described in any embodiment of the present application.
  • the above-mentioned electronic device is, for example, a chip.
  • FIG. 1 is a schematic diagram of the architecture of a communication system provided by an embodiment of the present application.
  • FIG. 2 is a schematic diagram of a scenario of information encoding and decoding provided by an embodiment of the present application
  • FIG. 3 is a schematic structural diagram of a terminal 100 provided by an embodiment of the present application.
  • FIG. 4 is a schematic structural diagram of a network device 200 provided by an embodiment of the present application.
  • 5-10 are schematic flowcharts of control methods for some information encoding provided by embodiments of the present application.
  • FIG. 11 is a schematic flowchart of a screening method provided in an embodiment of the present application.
  • FIGS. 12-13 are schematic diagrams of the generation process of some artificial intelligence AI determinators provided by the embodiments of the present application.
  • FIG. 14 is a schematic diagram of a generation process of an AI codec provided by an embodiment of the present application.
  • the embodiments of the present application provide an information encoding control method, which is applied to a communication system including a network device and a terminal.
  • the network device can send the first configuration information to the terminal, and the terminal can judge whether the information to be transmitted is suitable for the artificial intelligence (artificial intelligence, AI) model for information encoding according to the first configuration information (it can also be understood as being based on AI technology. encoder, referred to as AI encoder). If the information to be transmitted is suitable for the AI encoder, the terminal can use the AI encoder to encode the information to be transmitted. If the information to be transmitted is not suitable for the AI encoder, the terminal can use the standby encoder to encode the information to be transmitted. Therefore, an appropriate information encoding scheme can be selected quickly and efficiently, so as to avoid distortion of the information to be transmitted after encoding and decoding, thereby avoiding affecting the performance of the communication system and improving the transmission quality.
  • AI artificial intelligence
  • an AI encoder can correspond to an AI model for information decoding (it can also be understood as a decoder based on AI technology, referred to as AI decoder), and an AI encoder and corresponding AI decoder can be referred to as It is a group of AI codecs (which can also be understood as codecs based on AI technology). Different groups of AI codecs have different structures or the same structure but different model parameters (coefficients). Different groups of AI codecs have different identifiers. The AI encoders in different groups of AI codecs may be the same but the AI decoders are different, or the AI decoders in different groups of AI codecs may be the same but the AI encoders may be different.
  • the information encoded by the AI encoder needs to be decoded by the corresponding AI decoder in the same group, and other decoders cannot decode (or the decoded information has a large deviation from the original information to be transmitted).
  • the AI encoder is usually trained as an autoencoder together with the corresponding AI decoder, that is, the input of the AI encoder and the output of the AI decoder are both set to the information to be encoded during training (understandable).
  • the training data set of the AI encoder, the AI decoder or the set of AI codecs obtain the parameters (coefficients) in the AI model (including the AI encoder and/or the AI decoder) through error back propagation , so that the information obtained after AI encoding and AI decoding of the information to be encoded is as consistent as possible with the information to be encoded during inference.
  • the information to be encoded is applicable to the AI encoder, that is, the representation is also applicable to the AI decoder corresponding to the AI encoder. It can also be understood that the information to be encoded is applicable to the corresponding AI codec (that is, including the above AI encoder and corresponding AI decoder). Therefore, the terminal judges whether the information to be transmitted is suitable for the AI encoder according to the first configuration information, which can be understood as: the terminal judges whether the information to be transmitted is suitable for the AI encoder and the corresponding AI decoder according to the first configuration information (ie AI codec).
  • a backup encoder can also correspond to a decoder (referred to as a backup decoder), and the information encoded by the backup encoder also needs to be decoded by the corresponding backup decoder, and cannot be decoded by other decoders.
  • the decoding scheme of the standby decoder depends on the encoding scheme of the corresponding standby encoder.
  • the standby encoder is an encoder that uses a traditional encoding scheme such as a codebook
  • the corresponding standby decoder is a decoder that uses a traditional decoding scheme such as a codebook.
  • the backup encoder is another AI encoder with better generalization
  • the corresponding backup decoder is another AI decoder with better generalization.
  • the terminal can feed back the encoded information to the network device, and send a notification indicating the encoder used by the terminal to the network device, and the network device can determine the corresponding decoder according to the notification, and use the decoder to decode to obtain the terminal. reported information.
  • the encoder used by the terminal is an AI encoder
  • the decoder used by the network device is the AI decoder corresponding to the AI encoder.
  • the encoder used by the terminal is the backup encoder
  • the decoder used by the network device is the backup decoder corresponding to the backup encoder.
  • the information to be transmitted is the information sent by the terminal to the network device, which may be measurement information and state information such as channel state information (CSI). Not limited to this, it may also be service data, such as audio data, video data, text data, and so on.
  • CSI channel state information
  • the AI encoder or the backup encoder used by the terminal may be sent by the network device to the terminal.
  • the AI encoder or backup encoder used by the terminal may be pre-negotiated and configured by the terminal and the network device.
  • the communication system may be a wireless communication system, such as but not limited to global system for mobile communications (GSM), code division multiple access (CDMA), wideband code division multiple access (wideband code division multiple access, WCDMA), time division synchronous code division multiple access (time division synchronous code division multiple ac, TD-SCDMA), long term evolution (long term evolution, LTE), new radio access (new radio, NR) Or other future network systems.
  • GSM global system for mobile communications
  • CDMA code division multiple access
  • WCDMA wideband code division multiple access
  • WCDMA wideband code division multiple access
  • time division synchronous code division multiple access time division synchronous code division multiple ac
  • LTE long term evolution
  • new radio access new radio
  • the network device may be a device for sending or receiving information
  • the network device is an access network device
  • the network device is a core network device.
  • the base station is a device deployed in a radio access network (radio access network, RAN) to provide a wireless communication function.
  • radio access network radio access network, RAN
  • the names of base stations may be different.
  • a base transceiver station in GSM or CDMA
  • a node B node B (node B, NB) in WCDMA
  • an evolved base station evolved node B, eNodeB
  • Next-generation base stations g node B, gNB
  • the network device may be the base station 110 shown in FIG. 1 below, or the core network 120.
  • the terminal may be a device with a wireless communication function, and optionally, the terminal is a UE.
  • a terminal may also be referred to as a mobile station, an access terminal, a user agent, or the like.
  • the terminal is a terminal in the form of a handheld device, a wearable device, a computing device, a portable device, or a vehicle-mounted device.
  • the terminal is specifically a device such as a cellular phone, a smart phone, smart glasses, a laptop computer, a personal digital assistant, or a cordless phone.
  • the terminal may be the terminal 100 shown in FIG. 1 below.
  • FIG. 1 is a schematic structural diagram of a communication system provided by an embodiment of the present application.
  • the communication system may include a terminal 100 , a base station 110 and a core network 120 .
  • the core network 120 may be connected to at least one base station 110, the base station 110 may provide wireless communication services for at least one terminal 100, and the terminal 100 may be connected to at least one base station 110 through an air interface.
  • the core network 120 is a key control node in the communication system, and is mainly responsible for signaling processing functions, such as but not limited to implementing functions such as access control, mobility management, and session management.
  • At least one base station 110 may constitute a RAN node.
  • the core network 120 may be referred to as a 5G core network (5G Core, 5GC) 120, and the base station 110 may be referred to as a gNB 110.
  • 5G Core 5G Core
  • the NR-RAN node may include at least one gNB 110 connected to the 5GC 120 through an NG interface, and at least one gNB 110 in the NR-RAN node may connect and communicate through an Xn-C interface.
  • the terminal 100 can be connected to the gNB 110 through the Uu interface.
  • the core network 120 may send downlink information to the terminal 100 through the base station 110 , and the terminal 100 may also send uplink information to the core network 120 through the connected base station 110 . Wherein, when the terminal 100 is within the coverage of the base station 110, it needs to be connected to the base station 110 after random access and other operations. interact.
  • terminal 100 the base station 110, and the core network 120 shown in FIG. 1 are only used as examples, and are not limited in this embodiment of the present application.
  • FIG. 2 exemplarily shows a schematic diagram of a scenario of information encoding and decoding.
  • the scenario shown in FIG. 2 includes a terminal 100 and a network device 200, where the terminal 100 and the network device 200 can be connected and communicated.
  • the network device 200 may be deployed with an AI encoder and a corresponding AI decoder.
  • the network device 200 may send the AI encoder to the terminal 100, so that the subsequent terminal 100 uses the AI encoder to perform information encoding on the information to be transmitted (ie, information A).
  • the terminal 100 can determine whether the information A is suitable for the AI encoder sent by the network device 200 , and when applicable, the terminal 100 can use the AI encoder to encode the information A, and encode the encoded The information is sent to the network device 200. Then, the network device 200 can decode the encoded information using the AI decoder to obtain recovered information (ie, information A').
  • FIG. 3 shows a schematic structural diagram of a terminal 100 .
  • the terminal 100 may be the terminal 100 shown in FIG. 1 or the terminal 100 shown in FIG. 2 .
  • the terminal 100 may include a processor 110, a memory 120, and a transceiver 130, and the processor 110, the memory 120, and the transceiver 130 are connected to each other through a bus.
  • the processor 110 may be one or more central processing units (central processing units, CPUs). In the case where the processor 110 is a CPU, the CPU may be a single-core CPU or a multi-core CPU. In some embodiments, the processor 110 may include multiple processing units, such as an application processor (AP), a modem (modem), and the like. Wherein, different processing units may be independent devices, or may be integrated in one or more processors.
  • the memory 120 may include, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read only memory (EPROM), Or portable read-only memory (compact disc read-only memory, CD-ROM).
  • the memory 120 is used to store relevant computer programs and information, optionally, the memory 120 is used to store the AI encoder sent by the network device, optionally, the memory 120 is used to store the information sent by the network device for judging whether there is information to be transmitted
  • the AI model of the applicable AI encoder (referred to as the AI decider).
  • the transceiver 130 is used to receive and transmit information.
  • transceiver 130 may include a wireless transceiver and a mobile transceiver.
  • the terminal 100 may implement mobile communication technologies such as GSM, CDMA, WCDMA, SCDMA, UMTS, LTE, and NR through a modem processor and a mobile transceiver.
  • the terminal 100 can communicate with the network device through the modem processor and the mobile transceiver, for example, transmit measurement information such as CSI, status information, AI encoder, and the like.
  • the processor 110 in the terminal 100 is configured to read the computer program code stored in the memory 120, and execute the steps performed by the terminal in the information encoding control method shown in Figs. 5-10 and the screening method shown in Fig. 11 .
  • FIG. 4 shows a schematic structural diagram of a network device 200 .
  • the network device 200 may be the base station 110 or the core network 120 shown in FIG. 1 , or may be the network device 200 shown in FIG. 2 .
  • the network device 200 may include a processor 210, a memory 220, and a transceiver 230, and the processor 210, the memory 220, and the transceiver 230 are connected to each other through a bus.
  • the processor 210 may be one or more CPUs, and if the processor 210 is one CPU, the CPU may be a single-core CPU or a multi-core CPU. In some embodiments, the processor 210 may include multiple processing units, wherein different processing units may be independent devices or may be integrated into one or more processors. Optionally, the processor 210 may be trained to generate an AI codec. Optionally, the processor 210 can be trained to generate an AI decider.
  • the memory 220 may include, but is not limited to, RAM, ROM, EPROM, or CD-ROM. The memory 220 is used to store relevant computer programs and information, optionally, the memory 220 is used to store the acquired AI codec, optionally, The memory 220 is used to store the acquired AI decider.
  • the transceiver 230 is used to receive and transmit information.
  • the network device 200 may implement mobile communication technologies such as GSM, CDMA, WCDMA, SCDMA, UMTS, LTE, and NR through the processor 210 and the transceiver 230 .
  • the network device 200 may communicate with the terminal through the processor 210 and the transceiver 230, for example, transmit measurement information such as CSI, status information, AI encoder, and the like.
  • the processor 210 in the network device 200 is configured to read the computer program code stored in the memory 220, and execute the steps performed by the network device in the information encoding control method shown in FIGS. 5-10 and the screening method shown in FIG. 11 .
  • the method can be applied to the communication system shown in FIG. 1 , and can also be applied to the scenario shown in FIG. 2 .
  • the terminal in this method may be the terminal 100 shown in FIG. 3 .
  • the network device in this method may be the network device 200 shown in FIG. 4 .
  • FIG. 5 is a schematic flowchart of an information encoding control method provided by an embodiment of the present application. The method may include but is not limited to the following steps:
  • S101 The network device sends first configuration information to the terminal.
  • the method may include: connecting a network device and a terminal, and the connection method may refer to the descriptions of FIG. 1 to FIG. 4 above.
  • the network device is a base station
  • the terminal enters the range covered by the base station and initiates random access to the base station. After the random access is successful, the terminal is connected to the base station and can communicate.
  • the method may include: the network device and the terminal negotiate to determine to enable the information encoding feedback mechanism based on the AI encoder. Subsequent network devices can send the AI encoder to the terminal, and the terminal can use the AI encoder for encoding at the time of negotiation.
  • the time of negotiation is, for example, but not limited to, the time when the terminal determines that there is an AI encoder applicable to the information to be transmitted according to the first configuration information (for example, after S102), or the time when the information indicating the used encoder sent by the network device is received ( For example, after receiving the sixth indication information shown in Figure 10 below).
  • the first configuration information may include information of N groups of AI codecs, may also be understood as information of N AI encoders, or may be understood as information of N AI decoders, wherein the N AI encoders The encoders correspond to the N AI decoders respectively, and the N groups of AI codecs may respectively include the N AI encoders and the N AI decoders.
  • the content included in the first configuration information can be in three cases, as follows:
  • the first configuration information may include statistical information of training data sets of N groups of AI codecs (which may also be understood as N AI codecs) acquired by the network device, where N is a positive integer.
  • the first configuration information may include N groups of second configuration information, the N groups of second configuration information respectively correspond to the above N groups of AI codecs, and each group of second configuration information may include the training of the corresponding AI codecs.
  • Statistics for the dataset may include, but is not limited to, the mean value and/or the distribution parameters of the mathematical distribution, wherein the distribution parameters of any mathematical distribution may be determined according to the characteristics of the mathematical distribution.
  • the statistics for a set of AI codecs include the mean and variance of two normal distributions, both of which have different mean and variance.
  • the statistics for a set of AI codecs include the mean and variance of a normal distribution, and the mean and ⁇ of a Poisson distribution.
  • the statistical information may be of multiple different dimensions.
  • the dimension of the training dataset is [L,K 1 ,K 2 ,...,K z ], where L,K 1 ,K 2 ,...,K z are all positive integers, K 1 ,K 2 ,... , K z is the feature dimension of the training dataset.
  • L is the statistical dimension of the training dataset.
  • the mean of the training data set and the distribution parameters of the mathematical distribution can be the dimensions along K 1 , K 2 ,...,K z , respectively, or can also be the combined dimension of K 1 ⁇ K 2 ⁇ ... ⁇ K z , not limited to Here, it may also be other permutation and combination dimensions, which are not limited in this application.
  • Case 2 is similar to case 1, except that the statistical information in the first configuration information is subjected to first preprocessing, and the first preprocessing may include at least one of the following: translation, scaling, Fourier transform, compressed sensing Transformation (i.e. multiplication by compressed sensing measurement matrix), truncation, AI model processing, etc.
  • the first preprocessing may further include corresponding processing parameters, such as the direction and value of the translation.
  • the AI model processing can be performed through the AI model or the shallow part of the AI model, such as feature extraction, to achieve dimensionality reduction.
  • the AI model here can be different from the AI model of the AI encoder and the AI model of the AI decoder.
  • the information to be processed is a CSI matrix H with a dimension of R ⁇ W
  • the CSI matrix H can be processed by a fully connected linear layer of an AI model (ie, compressed sensing coding), specifically including multiplying by a dimension.
  • U of E ⁇ R where E can be much smaller than R.
  • the processed matrix O U ⁇ H
  • the dimension of the matrix O is E ⁇ W.
  • the decoded CSI matrix H′ LAMP(U ⁇ H).
  • the matrix U may be obtained by training as a layer of neural network, that is, based on the above formula for obtaining the decoded CSI matrix, the measurement matrix U is obtained by training with the known CSI matrix H.
  • the measurement matrix in the compressed sensing transformation may be randomly generated, for example by a Gaussian distribution.
  • preprocessing It is not limited to this, and other types of preprocessing are also possible, and the specific content of the preprocessing is not limited in this application.
  • the first configuration information may further include first preprocessed information, so that the terminal may determine the first preprocessed content according to the first configuration information.
  • each group of second configuration information may specifically include corresponding first preprocessed information.
  • the method before S101, may further include: the terminal receives the first preprocessed information sent by the network device.
  • the method may further include: the network device and the terminal negotiate and determine the content of the first preprocessing, that is, the first preprocessing is predefined.
  • the content of the first preprocessing that the statistical information of the training data sets of different AI encoders undergo may be different, for example, the first preprocessing indicated by different sets of second configuration information may be different.
  • the terminal when judging whether the information to be transmitted is suitable for the AI encoder to be applied (that is, any AI encoder in the above-mentioned N groups of AI codecs), the terminal needs to first perform a first preprocessing on the information to be transmitted, Then, a judgment process is performed based on the first preprocessed information, as shown in S102.
  • the terminal uses the AI encoder (corresponding to the first preprocessing here) to encode the information to be transmitted, it needs to first perform the first preprocessing on the information to be transmitted, and then use the AI encoder to perform the first preprocessing.
  • the latter information is encoded, specifically as shown in S103.
  • the terminal needs to first perform a judgment process on the information to be transmitted when performing the judgment process and encoding using the AI encoder.
  • the first configuration information may further include an indication of the type of statistical information.
  • it may specifically be an indication of the type of statistical information included in each group of second configuration information, and the types of statistical information included in different groups of second configuration information may be different.
  • the method may further include: the network device and the terminal negotiate to determine the type of statistical information, that is, the type of statistical information is predefined.
  • the first configuration information may further include a set of judgment thresholds, and optionally, may further include: judging whether the information to be transmitted is suitable for the information to be transmitted according to the set of judgment thresholds
  • the judgment method of the applied AI encoder before S101, the method may further include: the network device and the terminal negotiate and determine a set of judgment thresholds, optionally and the above judgment method, that is, the judgment threshold is predefined, optionally, the judgment method is predefined.
  • the above-mentioned set of judgment thresholds may include at least one judgment threshold.
  • the method before S101, the method may further include: the terminal receives the judgment threshold and/or the judgment method sent by the network device. Optionally, for different terminals, the judgment threshold determined by the network device may be different.
  • the first configuration information may include an AI decider, that is, an AI model for judging whether the information to be transmitted is suitable for the AI encoder to be applied.
  • the input of the AI decider is information to be transmitted by the terminal.
  • the input of the AI decider is obtained after the information to be transmitted is subjected to the second preprocessing.
  • There are at least one output port of the AI decider for example, N, respectively corresponding to the above-mentioned N groups of AI codecs (it can also be understood as corresponding to N AI codecs).
  • an output value of the AI decider characterizes whether the information to be transmitted is applicable to this one AI encoder.
  • the N output values of the AI determinator can represent the situation that the information to be transmitted is applicable to the N AI encoders, and the output value of any output port of the AI determinator can represent the information to be transmitted.
  • the information is applicable to the probability value of the AI encoder corresponding to the output port. The higher the probability value, the more suitable the information to be transmitted is for the AI encoder, that is, the AI codec corresponding to the output port is applied to the information to be transmitted. The effect of codec is expected to be better.
  • the first configuration information may further include indication information of the output of the AI decider.
  • the indication information may include content represented by the output of the AI decider, for example, representing whether the information to be transmitted is suitable for the AI encoder to be applied, or a probability value representing that the information to be transmitted is suitable for the AI encoder to be applied.
  • the indication information can also include the threshold of the output of the AI judging device (referred to as the output threshold), and judge whether the information to be transmitted is suitable for the judgment of the AI encoder to be applied according to the output threshold and the output value of the AI judging device. method.
  • the method may further include: the network device and the terminal negotiate to determine the indication information output by the AI decider, that is, the indication information is predefined. In other embodiments, before S101, the method may further include: the terminal receiving indication information of the output of the AI decider sent by the network device.
  • the output value of the AI decider is 0 or 1 respectively, wherein, when the output is 1, the information to be transmitted is applicable to this AI encoder, and when the output is 0, the information to be transmitted is represented. Not applicable to this one AI encoder.
  • the value range of the output of the AI determinator is [0, 1]. The higher the output value, the more suitable the first information is for the AI encoder, that is, the AI codec corresponding to the output port is applied to the information to be transmitted. The effect of information encoding and decoding is expected to be better.
  • the terminal can determine that the information to be transmitted is suitable for the AI encoder, otherwise the information to be transmitted is not suitable for the AI encoder.
  • the information to be transmitted is applicable to the AI encoder corresponding to the i-th output port, where the value range of i is [1, N], and I t is the output threshold of the AI determinator.
  • the information to be transmitted is most suitable for the AI encoder corresponding to the ith output port, that is, the AI corresponding to the ith output port is applied.
  • the codec is expected to perform the best information encoding and decoding of the information to be transmitted.
  • outputs of other AI deciders please refer to the description of the outputs of the AI deciders in S102, which will not be described in detail for the time being.
  • the first configuration information may further include second preprocessed information, so that the terminal may determine the second preprocessed content according to the first configuration information.
  • the method before S101, the method may further include: the network device and the terminal negotiate and determine the content of the second preprocessing, that is, the second preprocessing is predefined.
  • the method before S101, the method may further include: the terminal receives the second preprocessed information sent by the network device.
  • the first preprocessing and the second preprocessing may be the same or different. For an example of the content of the second preprocessing, please refer to the above-mentioned example of the first preprocessing.
  • the first configuration information may also include first pre-processed information or pre-defined first pre-processing, so that when the terminal uses the AI encoder to encode the information to be transmitted, the information to be transmitted is firstly encoded.
  • the first preprocessed information is performed, and then the AI encoder is used to encode the first preprocessed information, as specifically shown in S103.
  • the first configuration information includes the information of the second preprocessing or pre-defined the second preprocessing, when the terminal determines whether the information to be transmitted is suitable for the AI encoder to be applied, it needs to perform the second preprocessing on the information to be transmitted first. Preprocessing is performed, and then a judgment process is performed based on the information after the second preprocessing, as shown in S102.
  • the first configuration information may further include performance requirements of N AI encoders, such as but not limited to including storage space capacity requirements, computing power requirements, delay requirements, etc. .
  • each group of second configuration information may specifically include the performance requirements of the corresponding AI encoder.
  • the method before S101, may further include: the network device sends the performance requirements of the N AI encoders to the terminal. In other embodiments, before S101, the method may further include: the network device and the terminal negotiate and determine the performance requirements of the N AI encoders, that is, the performance requirements are predefined.
  • the first configuration information may further include an alternate coding scheme (ie, a scheme for coding by using a spare encoder), and optionally, and a usage manner of the alternate coding scheme.
  • the method may further include: the network device and the terminal negotiate to determine an alternate coding scheme, optionally, and a usage manner of the alternate coding scheme, that is, the alternate coding scheme is predefined, optionally , the usage is predefined.
  • the method may further include: the terminal receives an alternate coding scheme sent by the network device, optionally, and a usage manner.
  • the backup encoder may be, but is not limited to, an encoder using a traditional encoding scheme such as a codebook, or other AI encoder with better generalization.
  • the usage manner may include: when the terminal determines that there is no AI encoder applicable to the information to be transmitted, the terminal may use a backup encoder to encode the information to be transmitted. And/or, although the terminal determines that there is an AI encoder suitable for the information to be transmitted, but the terminal has not yet received the AI encoder, the terminal uses a backup encoder to encode the information to be transmitted. And/or, although the terminal determines that there is an AI encoder suitable for the information to be transmitted, but the terminal performance does not meet the performance requirements of the AI encoder, the standby encoder is used to encode the information to be transmitted.
  • the terminal determines, according to the first configuration information, whether there is an AI encoder to which the first information is applicable.
  • S102 is an optional step.
  • the terminal determines, according to the first configuration information, whether the first information to be transmitted is suitable for the AI encoder to be applied (ie, any AI encoder in the above-mentioned N groups of AI codecs).
  • the first information is the information to be transmitted, which may be, but is not limited to, measurement information such as CSI, status information, and service data such as audio data and text data.
  • the terminal may judge whether the first information is applicable to this AI encoder according to the first configuration information, and optionally, when N is greater than 1, the terminal may judge the above N according to the first configuration information.
  • a group of AI codecs which can also be understood as N AI encoders
  • an AI encoder to which the first information is applicable.
  • the above-mentioned N AI encoders may be sent by the network device to the terminal, optionally, the network device may send N AI encoders to the terminal before S102, optionally, it may be S102 (the terminal determines the information to be transmitted) After the applicable AI encoder), the terminal sends information to the network device to request to configure the AI encoder, and the network device sends the AI encoder to the terminal in response to the terminal's request.
  • the terminal may determine whether the performance requirements of the N AI encoders are satisfied according to the first configuration information, and if none of them are satisfied, the terminal may determine that the used encoder is a backup encoder; if the terminal determines that the M AI encodings are satisfied If the performance requirements of the encoder are met, it can be determined according to the first configuration information whether there is an AI encoder to which the first information is applicable among the M AI encoders, where M is less than or equal to N. Exemplarily, if there is an AI encoder to which the first information is applicable in the M AI encoders, the terminal may determine that the used encoder is the AI encoder, and if it does not exist, the terminal may determine that the used encoder is a backup encoder. .
  • the terminal may also first determine the AI encoder to which the first information is applicable, and then determine whether the performance requirements of the AI encoder are met according to the first configuration information. Exemplarily, when not satisfied, the terminal may determine that the used encoder is a backup encoder; when satisfied, the terminal may determine that the used encoder is the AI encoder.
  • the terminal meeting the performance requirements of the AI encoder may include at least one of the following: the terminal meets the storage capacity requirement of the AI encoder, the terminal meets the computing capacity requirement of the AI encoder, and the terminal meets the delay requirement of the AI encoder.
  • the terminal may determine that the storage capacity requirement of the AI encoder is met.
  • the parameter representing the computing capability of the terminal for example, the amount of computation per unit time, etc.
  • the terminal can determine that the computing capability requirement of the AI encoder is met.
  • the delay for the terminal to complete information encoding based on the AI encoder is less than the delay requirement of the AI encoder, the terminal can determine that the delay requirement of the AI encoder is met.
  • meeting the performance requirements of the AI encoder is only a precondition for using the AI encoder, that is, if the performance requirements of the AI encoder are not met, the AI encoder cannot be used, and if the performance requirements of the AI encoder are met Determine whether to use the AI encoder according to the actual situation (for example, whether the terminal receives the instruction of the AI encoder or the network device).
  • the terminal may determine whether there is an AI encoder applicable to the first information in different ways, which may specifically include the following three situations:
  • Case 1 when the first configuration information is shown in Case 1 of S101 and includes the mean value of the training data set, the terminal may calculate the cosine similarity (CS) between the first information and the mean value.
  • the terminal may first perform the first preprocessing on the first information, and then calculate that the first information has undergone the first preprocessing. The cosine similarity between the processed information and the mean. Finally, the terminal can judge whether the first information is suitable for the AI encoder to be applied according to the calculated cosine similarity and the corresponding judgment threshold.
  • the terminal may calculate the cosine similarity corresponding to this AI encoder, and judge whether the first information is applicable to the AI encoder according to the relationship between the cosine similarity and the judgment threshold.
  • a set of judgment thresholds predefined or indicated in the first configuration information includes a judgment threshold CS t , and when the calculated cosine similarity CS 0 >CS t indicates that the first information is applicable to the AI encoder (that is, applicable to this AI encoder) group AI codec), otherwise the first information signifies that the AI codec is not applicable.
  • the set of judgment thresholds includes two judgment thresholds CS t1 and CS t2 , when the calculated cosine similarity CS 0 ⁇ CS t1 indicates that the first information is not applicable to the AI encoder, when CS t2 ⁇ CS 0 ⁇ CS t1 Characterizing that the first information is suitable for the AI encoder, but the effect of applying the set of AI codecs for encoding and decoding the first information is expected to be poor, and when CS 0 >CS t2 , characterizing the first information is applicable to the AI encoder, and The effect of encoding and decoding the first information by applying the group of AI codecs is expected to be good.
  • the judgment method defined in advance or indicated in the first configuration information may include: determining that the applied encoder is the AI encoder only when CS 0 >CS t2 , or determining that the applied encoder is the AI when CS 0 ⁇ CS t1 Encoder.
  • the terminal can calculate the corresponding cosine similarities of the N AI encoders respectively, and judge whether there is an AI encoder for which the first information is applicable according to the relationship between the N cosine similarities and the judgment threshold, Optionally, and an AI encoder to which the first information applies.
  • the terminal may determine that the first information is applicable to the AI encoder corresponding to the cosine similarity.
  • the terminal may determine that the cosine similarity and the judgment threshold meet the preset threshold conditions, and the AI encoder with the largest cosine similarity is the most suitable AI encoder for the first information, that is, the AI encoder of the group is applied to the first information.
  • the AI encoder with the largest cosine similarity is the most suitable AI encoder for the first information, that is, the AI encoder of the group is applied to the first information. The effect of encoding and decoding a message is expected to be the best.
  • Case 2 when the first configuration information is shown in case 1 of S101, and includes complete distribution parameters of the mathematical distribution (for example, including the mean and variance matrix that can jointly form a multi-dimensional Gaussian distribution), the terminal can calculate the first information in the mathematical distribution.
  • a probability parameter in a distribution such as a probability density function (PDF) or a probability mass function (PMF).
  • PDF probability density function
  • PMF probability mass function
  • the terminal may first perform the first preprocessing on the first information to obtain the first processing information, and then calculate The probability parameter of the first processing information in the mathematical distribution. Finally, the terminal can judge whether the first information is suitable for the AI encoder to be applied according to the calculated probability parameter and the corresponding judgment threshold.
  • the terminal may calculate the probability parameter corresponding to this AI encoder, and judge whether the first information is applicable to the AI encoder according to the relationship between the probability parameter and the judgment threshold.
  • the terminal may calculate probability parameters corresponding to N AI encoders respectively, and judge whether there is an AI encoder applicable to the first information according to the relationship between the N probability parameters and the judgment threshold, and optionally , and the AI encoder to which the first information applies.
  • the terminal may determine that the first information is applicable to the AI encoder corresponding to the probability parameter.
  • the terminal may determine that the probability parameter and the judgment threshold meet the preset threshold condition, and the AI encoder with the largest probability parameter is the most suitable AI encoder for the first information, that is, the AI encoder/decoder of the group is applied to the first information.
  • the effect of encoding and decoding is expected to be the best.
  • the specific example is similar to the example shown in the above case 1, and is not repeated here.
  • the terminal may set the input of the AI decider as the first information to obtain the output value of the AI decider.
  • the terminal may perform second preprocessing on the first information to obtain second processing information, and then set the input of the AI decider to undergo the second processing information to obtain the output value of the AI decider.
  • the terminal may determine whether there is an AI encoder to which the first information is applicable according to the output value of the AI determiner.
  • the AI decider is a single output port, that is, it has one output value.
  • there are N values of the output values of the AI decider for example, [1, N], the output value of the AI decider is a positive integer), and these N values correspond to N AI encoders respectively.
  • the output value of the decider is a value corresponding to the first AI encoder, it indicates that the first AI encoder is an AI encoder to which the first information is applicable among the N AI encoders determined by the AI decider.
  • the output value of the AI decider has N+1 values (for example, [0, N], and the output value of the AI decider is a positive integer), wherein N values (for example, [1, N]) Corresponding to N AI encoders respectively, the N values are consistent with the above description of N values with N values, and will not be repeated here.
  • the output value of the AI determinator is one other than the N values
  • the characterizing AI decider determines that the first information is not applicable to the N AI encoders.
  • the AI decider is N output ports, that is, there are N output values, and the N output values correspond to N AI encoders respectively.
  • any one of the N output values has two values (for example, 0 or 1), and one of them (for example, 0) indicates that the first information is not applicable to the AI encoder corresponding to the output value, and in addition 1 (eg 1) indicates that the first information is applicable to the AI encoder corresponding to the output value.
  • the AI decider is an N+1 output port, that is, there are N+1 output values, wherein the N output values are consistent with the description of the N output values of the AI decider of the N output port, and are not repeated here.
  • the other 1 output value has 2 values (eg 0 or 1), of which 1 (eg 0) indicates that the first information is not applicable to N AI encoders, and the other 1 (eg 1) indicates that the first information is not applicable to the N AI encoders.
  • the first information applies to at least one of the N AI encoders.
  • the terminal may also determine whether there is an AI encoder to which the first information is applicable according to the relationship between the output value of the AI determiner and the output threshold.
  • N is equal to 1
  • the AI decider is a single output port, and the terminal can judge whether the first information is applicable to this AI encoder according to the output value of the AI decider.
  • the output value of the AI decider is 0 or 1 respectively.
  • the terminal can determine that the first information is applicable to the AI encoder, and when the output is 0, the terminal can determine that the first information is not applicable to the AI encoder.
  • the value range of the output of the AI decider is [0, 1], and the higher the output value, the more suitable the first information is for the AI encoder, that is, the first information is encoded and decoded by using the set of AI codecs. The effect is expected to be better.
  • the terminal may determine that the first information is applicable to the AI encoder, otherwise the first information is not applicable to the AI encoder. It is not limited to this, and there may be multiple output thresholds, and the judgment example is similar to the cases of multiple judgment thresholds shown in the above cases 1 and 2, and will not be repeated here.
  • the AI decider has multiple output ports, and each of the N output ports can output a probability value, which can be expressed as I 1 , I 2 , . . . , I N .
  • the effect of encoding and decoding is expected to be better, where the value range of i is [1, N].
  • the terminal may determine that the first information is applicable to the AI encoder corresponding to the ith output port.
  • the terminal can determine that the first information is most suitable for the AI encoder corresponding to the ith output port, that is, the application of the first information corresponding to the ith output port. It is expected that the effect of the AI codec in encoding and decoding the first information is the best.
  • S103 The terminal uses the first encoder to encode the first information and obtains the second information.
  • the first encoder may be an AI encoder for which the terminal determines that the first information is (most) applicable.
  • the terminal determines that the first information is (most) applicable to the first encoder in the N AI encoders, and the terminal has received the first encoder before S103, the terminal can use the first encoder to The information is encoded, and the specific process example can be seen in Figure 6 below.
  • the terminal may send information to the network device to request the network device to configure the first encoder, and receive the first encoder. After the encoder, the first encoder can be used to encode the first information. For a specific process example, see FIG. 7 below.
  • the terminal may directly use the first information as the input of the first encoder to obtain the output second information.
  • the terminal may also perform first preprocessing on the first information to obtain the first processing information, and then use the first processing information as the input of the first encoder to obtain the output second information.
  • the first encoder may be a backup encoder.
  • the terminal determines that there is an AI encoder to which the first information is applicable in the N AI encoders, but the terminal has not yet received the AI encoder, the terminal can use the first information as the input of the standby encoder to obtain the output.
  • the terminal determines that there is no AI encoder to which the first information is applicable among the N AI encoders, the terminal may use the first information as the input of the standby encoder to obtain the output.
  • a specific process example is shown in Figure 9 below.
  • Example 1 Assuming that N is equal to 1, the first configuration information obtained by the terminal is S101. As shown in the second case, it may specifically include: the first preprocessed training data set (assuming CSI) of the group of AI codecs. Mean value H TRN , the information of the first preprocessing (that is, the first preprocessing includes: Fourier transform (which can be converted to the angle domain and the time delay domain), for truncation of some high time delay data), the judgment threshold H Th , The backup encoder that uses the codebook for encoding, and how the backup encoder is used (that is, the backup encoder is used when there is no AI encoder suitable for the information to be transmitted).
  • the first preprocessed training data set assuming CSI
  • Mean value H TRN the information of the first preprocessing (that is, the first preprocessing includes: Fourier transform (which can be converted to the angle domain and the time delay domain), for truncation of some high time delay data), the judgment threshold H Th , The backup encoder that uses
  • the manner in which the terminal determines whether the first information is suitable for the AI encoder to be applied according to the first configuration information is as shown in the case 1 of S102 in FIG. 5 above.
  • the terminal performs the above-mentioned first preprocessing on the first information. , that is, Fourier transform is performed on the first information to convert it into the angle domain and the time delay domain, and then part of the high time delay data is truncated to obtain the first processing information H 1 .
  • the terminal calculates the cosine similarity CS H between the first processing information H 1 and the above-mentioned mean value H TRN , and if CS H >H Th , the terminal can determine that the first information is applicable to the AI encoder, and determine the encoder ( That is, the first encoder) is the AI encoder. At this time, in S103, the terminal may use the first processing information H 1 as the input of the AI encoder to obtain the output H FB . If CS H ⁇ H Th , the terminal may determine that the AI encoder is not applicable to the first information, and determine that the used encoder is a backup encoder. At this time, in S103, the terminal may use the backup encoder to encode the first information.
  • the example of the terminal judging whether there is an AI encoder suitable for the information to be transmitted in the N AI encoders is similar to the above example, with the difference that: the terminal can also determine when the cosine similarity is greater than the judgment threshold.
  • the AI encoder with the largest cosine similarity is determined among the at least one AI encoder of the terminal, and the AI encoder is the most suitable AI encoder for the first information determined by the terminal, and is also the used encoder.
  • Example 2 Assuming that N is greater than 1, the first configuration information obtained by the terminal is S101. As shown in case 3, it may specifically include: an AI decider with N output ports (wherein the N output ports correspond to N groups of AI codes respectively.
  • the decoder which can also be understood as corresponding to N AI encoders), outputs the threshold Th, and judges whether there is information to be transmitted in the N AI encoders according to the output threshold and the output value of the AI determinator (assuming it is CSI) Applicable
  • the judgment method of the AI encoder that is, when the output value of an output port of the AI determinator is greater than the output threshold Th and greater than the output values of the other N-1 output ports, the information to be transmitted is applicable to the AI corresponding to the port.
  • the information of the first preprocessing that is, the first preprocessing includes: Fourier transform, truncation for some high-latency data
  • the information of the second preprocessing the second preprocessing is consistent with the first preprocessing
  • the backup encoder that uses the codebook for encoding
  • the usage mode of the backup encoder that is, the backup encoder is used when there is no AI encoder suitable for the information to be transmitted.
  • the manner in which the terminal determines whether there is an AI encoder applicable to the first information among the N AI encoders according to the first configuration information is as shown in the third case of S102.
  • the terminal performs the above-mentioned second preprocessing on the first information. , that is, performing Fourier transform on the first information to convert it into the angle domain and the time delay domain, and then truncating part of the high time delay data to obtain the second processing information H 2 .
  • the terminal uses the second processing information H 2 as the input of the AI decider, so as to obtain the output values I 1 , I 2 , . . . , I N of the N output ports.
  • the terminal can determine the first information It is applicable to the AI encoder corresponding to the i-th output port, and the encoder used is determined to be the AI encoder.
  • the terminal may perform the above-mentioned first preprocessing on the first information to obtain the first processing information H1, wherein the first processing information is consistent with the second preprocessing, so the first processing information H1 and the second processing information are consistent.
  • the second processing information H2 is the same. Then, the terminal can use the first processing information as the input of the AI encoder to obtain the output H FB .
  • the terminal may determine that the first information is not applicable to the N AI encoders, and determine that the used encoder is a backup encoder. At this time, in S103, the terminal may use the backup encoder to encode the first information.
  • S104 The terminal sends the second information and the first indication information to the network device.
  • the terminal may also not send the first indication information to the network device.
  • the first indication information may be used to indicate whether the terminal uses an AI encoder.
  • the first indication information is also used to indicate the AI encoder specifically used by the terminal, for example, including the identifiers of a group of AI codecs to which the first information determined by the terminal is applicable .
  • the first encoder is the AI encoder.
  • the first indication information is also used to indicate that the terminal uses a backup encoder, that is, the first encoder is a backup encoder.
  • the network device may also obtain according to the first indication information: the second information is obtained by the terminal using the first encoder to encode, optionally, the first encoder is the (most) applicable AI encoder determined by the terminal for the first information.
  • the network device determines a first decoder corresponding to the first encoder according to the first indication information.
  • S105 is an optional step.
  • the network device may determine the first encoder used by the terminal according to the first indication information, and optionally, determine the first decoder corresponding to the first encoder.
  • the first encoder is an AI encoder
  • the first decoder is an AI decoder.
  • the first indication information includes the identification of a group of AI codecs, and the network device can determine that the first encoder is an AI encoder in the group of AI encoders according to the identification, and determine the AI corresponding to the AI encoder according to the identification.
  • the decoder is the first decoder.
  • the first encoder is a backup encoder
  • the first decoder is a corresponding backup decoder.
  • the first indication information includes the identifier of the first encoder, and the network device can determine, according to the identifier, that the first encoder is an encoder using the codebook encoding scheme, and determine that the decoder corresponding to the encoder using the codebook decoding scheme is first decoder.
  • S106 The network device uses the first decoder to decode the second information.
  • S106 is an optional step.
  • the above S102-S106 are the first information transmission process, that is, an information transmission process, wherein S102 may be the first judgment performed by the terminal after acquiring the first configuration information.
  • the terminal determines, according to the first configuration information, whether there is an AI encoder applicable to the third information.
  • the terminal uses the second encoder to encode the third information and obtains the fourth information.
  • S109 The terminal sends the fourth information and the second indication information to the network device.
  • the network device determines a second decoder corresponding to the second encoder according to the second indication information.
  • S111 The network device uses the second decoder to decode the fourth information.
  • the above-mentioned S107-S111 are the transmission process of the second information, which is also an information transmission process, which is similar to the transmission process shown in S102-S106, except that the transmission time is different from the transmission time of the above-mentioned first information, wherein, S107 may be the second judgment performed by the terminal after acquiring the first configuration information.
  • the terminal can continuously obtain the information to be transmitted, and periodically or trigger the judgment process, that is, according to the first configuration information, determine whether there is an AI encoder suitable for the information to be transmitted, wherein the triggering process is performed.
  • the formula means that the judgment is made every time the information to be transmitted is obtained.
  • the terminal may select an encoder to use according to the judgment result, that is, the terminal may perform multiple information transmission processes.
  • each information transmission process is similar to the above-mentioned one transmission process, but since the information to be transmitted may be different, the judgment result of each judgment and the encoder used may be different.
  • the first information and the third information may not be identically distributed, then the judgment results of S102 and S107 may be different, and the first encoder and the second encoder may be different (the first decoder and the second decoder are also different).
  • each information transmission process please refer to S203-S207 in FIG. 6 , S302-S308 in FIG. 7 , S402-S408 in FIG. 8 , and S502-S506 in FIG. 9 .
  • the terminal can determine whether there is an AI encoder suitable for the information to be transmitted, that is, to determine whether the information to be transmitted and the training data set of the AI encoder and decoder are equally distributed, so as to avoid passing the AI encoder and decoder. Encoding and decoding the information that is not the same distribution as the training data set, resulting in information distortion and uncontrollable deterioration of the performance of the communication system.
  • the information to be transmitted is usually large, and the terminal can quickly and flexibly switch the encoder used according to different information to be transmitted. Compared with the terminal sending unencoded or high-fidelity encoded information to the network device, the network device can determine it. And indicate the encoder used by the terminal, the transmission overhead and delay of the terminal's own judgment are smaller.
  • the network device may send D AI encoders to the terminal before the terminal determines whether there is an AI encoder applicable to the information to be transmitted according to the first configuration information, where D is a positive integer, and D is less than or equal to N,
  • D is a positive integer
  • D is less than or equal to N
  • FIG. 6 is a schematic flowchart of another information encoding control method provided by an embodiment of the present application. The method may include but is not limited to the following steps:
  • S201 The network device sends D AI encoders to the terminal.
  • S201 is an optional step.
  • the method may further include: the terminal reports the performance parameters of the terminal to the network device, and the network device determines whether the terminal meets the performance requirements of the N AI encoders according to the performance parameters of the terminal.
  • the D AI encoders are AI encoders whose terminals meet the performance requirements determined by the network device among the N AI encoders, and D is less than or equal to N.
  • the network device may also directly send the N AI encoders obtained by the network device to the terminal, where D is equal to N in this case.
  • S202 The network device sends the first configuration information to the terminal.
  • the first configuration information may include statistical information of N AI encoders, or may include statistical information of D AI encoders.
  • the first configuration information reference may be made to the description of S101 in FIG. 5 , and details are not repeated here.
  • S203 The terminal determines, according to the first configuration information, that the first information is applicable to the first AI encoder among the D AI encoders.
  • S203 is an optional step.
  • the judgment process shown in S203 is similar to that of S102 in FIG. 5 above, except that the terminal can make judgment based on D AI encoders instead of N AI encoders, and the judgment result is that the terminal determines that there is a first AI encoder.
  • the AI encoder to which the information applies, and the AI encoder is the first AI encoder.
  • S204 The terminal uses the first AI encoder to encode the first information to obtain fifth information.
  • S204 is the same as S103 in Fig. 5 above, except that the first encoder in S103 is the first AI encoder in S204.
  • S205 The terminal sends the fifth information and the third indication information to the network device.
  • the terminal may also not send the third indication information to the network device.
  • the third indication information may be used to indicate that the terminal uses an AI encoder, and the AI encoder used by the terminal is the first AI encoder.
  • the network device may obtain according to the third indication information: the fifth information is obtained by the terminal using the first AI encoder, optionally, the first AI encoder is the (most) applicable AI code for the first information determined by the terminal device.
  • the network device determines the first AI decoder corresponding to the first AI encoder according to the third indication information.
  • S206 is an optional step.
  • the network device may determine the first AI encoder used by the terminal according to the first indication information, and optionally, determine the first AI decoder corresponding to the same group as the first AI encoder.
  • the network device uses the first AI decoder to decode the fifth information.
  • S207 is an optional step.
  • the terminal may also perform other information transmission processes.
  • the terminal may determine whether there is an AI encoder to which the third information is applicable among the D AI encoders according to the first configuration information.
  • the terminal can use the AI encoder to encode the third information, And send the information indicating that the AI encoder is used to the network device.
  • the terminal can send information to the network device to request the network device to configure the AI encoder, and then use the AI encoder to The third information is encoded, and the indication information that the AI encoder is used is sent to the network device, and the specific process is similar to S302-S308 in Fig. 7 below.
  • the terminal can use the backup encoder to encode the third information and send the network device an indication that the backup encoder is used. information, and the specific process is similar to S402-S408 in Figure 8 below.
  • the terminal can fall back to the standby encoding scheme according to the first configuration information, that is, use the standby encoder to encode the third information, and send the used For the indication information of the standby encoder, the specific process is similar to S502-S506 in Fig. 9 below.
  • the terminal may still determine whether there is an AI encoder suitable for the information to be transmitted based on N AI encoders.
  • the terminal determines that the information to be transmitted is suitable for the AI encoder X (which belongs to an AI encoder other than the D AI encoders among the N AI encoders)
  • the terminal can send information to the network device to request the network device to configure the AI Encoder X, and then use AI encoder X for encoding, the specific process is similar to S302-S308 in Figure 7 below, or the terminal can use an alternate encoder for encoding, and the specific process is similar to S402-S408 in Figure 8 below.
  • the terminal when the terminal determines that the information to be transmitted is suitable for the first AI encoder, and the terminal does not receive the first AI encoder, the terminal may send information to the network device to request the network device to configure the first AI encoder and then use the first AI encoder to encode the information to be transmitted.
  • an example of an information transmission process ie, S102-S106 in the above Figure 5 can be seen in Figure 7 below.
  • FIG. 7 is a schematic flowchart of another information encoding control method provided by an embodiment of the present application. The method may include but is not limited to the following steps:
  • S301 The network device sends first configuration information to the terminal.
  • S301 is an optional step.
  • S101 in FIG. 5 For the description of the first configuration information, reference may be made to the description of S101 in FIG. 5 , and details are not repeated here.
  • S302 The terminal determines, according to the first configuration information, that the first information is applicable to the first AI encoder.
  • S302 is an optional step.
  • the determination process shown in S302 is the same as that of S102 in FIG. 5 above, except that the determination result is that the terminal determines that there is an AI encoder applicable to the first information, and the AI encoder is the first AI encoder. Since the terminal has not received the first AI encoder, the terminal may send information to the network device to request the network device to configure the first AI encoder, that is, S303 is performed.
  • S303 The terminal sends fourth indication information to the network device.
  • S303 is an optional step.
  • the fourth indication information may include the identifier of the AI codec group where the first AI encoder is located, and the network device may determine, according to the identifier, that the encoder that the terminal requests to configure is the AI encoder in the group of AI codecs. .
  • S304 is an optional step.
  • the network device may first determine whether the terminal meets the performance requirements of the first AI encoder according to the performance parameters of the terminal.
  • the corresponding indication information for example, an indication that the performance requirements are not met, etc.
  • S305 The terminal uses the first AI encoder to encode the first information to obtain fifth information.
  • S306 The terminal sends the fifth information and the third indication information to the network device.
  • the terminal may also not send the third indication information to the network device.
  • the network device determines the first AI decoder corresponding to the first AI encoder according to the third indication information.
  • S307 is an optional step.
  • the network device uses the first AI decoder to decode the fifth information.
  • S308 is an optional step.
  • S305-S308 are the same as S204-S207 in Fig. 6 above, and are not repeated here.
  • the fourth indication information may further include the judgment result of S302, that is, it is used to indicate that the terminal determines that the first information is (most) suitable for the first AI encoder.
  • the terminal may also not send the third indication information in S306, and the network device may determine the AI encoder required by the terminal according to the fourth indication information sent by the terminal.
  • the terminal may also perform other information transmission processes.
  • the terminal may determine whether there is an AI encoder to which the third information is applicable among the N AI encoders according to the first configuration information.
  • the terminal may not need to send the information requesting the configuration of the first AI encoder to the network device. , but directly use the first AI encoder to encode the third information, and send the indication information that the first AI encoder is used to the network device, reducing the transmission delay.
  • the specific process is the same as S203-S207 in Figure 6 above. similar.
  • the terminal can send information to the network device again to request the network device to configure the AI encoder, and then use the AI encoder.
  • the third information is encoded, and the indication information that the AI encoder is used is sent to the network device.
  • the terminal can use the backup encoder to encode the third information and send the network device an indication that the backup encoder is used. information, and the specific process is similar to S402-S408 in Figure 8 below.
  • the terminal can fall back to the standby encoding scheme according to the first configuration information, that is, use the standby encoder to encode the third information, and send the used For the indication information of the standby encoder, the specific process is similar to S502-S506 in Fig. 9 below.
  • the terminal when the terminal determines that the information to be transmitted is suitable for the first AI encoder, and the terminal does not receive the first AI encoder, the terminal may first use the backup encoder to encode the information to be transmitted, thereby reducing The transmission delay is used to avoid affecting the subsequent information transmission process.
  • an example of an information transmission process ie, S102-S106 in the above Figure 5 can be seen in Figure 8 below.
  • FIG. 8 is a schematic flowchart of another information encoding control method provided by an embodiment of the present application. The method may include but is not limited to the following steps:
  • S401 The network device sends first configuration information to the terminal.
  • S401 is an optional step.
  • S101 in FIG. 5 For the description of the first configuration information, reference may be made to the description of S101 in FIG. 5 , and details are not repeated here.
  • S402 The terminal determines, according to the first configuration information, that the first information is applicable to the first AI encoder.
  • S402 is an optional step.
  • the determination process shown in S402 is the same as that of S102 in FIG. 5 above, except that the determination result is that the terminal determines that there is an AI encoder applicable to the first information, and the AI encoder is the first AI encoder. Since the terminal has not received the first AI encoder, the terminal may send the network device information to request the network device to configure the first AI encoder, that is, perform S405.
  • S403 The terminal uses the standby encoder to encode the first information and obtains the sixth information.
  • S404 The terminal sends sixth information and fifth indication information to the network device.
  • the fifth indication information may be used to indicate that the terminal uses an alternate encoder, optionally, does not use an AI encoder.
  • the network device may obtain according to the fifth indication information: the sixth information is obtained by the terminal using the standby encoder.
  • the terminal may also not send the fifth indication information to the network device.
  • S405 The terminal sends fourth indication information to the network device.
  • S405 is an optional step.
  • the fourth indication information may include the identifier of the AI codec group where the first AI encoder is located, and the network device may determine, according to the identifier, that the encoder that the terminal requests to configure is the AI encoder in the group of AI codecs. .
  • the terminal recognizes that the information to be transmitted in the near future is not much different. For example, the terminal has been indoors in the gym for 10 minutes. When the user moves indoors in the gym, the CSI to be transmitted by the terminal is equally distributed. Therefore, the terminal can request the network device to configure the first AI encoder through the fourth indication information, so that the information applicable to the first AI encoder can be directly encoded by the first AI encoder in the future, without the need to request again, reducing the transmission time. extension.
  • S406 is an optional step.
  • the network device may first determine whether the terminal meets the performance requirements of the first AI encoder according to the performance parameters of the terminal.
  • the corresponding indication information for example, an indication that the performance requirements are not met, etc.
  • the terminal determines that the first information is (most) suitable for the first AI encoder, since the terminal has not received the first AI encoder, the terminal can use the standby encoder to encode the first information (that is, perform S403). ), optionally, information may be sent to the network device to request to configure the first AI encoder (ie, perform S405).
  • S406 is after S403 and S405, but the specific moment when the network device sends the first AI encoder is not limited, for example, the sequence of S406 and S404, S407, and S408 is not limited.
  • the sequence of S403 and S405 is not limited.
  • the terminal may perform encoding first and then request to configure the first AI encoder, that is, S403 is before S405, but the order in which the terminal sends the encoded sixth information and requests to configure the first AI encoder is not limited. That is, the order of S404 and S405 is not limited.
  • the terminal may first send information to the network device to request to configure the first AI encoder, and perform encoding before receiving the first AI encoder, that is, S405 is before S403, and S406 is after S403.
  • the network device determines a backup decoder corresponding to the backup encoder according to the fifth indication information.
  • S407 is an optional step.
  • S408 The network device uses the backup decoder to decode the sixth information.
  • S408 is an optional step.
  • the terminal may also perform other information transmission processes.
  • the terminal may determine, according to the first configuration information, whether there is an AI encoder to which the third information is applicable among the N AI encoders.
  • the terminal may not need to send information requesting the configuration of the first AI encoder to the network device, Instead, the first AI encoder is directly used to encode the third information, and the indication information that the first AI encoder is used is sent to the network device, which reduces the transmission delay.
  • the specific process is similar to S203-S207 in Figure 6 above. .
  • the terminal can send information to the network device to request to configure the AI encoder, and then use the AI encoder for the third AI encoder.
  • the information is encoded, and the indication information that the AI encoder is used is sent to the network device.
  • the specific process is similar to S302-S308 in Figure 7 above.
  • the terminal can use the backup encoder to encode the third information and send the network device an indication that the backup encoder is used. information, reduce the transmission delay, and ensure that the process of information coding and feedback is not interrupted.
  • the terminal can fall back to the standby encoding scheme according to the first configuration information, that is, use the standby encoder to encode the third information, and send the used For the indication information of the standby encoder, the specific process is similar to S502-S506 in Fig. 9 below.
  • the terminal when the terminal determines that there is no AI encoder suitable for the information to be transmitted among the N AI encoders, the terminal can use a backup encoder to encode the information to be transmitted, so as to avoid misuse of the AI encoder to bring about
  • the example of an information transmission process ie, S102-S106 in the above Figure 5 can be seen in Figure 9 below.
  • FIG. 9 is a schematic flowchart of another information encoding control method provided by an embodiment of the present application. The method may include but is not limited to the following steps:
  • S501 The network device sends first configuration information to the terminal.
  • S501 is an optional step.
  • S501 is an optional step.
  • S101 in FIG. 5 For the description of S101 in FIG. 5 , and details are not repeated here.
  • S502 The terminal determines, according to the first configuration information, that there is no AI encoder applicable to the first information.
  • S502 is an optional step.
  • the determination process shown in S502 is the same as that of S102 in FIG. 5 above, except that the determination result is that the terminal determines that there is no AI encoder to which the first information is applicable.
  • S503 The terminal uses the standby encoder to encode the first information and obtain sixth information.
  • S504 The terminal sends sixth information and fifth indication information to the network device.
  • the fifth indication information may be used to indicate that the terminal uses an alternate encoder, optionally, does not use an AI encoder.
  • the network device may obtain according to the fifth indication information: the sixth information is obtained by the terminal using the standby encoder.
  • the terminal may also not send the fifth indication information to the network device.
  • the network device determines a backup decoder corresponding to the backup encoder according to the fifth indication information.
  • S505 is an optional step.
  • S506 The network device uses the backup decoder to decode the sixth information.
  • S506 is an optional step.
  • the backup encoder can be an encoder using traditional encoding schemes such as codebooks, or other AI encoders with better generalization. If the backup encoder is another AI encoder with better generalization, the backup encoder may be sent by the network device before the terminal determines whether there is an AI encoder suitable for the information to be transmitted, or it may be the terminal that determines whether there is an AI encoder to be transmitted. The information is sent by the network device after the applicable AI encoder.
  • the terminal may also perform other information transmission processes.
  • the terminal may determine, according to the first configuration information, whether there is an AI encoder to which the third information is applicable among the N AI encoders.
  • the terminal may directly use the AI encoder to configure the AI encoder without sending the information requesting the configuration of the AI encoder to the network device.
  • the third information is encoded, and the indication information that the AI encoder is used is sent to the network device, and the specific process is similar to S203-S207 in Fig. 6 above.
  • the terminal can send information to the network device to request to configure the AI encoder, and then use the AI encoder for the third AI encoder.
  • the information is encoded, and the indication information that the AI encoder is used is sent to the network device.
  • the specific process is similar to S302-S308 in Figure 7 above.
  • the terminal can use the standby encoder to encode the third information, and send the information using the standby encoder to the network device.
  • Indication information the specific process is similar to S402-S408 in Fig. 8 above.
  • the terminal continues to use the backup encoder to encode the third information, and sends the indication information that the backup encoder is used to the network device.
  • the terminal may, after judging whether there is an AI encoder suitable for the information to be transmitted (for example, after S102 in FIG. 5 above), the judgment result obtained in this judgment process, optionally and related The parameters are reported to the network device, and the network device decides the encoder used by the terminal.
  • an AI encoder suitable for the information to be transmitted for example, after S102 in FIG. 5 above
  • the parameters are reported to the network device, and the network device decides the encoder used by the terminal.
  • Figure 10 A specific example is shown in Figure 10 below.
  • FIG. 10 is a schematic flowchart of another information encoding control method provided by an embodiment of the present application. The method may include but is not limited to the following steps:
  • S601 The network device sends first configuration information to the terminal.
  • the first configuration information may further include judgment criteria.
  • the method may further include: the network device and the terminal negotiate to determine the judgment criterion, that is, the judgment criterion is predefined.
  • the method may further include: the terminal receives the judgment criterion sent by the network device.
  • the judgment criterion is a judgment process for the terminal to judge whether there is an AI encoder suitable for the information to be transmitted according to the judgment threshold (for example, S102 in Fig. 5 above, that is, S602).
  • the judgment criterion may be a soft decision criterion or a hard decision criterion.
  • S602 The terminal determines whether there is an AI encoder applicable to the first information according to the first configuration information.
  • S602 is an optional step.
  • S602 is similar to S102 of Fig. 5 above, the difference is: in S102 of Fig. 5 above, the terminal judges and decides the encoder that is finally used by itself, so the judgment criterion is a hard judgment criterion by default, and the final judgment result is is the decision result (an encoder, i.e. the encoder used). In S602, the terminal only judges by itself but does not make a decision.
  • the judgment result is also an encoder, but when the judgment criterion is a soft judgment criterion, there can be multiple judgment results, and any judgment result corresponds to one AI encoder, indicating whether the information to be transmitted is suitable for this AI encoder.
  • S603 The terminal sends the first judgment information to the network device.
  • the first judgment information may include the judgment result, and optionally related parameters obtained in the judgment process (ie S602 ), such as the obtained cosine similarity, probability parameter, and output value of the AI decider.
  • the first judgment information may include: the identifier of the AI codec group where the AI encoder is located and An indication that the information to be transmitted is applicable to this AI encoder (ie, the judgment result). If the terminal determines that there is no AI encoder applicable to the information to be transmitted, the first judgment information may include: an indication (ie, a judgment result) indicating that the information to be transmitted is not applicable to any AI encoder.
  • the first judgment information may include: an indication of whether the information to be transmitted is applicable to N AI encoders or the information to be transmitted is applicable to N AI codes.
  • the probability value of the device that is, the N judgment results
  • the first judgment information may include: indications of S AI encoders to which the information to be transmitted is applicable (that is, S judgment results, where S is less than or equal to N) and indications that the information to be transmitted is applicable to these S AI codes. indicator of the degree of the device.
  • the first judgment information may include: the information to be transmitted is applicable to 2 AI encoders (AI encoder 1 and AI encoder 2 respectively), wherein the information to be transmitted is applicable to AI encoder 1 and AI encoder 2
  • the indexes of the degree of 1 and 2 are respectively: 1 and 2, wherein the larger the index value is, the higher the degree of applicability is, so the information to be transmitted is more suitable for AI encoder 2.
  • the method for the terminal to judge whether there is an AI encoder to which the first information is applicable is shown in S102 of the above Fig. 5 Scenario 1, the first judgment information may also include the obtained cosine similarity, for example, the first information and N The largest S among the cosine similarities of the statistical information of the AI encoder, where S is a positive integer, and 1 ⁇ S ⁇ N.
  • the method for the terminal to judge whether there is an AI encoder applicable to the first information is as shown in the second case of S102 in FIG. 5
  • the first judgment information may also include the acquired probability parameter, for example, the first information passes through N AI encoders.
  • the largest T among the N PDFs calculated by the statistical information of the encoder where T is a positive integer and 1 ⁇ T ⁇ N.
  • the method for the terminal to determine whether there is an AI encoder to which the first information is applicable is shown in S102 in S102 in the above Figure 5.
  • the first determination information may also include an output value of at least one output port of the AI determiner, for example, The output values of the N output ports, or the output values greater than the output threshold, or the largest Y output values among the N output values, where Y is a positive integer, and 1 ⁇ Y ⁇ N.
  • the network device determines the third encoder according to the first judgment information.
  • S604 is an optional step.
  • the third encoder is an AI encoder or a backup encoder.
  • the network device can be based on the actual situation (for example, whether the terminal meets the performance requirements of the AI encoder). ) decides whether the encoder used by the terminal is the AI encoder or an alternate encoder.
  • the network device may decide that the encoder used by the terminal is a backup encoder.
  • the network device can decide the encoder used by the terminal based on the actual situation (for example, whether the terminal meets the performance requirements of the AI encoder) and the first judgment information.
  • the manner of the encoder used in the terminal decision in S102 is the same, and details are not repeated here.
  • the network device sends the sixth indication information and the third encoder to the terminal.
  • S605 is an optional step.
  • the sixth indication information is used to instruct the terminal to use the third encoder for encoding.
  • S606 The terminal uses the third encoder to encode the first information to obtain seventh information.
  • the third encoder is an AI encoder
  • the terminal may perform first preprocessing on the first information, and then input the first preprocessed information into the AI encoder to obtain the output seventh information .
  • the third encoder is an AI encoder
  • the sixth indication information may include first preprocessing information corresponding to the AI encoder, and the terminal may first perform first preprocessing on the first information, and then The first preprocessed information is then input into the AI encoder to obtain the output seventh information.
  • S607 The terminal sends seventh information to the network device.
  • the network device uses a third decoder corresponding to the third encoder to decode the seventh information.
  • S608 is an optional step.
  • the network device since the network device sends the sixth indication information to the terminal, the network device may default that the third encoder indicated in the sixth indication information is the encoder used by the terminal.
  • the network device may directly use the third encoder.
  • the third decoder corresponding to the three encoders decodes the seventh information.
  • S607 may further include information indicating that the encoder used when the terminal encodes the seventh information is the third encoder, and the network device may determine the third decoder corresponding to the third encoder according to the information, And use the third decoder to decode the seventh information.
  • the network device may send the AI encoder to the terminal before the terminal determines whether there is an AI encoder applicable to the information to be transmitted according to the first configuration information. For example, before the above S601, the network device sends D AI encoders to the terminal, then in S605, the network device may not need to send a third encoder (AI encoder), the specific process is similar to that in Figure 6 above, and will not be repeated.
  • AI encoder the third encoder
  • the terminal may send information to the network device after receiving the sixth indication information to request to configure the third encoder
  • the third encoder for example, after S605 and before S606, the instruction information requesting to configure the third encoder is sent to the network device.
  • the specific process is similar to that in FIG. 7 above, and will not be repeated here.
  • the third encoder is a backup encoder, and the network device may not send the third encoder in S605.
  • the terminal may also perform other information transmission processes.
  • the terminal can determine the encoder used by the network device to determine the terminal as in S602-S608 above, or it can be determined by the terminal as in S102-S106 in Figure 5 above.
  • S203-S207 in Fig. 6, S302-S308 in Fig. 7, S402-S408 in Fig. 8, and S502-S506 in Fig. 9 are shown.
  • the encoder used by the terminal may also be decided by the network device to reduce the processing pressure of the terminal. Moreover, the terminal does not need to send the information indicating the used encoder, thereby reducing the transmission overhead. If the encoder used by the decision-making terminal is an AI encoder that is not sent by the network device, the network device can directly send the AI encoder without the terminal sending the information requesting the configuration of the AI encoder, thereby reducing transmission overhead.
  • network devices can be trained to generate AI codecs.
  • the training data set used by the network device to train the AI codec may be the information to be transmitted sent by the terminal to the network device, and the terminal may filter the information to be transmitted, in order to train the AI codec with high value data (not yet available). Encoded or high-fidelity encoding) is sent to the network device, thereby improving the generalization of the AI codec.
  • Figure 11 A specific example is shown in Figure 11 below.
  • FIG. 11 is a schematic flowchart of a screening method provided by an embodiment of the present application.
  • the method can be applied to the communication system shown in FIG. 1 , and can also be applied to the scenario shown in FIG. 2 .
  • the terminal in this method may be the terminal 100 shown in FIG. 3 .
  • the network device in this method may be the network device 200 shown in FIG. 4 .
  • the method may include but is not limited to the following steps:
  • S701 The network device sends first configuration information to the terminal.
  • the first configuration information may further include a judgment criterion for the terminal to judge whether to send the information to be transmitted, and the judgment criterion may be a hard decision criterion or a soft decision criterion.
  • the terminal may directly determine whether to send the information to be transmitted according to whether the information to be transmitted is suitable for the AI codec.
  • the terminal may determine the sending probability of the information to be transmitted according to a preset rule, and then send the information to be transmitted according to the sending probability.
  • the first configuration information may also include the above-mentioned preset rules, such as the mapping relationship between the relevant parameters obtained in the judgment process (such as cosine similarity, probability parameter or the output value of the AI judger) and the transmission probability (such as proportional,
  • the mapping relationship can be represented by functions and/or parameters).
  • the above preset rules are pre-negotiated and configured by the terminal and the network device.
  • S702 The terminal determines to send the eighth information according to the first configuration information.
  • S702 is an optional step.
  • the terminal may judge whether the information to be transmitted can be sent according to the first configuration information, that is, whether the information to be transmitted is of high value for training the AI codec.
  • the terminal may judge whether the information to be transmitted can be sent according to the first configuration information, that is, whether the information to be transmitted is of high value for training the AI codec.
  • the terminal may judge whether the information to be transmitted can be sent according to the first configuration information, that is, whether the information to be transmitted is of high value for training the AI codec.
  • you want to improve the generalization of the AI codec you need to set the input to data that is not the same distribution as the existing training data set when training the AI codec, which means that the AI codec is not applicable. data is more valuable for training AI codecs.
  • the terminal can judge whether the information to be transmitted is applicable to the AI codec according to the first configuration information, and if not applicable, then The information to be transmitted can be sent to the network device, and the judgment method is similar to the judgment method shown in S102 of FIG. 5 above, but the required judgment result is opposite.
  • the terminal may determine the transmission probability of the information to be transmitted according to the relevant parameters obtained in the judgment process based on preset rules, wherein , the larger the above related parameters, the lower the transmission probability.
  • Example 1 Assume that the first configuration information is shown in the second case of S101, and includes the mean value of the training data set that has undergone the first preprocessing. N is equal to 1, and a set of judgment thresholds predefined or indicated in the first configuration information includes a judgment threshold CS t .
  • the terminal may first perform first preprocessing on the information to be transmitted, and then calculate the cosine similarity CS 0 between the first preprocessed information and the mean.
  • the terminal can determine that the information to be transmitted is not applicable to the set of AI codecs, that is, to train this set of AI codecs The value of the device is high, so it can be determined to send the information to be transmitted, otherwise it is determined not to send the information to be transmitted. If the above judgment criterion in the first configuration information is a soft decision criterion, the terminal may determine the transmission probability of the information to be transmitted according to the cosine similarity CS 0 obtained by the above calculation.
  • the terminal may determine whether to transmit the information to be transmitted according to the sampling of the transmission probability P.
  • the terminal may also send the transmission probability P to the network device, and the network device determines whether to allow the terminal to send the eighth information according to sampling of the transmission probability P. For example, when the sample value characterizes the transmission, the information to be transmitted is transmitted; otherwise, the information to be transmitted is not transmitted.
  • Example 2 It is assumed that the first configuration information is shown in the first case of S101, and includes the complete distribution parameters of the mathematical distribution of the training data set. N is equal to 1, and a set of judgment thresholds predefined or indicated in the first configuration information includes two judgment thresholds PDF t1 and PDF t2 .
  • the terminal can calculate the PDF of the information to be transmitted in the mathematical distribution, that is, PDF 0 .
  • the terminal can determine that the information to be transmitted is applicable to this group of AI codecs, but apply this group of AI codecs The effect of encoding and decoding the information to be transmitted is expected to be poor; if PDF 0 >PDF t2 , the terminal can determine that the information to be transmitted is suitable for this group of AI codecs, and apply this group of AI codecs to the information to be transmitted. The effect of encoding and decoding is expected to be good. In both cases, the terminal can determine not to send the information to be transmitted.
  • the terminal can determine that the information to be transmitted is not applicable to this group of AI codecs, that is, the information to be transmitted is information of high value for training this group of AI codecs, so it can determine to send the information to be transmitted. information. If the above judgment criterion in the first configuration information is a soft decision criterion, the terminal may determine the sending probability of the information to be transmitted according to a preset rule.
  • the terminal may determine whether to transmit the information to be transmitted according to the sampling of the transmission probability P.
  • the terminal may also send the transmission probability P to the network device, and the network device determines whether to allow the terminal to send the information to be transmitted according to the sampling of the transmission probability P. For example, when the sample value characterizes the transmission, the information to be transmitted is transmitted; otherwise, the information to be transmitted is not transmitted.
  • Example 3 Assume that the first configuration information is S101 as shown in Case 3, and N is greater than 1.
  • the terminal can set the input of the AI decider as the information to be transmitted, so as to obtain the outputs I 1 , I 2 , . . . , IN of the N output ports of the AI decider.
  • the terminal can determine that the information to be transmitted is not applicable to the group of AI codecs, that is, the value of training the group of AI codecs is high, so the terminal can determine to send the information to be transmitted.
  • the terminal may determine whether to transmit the information to be transmitted according to the sampling of the transmission probability P.
  • the terminal may also send the transmission probability P to the network device, and the network device determines whether to allow the terminal to send the information to be transmitted according to the sampling of the transmission probability P. For example, when the sample value characterizes the transmission, the information to be transmitted is transmitted; otherwise, the information to be transmitted is not transmitted.
  • the terminal determines that the eighth information is data of high value for training the AI codec, so the terminal may request the network device to send the eighth information, that is, perform S702.
  • S703 The terminal sends first notification information to the network device.
  • the first notification information may be used to request the network device to send data, for example, to request transmission resources.
  • the first notification information may specifically include the judgment result obtained in the above judgment process (ie, S702 ), that is, the eighth information to be transmitted is of high value for training the AI codec.
  • the first notification information may specifically include the transmission probability obtained in the above judgment process (ie, S702).
  • the first notification information may specifically include relevant parameters obtained in the above judgment process (ie S702 ), for example: the calculated cosine similarity, probability parameters (such as the above PDF 0 or PMF 0 ), the output value of the AI decider .
  • the network device In response to the first notification information, the network device sends the second notification information to the terminal.
  • S704 is an optional step.
  • the network device may determine whether to allow the first notification information (for example, including the judgment result obtained in S702, the transmission probability, and/or related parameters) and the actual situation of resource utilization (for example, whether there is an available uplink transmission resource).
  • the terminal sends data.
  • the first notification information includes the judgment result obtained in S702, and the network device may allow the terminal to send data when determining that the transmission resource is available.
  • the first notification information includes the transmission probability obtained in S702, and the network device may determine whether to allow the terminal to send the eighth information according to sampling of the transmission probability.
  • the first notification information includes the relevant parameters obtained in S702, and the network device can judge whether the eighth information is data of high value for training the AI codec according to the relevant parameters reported by the terminal, that is, whether it can be sent, and the judgment method can be the same as that in S702.
  • the judgment process shown is the same, except that the terminal can only calculate the relevant parameters in S702, but does not judge whether the eighth information is valuable data for training the AI codec, that is, S702 does not need to obtain the judgment result.
  • the network device may allow the terminal to send the eighth information when the uplink transmission resource is available.
  • the second notification information may be used to indicate that the terminal is allowed to send data, optionally and related information of transmission resources allocated by the network device to the terminal, such as an identifier of a transmission frequency band.
  • the second notification information may also be used to instruct the terminal to reject data transmission by the terminal.
  • the terminal may cancel data transmission when receiving the second notification information.
  • S705 The terminal sends eighth information to the network device.
  • the terminal can send the uncoded eighth information or the high-fidelity encoded eighth information to the network device to ensure the integrity of the data and improve the training AI. Codec accuracy.
  • S706 The terminal determines not to send the ninth information according to the first configuration information.
  • S706 is an optional step.
  • the above S702-S705 may be the first time after the terminal obtains the first configuration information to determine whether to send the information to be transmitted, and the result of the determination is that it is determined to send, and the above S706 may be the second time after the terminal obtains the first configuration information. It is judged whether to send the information to be transmitted, and the judgment result is that it is determined not to send.
  • the terminal may continuously obtain the information to be transmitted, and periodically or trigger multiple determinations, that is, determine whether to loop the information to be transmitted according to the first configuration information. Wherein, each judgment is similar to the above-mentioned first judgment and second judgment, but since the information to be transmitted may be different, the judgment result obtained each time may be different, for example, it is determined to send or not to send.
  • the information to be transmitted is usually large, and the terminal can filter the information to be transmitted, so as to send the data with high value for training the AI codec to the network device, so as to avoid the terminal from sending data for training AI codecs.
  • Data with low codec value is sent to network equipment, and the performance of AI codec is poor due to the unbalanced training data set, which also avoids the waste of uplink bandwidth.
  • the above-mentioned information of the AI codec (for example, the identifier of the AI codec) can also be understood as the information of the AI encoder in the AI codec (for example, the identifier of the AI encoder), or can also be understood as In order to be the information of the AI decoder in the AI codec (for example, the identifier of the AI decoder), it can also be understood in reverse.
  • the AI decider sent by the network device to the terminal may be generated by independent training, and a specific example is shown in Figure 12 below.
  • FIG. 12 exemplarily shows a generation process of an AI decider.
  • the training data set can be divided into two sets: R(1) and R(2), where R(1) can be the same distributed data as the training data set of this group of AI codecs, for example, it is the set of training data sets of this group of AI codecs, R(2) can be It is data that is not the same distribution as the training data set of this group of AI codecs, such as a set of training data sets for other groups of AI codecs.
  • the AI model of the AI decider can be set to at least one layer of neural network, and the last layer of the AI model is assumed to be set to a sigmoid function, and the value range of the output value is [0, 1].
  • the input data type of the AI decider can be set to CSI, and the output port can be one.
  • the output can be set to 1 when the input is set to the data in R(1), and the output can be set when the input is set to the data in R(2) is 0.
  • the information to be transmitted may be set as the input of the AI decider to obtain the output value of the AI decider.
  • the output value is greater than the output threshold (for example, 0.5), it can be indicated that the information to be transmitted is suitable for the AI encoder, otherwise the information to be transmitted is not suitable for the AI encoder.
  • the training data set can be divided into N sets: H(1), H(2), .
  • a set of sets, i is a positive integer, and the value range of i is [1, N].
  • the AI model of the AI decider can be set to at least one layer of neural network, and the last layer of the AI model is assumed to be set to the Softmax function, and the output value range of each output port is [0, 1] , and the sum of all port output values is 1.
  • the input data type of the AI decider can be set to CSI, and the output ports can be N: port 1, port 2, ...
  • port N where port i can correspond to the ith group of AI codecs, that is, output ports
  • the output value of i characterizes whether the information to be transmitted is suitable for the i-th group of AI codecs.
  • the output value of output port i when training the AI decider, when the input is set to the data in H(i), the output value of output port i can be set to 1, and the output values of other output ports can be set to 0.
  • the output value of output port 1 is set to 1, and the output value of the other N-1 output ports is set to 0.
  • the AI decider obtained in this way can also be understood as an N classification decider.
  • the information to be transmitted may be set as the input of the AI decider, so as to obtain the output values of the N output ports of the AI decider.
  • the most suitable AI codec for the information to be transmitted can be determined by comparing the output values of the N output ports, for example, when the output value of output port i is greater than the output threshold (for example, 0.5), and greater than the other N-1 output ports
  • the output value of can characterize the information to be transmitted that is most suitable for the i-th group of AI codecs.
  • the above A is “greater than” B, and can also be replaced with (-A) “less than” (-B).
  • the above-mentioned “greater than” a certain threshold value can also be replaced with “greater than or equal to”, or vice versa.
  • the above-mentioned “less than” a certain threshold can also be replaced with “less than or equal to”, or vice versa.
  • the AI decider sent by the network device to the terminal may be obtained by migration learning, and a specific example is shown in Figure 13 below.
  • Fig. 13 exemplarily shows a generation process of yet another AI decider.
  • N AI encoders (AI encoder 1, AI encoder 2, . . . , AI encoder N) can share a shared backbone network with the AI decider, optional It is possible to uniformly train an AI codec with all the data, and then keep the shallow network of the AI encoder part in the AI codec (such as the first F layer network, F is a positive integer, and F is smaller than the AI encoder's total number of layers) as the shared backbone network.
  • the shared backbone network unchanged (that is, the parameters will not be updated in subsequent training), then connect to different AI encoder sub-networks and AI decoders respectively, and use the training data sets corresponding to each group of AI encoders and decoders.
  • the training generates the final sets of AI codecs.
  • keep the shared backbone network unchanged connect to the AI decider sub-network later, and use all data to train to generate an N-class decider (ie, AI decider). That is, the training of the shared backbone network needs to be completed before the training of each sub-network.
  • the shared backbone network can be kept unchanged to train the sub-network k. For a specific example, see (B) of FIG.
  • the shared backbone network may include at least one layer (optionally, any layer may also be a module including multiple layers), and the sub-network may be one layer (optionally, this layer may also be a layer including multi-layer modules).
  • the last layer of the sub-network k can be set as the Softmax function.
  • the output value of output port i can be set to 1
  • the output value of other output ports can be set to 0.
  • the output value of output port N is set to 1
  • the output value of the other N-1 output ports is set to 0.
  • only the parameters of the sub-network k are updated during the training process, and the parameters of the shared backbone network are not updated.
  • the information to be transmitted may be set as the input of the AI decider to obtain the output values of the N output ports of the AI decider.
  • the most suitable AI encoder for the information to be transmitted can be determined by comparing the output values of the N output ports. For a specific example, please refer to the description of (B) of FIG. 12 above.
  • AI encoder i and AI decoder i corresponding to the same group are trained together to form a large AI model.
  • the input is set to H(i)
  • the output is also set to the above-mentioned H(i), that is, the input data and the output data are the same.
  • the parameters of the sub-network i and the AI decoder i are updated during the training process, and the parameters of the shared backbone network are not updated.
  • the corresponding AI encoder i and AI decoder i can be obtained.
  • a second preprocessing may be performed on the training data set, and then the second preprocessed data is used as the input of the AI decider.
  • the information to be transmitted may also be subjected to second preprocessing first, and then the second preprocessed information may be used as the input of the AI decider to obtain the output.
  • the network device after the network device sends the AI decider or any AI encoder to the terminal, if it subsequently sends the AI decider or other AI encoder to the terminal, it can only send the corresponding sub-network without sending the entire AI model, reducing Small transmission resources.
  • the terminal uses the AI judging device to determine whether there is any AI encoder suitable for the information to be transmitted (referred to as information B), that is, the information B has been processed by the shared backbone network (the processed information is abbreviated as information C), then
  • the subsequent terminal uses any AI encoder to encode the information B, it does not need to process the information B through the entire AI model of the AI encoder, and can directly input the information C into the sub-network of the AI encoder, which greatly reduces the calculated amount.
  • AI codecs can also be independently trained and generated.
  • FIG. 14 A specific example is shown in FIG. 14 below.
  • FIG. 14 exemplarily shows a generation process of an AI codec.
  • AI encoder i and AI decoder i corresponding to the same group are trained together to form a large AI model.
  • the AI model is CsiNet for implementing CSI encoding and decoding.
  • H(i) may be CSI or preprocessed CSI
  • the output H'(i) of the AI encoder may be CSI encoding information or an index corresponding to the encoding information.
  • H'(i) The dimension of '(i) is greatly reduced.
  • the output H'(i) of the AI encoder can be used as the input of the AI decoder again, and the output is the recovered CSI.
  • the parameters of both AI encoder i and AI decoder i can be updated, or only the parameters of AI encoder i or AI decoder i can be updated as required.
  • the corresponding AI encoder i and AI decoder i can be obtained.
  • the terminal establishes a connection with the base station after entering the range covered by the base station, and the base station can send the first configuration information to the terminal after the connection.
  • the terminal In the case of maintaining the connection with the base station, it is assumed that the terminal is located in the geographic location A and indoors at the first moment, and at this time, the terminal needs to send the first CSI to the base station.
  • the terminal can determine whether there is an AI encoder suitable for the first CSI according to the first configuration information. Assuming that the first CSI is most suitable for the first AI encoder, the terminal can send information requesting the configuration of the first AI encoder to the base station, and then use The first AI encoder encodes the first CSI.
  • the terminal may send the encoded CSI and information indicating that the first AI encoder is used to the base station, and the base station may use the AI decoder corresponding to the first AI encoder to decode to obtain the first CSI reported by the terminal.
  • the terminal may move, and it is assumed that the terminal is located at the geographic location B and is located outdoors at the second moment, and at this time, the terminal needs to send the second CSI to the base station.
  • the terminal may determine whether there is an AI encoder applicable to the second CSI according to the first configuration information, and if there is no AI encoder applicable to the second CSI, the terminal may use a backup encoder to encode the second CSI. Then the terminal can send the encoded CSI and the information indicating that the backup encoder is used to the base station, and the base station can use the backup decoder corresponding to the backup encoder to decode to obtain the second CSI reported by the terminal.
  • the information of the AI codec (for example, the above-mentioned first configuration information) and the AI encoder received by the terminal may not be sent by the network device, but sent by other devices, such as the network.
  • the device is a core network device
  • the information of the AI codec and the AI encoder can be sent to the base station first, and then the base station sends it to the terminal according to the actual situation.
  • the input of the AI encoder can be information to be transmitted by the terminal, or information obtained after the information to be transmitted by the terminal is subjected to the first preprocessing, or information to be transmitted by the terminal obtained through other processing, such as
  • the precoding matrix indicator (PMI) obtained after the CSI is decomposed by the matrix, or the correlation matrix of the CSI, etc., are not limited in this application.
  • the input of the AI decider may be the information to be transmitted by the terminal, or the information obtained after the second preprocessing of the information to be transmitted by the terminal, or the information to be transmitted by the terminal obtained through other processing, such as
  • the PMI obtained after the CSI is decomposed by the matrix, or the correlation matrix of the CSI, etc., are not limited in this application.
  • some modules in the AI decider may be AI models, and other modules may be other processing modules, that is, some or all modules in the AI decider are AI models.
  • the terminal in this embodiment of the present application may also be other devices, such as a base station, core network device, etc., and the description of the execution operation can refer to the description of the terminal.
  • the network device may also be other devices, such as a terminal, etc. For relevant descriptions of performing operations and the like, please refer to the description of the network device.

Landscapes

  • Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Quality & Reliability (AREA)
  • Theoretical Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Medical Informatics (AREA)
  • Evolutionary Computation (AREA)
  • Databases & Information Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Power Engineering (AREA)
  • Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • Acoustics & Sound (AREA)
  • Physics & Mathematics (AREA)
  • Human Computer Interaction (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Computational Linguistics (AREA)
  • Compression, Expansion, Code Conversion, And Decoders (AREA)

Abstract

本申请实施例提供一种信息编码的控制方法，应用于通信技术领域的终端，该方法包括：接收第一配置信息，第一配置信息用于配置N个AI编码器的N组参数或AI判决器，AI判决器用于确定N个AI编码器中第一信息适用的AI编码器，和/或用于确定N个AI编码器不适用于第一信息编码；向网络设备发送第一指示信息，第一指示信息用于指示第一编码器用于第一信息编码，第一编码器是根据第一配置信息和第一信息确定的，第一编码器是N个AI编码器中的编码器，或者是和N个AI编码器不同的第二编码器。本申请实施例可以快速高效地选择合适的编码器，避免待传输的信息经过编码和解码后出现失真，提高传输质量。

Description

信息编码的控制方法及相关装置
本申请要求于2021年02月24日提交中国专利局、申请号为202110206769.9、申请名称为“一种控制方法、终端及网络设备”的中国专利申请的优先权，本申请要求于2021年04月06日提交中国专利局、申请号为202110376449.8、申请名称为“信息编码的控制方法及相关装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及通信技术领域，尤其涉及一种信息编码的控制方法及相关装置。
背景技术
在无线通信系统中，两个设备之间往往会传输一些测量信息、状态信息等，以此实现后续的信号传输，这样可以提高通信系统性能。例如，两个设备可以分别为发送信号的发送设备(如基站)，以及接收信号的接收设备(如终端)。接收设备可以将信道状态信息(channel state information，CSI)反馈给发送设备，发送设备可以基于CSI对待传输的无线信号进行多天线系统中的预编码，经过预编码后的无线信号可以对抗信道失真，提升信道容量。而上述测量信息、状态信息等通常较大，因此传输之前往往需要进行编码或压缩，从而节省空口资源和传输开销。
基站可以向终端发送用于信息编码的人工智能(artificial intelligence，AI)模型(简称AI编码器)，终端可以使用AI编码器对待传输的信息(例如上述测量信息、状态信息)进行编码，并将编码后的信息反馈给基站，基站通过和AI编码器对应的AI解码器解码获得终端上报的信息(例如上述测量信息、状态信息)。但数据驱动的AI编码器和AI解码器的存在泛化性问题，若待传输的信息和用于训练AI编码器和对应的AI解码器的数据不是同分布的，那么待传输的信息经过AI编码和AI解码之后可能会失真，进而影响通信系统性能。
发明内容
本申请实施例公开了一种信息编码的控制方法及相关装置，可以快速高效地选择合适的信息编码方案，避免待传输的信息经过编码和解码后出现失真，从而避免影响通信系统的性能，提高传输质量。
第一方面，本申请实施例提供了一种信息编码的控制方法，应用于终端，该方法包括：接收第一配置信息，第一配置信息用于配置N个AI编码器的N组参数，N为大于1的正整数；向网络设备发送第一指示信息，第一指示信息用于指示第一编码器用于第一信息编码，第一编码器是根据N组参数和第一信息确定的，第一编码器是N个AI编码器中的编码器，或者，第一编码器是和N个AI编码器不同的第二编码器。
其中，N个AI编码器的N组参数，可以替换为：N个AI解码器的N组参数，也可以替换为N个AI编解码器的N组参数。其中N个AI编码器分别对应N个AI解码器，一个AI编解码器包括一个AI编码器和对应的一个AI解码器。
可选地，第二编码器为基于码本等传统编码方案的编码器，或者其他泛化性更好的AI编码器。
本申请中，用于第一信息编码的第一编码器是基于N组参数确定的，第一编码器是N个AI编码器中的一个AI编码器或者N个AI编码器中以外的第二编码器，可以理解为是用于第一信息编码的第一编码器是经过判断得到的适用于第一信息编码的，从而避免通过第一信息不适用的AI编解码器(或者AI编码器和AI解码器)对第一信息进行编码和解码带来的信息失真，以及通信系统的性能恶化不可控的问题。
在一种可能的实现方式中，该方法还包括：发送第二信息，第二信息是基于第一编码器对第一信息的编码确定的。
在一种可能的实现方式中，第一编码器是根据第一判断参数和第一判断阈值的关系确定的，第一判断参数是根据N组参数和第一信息确定的，第一判断参数为余弦相似度CS、概率密度函数PDF、概率质量函数PMF或者欧式距离。
可选地，第一判断参数是根据N组参数、第一信息和第一预处理确定的，第一预处理包括以下至少一项：平移、缩放、傅里叶变换、压缩感知变换、截断、AI模型处理，以及对应的处理参数。
可选地，N组参数为N个AI编码器的训练数据集的统计信息，统计信息包括均值和/或数学分布的分布参数。
在一种可能的实现方式中，第一信息为信道状态信息CSI或上行数据。
可选地，第一信息为CSI，该方法还包括：接收CSI参考信号CSI-RS，第一信息是根据CSI-RS的测量结果确定的。
在一种可能的实现方式中，该方法还包括：向网络设备发送第二指示信息，第二指示信息用于指示第一编码器，或者用于指示N个AI编码器中的第一编码器适用于第一信息编码，或者用于指示N个AI编码器不适用于第一信息编码；接收第二配置信息，第二配置信息用于配置第一编码器，第二配置信息是根据第二指示信息确定的。
可选地，发送第二指示信息是在发送第二信息之前的。
本申请中，终端使用的第一编码器可以是向网络设备实时请求获得的，终端可以无需提前存储第一编码器，减小终端的存储压力。
在一种可能的实现方式中，该方法还包括：接收第三配置信息，第三配置信息用于配置N个AI编码器，和/或，N个AI编码器中的第一编码器，和/或，第二编码器。
可选地，接收第三配置信息是在发送第二信息之前的。
第二方面，本申请实施例提供了又一种信息编码的控制方法，应用于终端，该方法包括：接收第一配置信息，第一配置信息用于配置AI判决器，AI判决器用于确定N个AI编码器中第一信息适用的AI编码器，和/或用于确定N个AI编码器不适用于第一信息编码，N为大于1的正整数；向网络设备发送第一指示信息，第一指示信息用于指示第一编码器用于第一信息编码，第一编码器是根据AI判决器和第一信息确定的，第一编码器是N个AI编码器中的编码器，或者，第一编码器是和N个AI编码器不同的第二编码器。
其中，AI判决器用于确定N个AI编码器中第一信息适用的AI编码器，和/或用于确定N个AI编码器不适用于第一信息编码，可以替换为：AI判决器用于确定N个AI编解码器中第一信息适用的AI编解码器，和/或用于确定N个AI编解码器不适用于第一信息编解码。也可以替换为：AI判决器用于确定N个AI解码器中第一信息适用的AI解码器，和/或用于确定N个AI解码器不适用于第一信息解码。其中N个AI编码器分别对应N个AI解码器，一个AI编解码器包括一个AI编码器和对应的一个AI解码器。
可选地，第二编码器为基于码本等传统编码方案的编码器，或者其他泛化性更好的AI编码器。
本申请中，用于第一信息编码的第一编码器是基于AI判决器确定的，第一编码器是N个AI编码器中的一个AI编码器或者N个AI编码器中以外的第二编码器，可以理解为是用于第一信息编码的第一编码器是经过判断得到的适用于第一信息编码的，从而避免通过第一信息不适用的AI编解码器(或者AI编码器和AI解码器)对第一信息进行编码和解码带来的信息失真，以及通信系统的性能恶化不可控的问题。
在一种可能的实现方式中，该方法还包括：发送第二信息，第二信息是基于第一编码器对第一信息的编码确定的。
在一种可能的实现方式中，第一编码器是根据AI判决器的输出确定的，AI判决器的输出是根据第一信息和AI判决器得到的。
可选地，AI判决器的输出是根据第一输入信息和AI判决器得到的，第一输入信息是第一信息或第一信息经过第二预处理后得到的，第二预处理包括以下至少一项：平移、缩放、傅里叶变换、压缩感知变换、截断、AI模型处理，以及对应的处理参数。
可选地，AI判决器的输出是使用AI判决器对第一输入信息进行处理得到的。
可选地，AI判决器的输出是将第一输入信息作为AI判决器的输入得到的输出。
在一种可能的实现方式中，该方法还包括：第一信息为信道状态信息CSI或上行数据。
可选地，第一信息为CSI，该方法还包括：接收CSI参考信号CSI-RS，第一信息是根据CSI-RS的测量结果确定的。
在一种可能的实现方式中，该方法还包括：向网络设备发送第二指示信息，第二指示信息用于指示第一编码器，或者用于指示N个AI编码器中的第一编码器适用于第一信息编码，或者用于指示N个AI编码器不适用于第一信息编码；接收第二配置信息，第二配置信息用于配置第一编码器，第二配置信息是根据第二指示信息确定的。
可选地，发送第二指示信息是在发送第二信息之前的。
本申请中，终端使用的第一编码器可以是向网络设备实时请求获得的，终端可以无需提前存储第一编码器，减小终端的存储压力。
在一种可能的实现方式中，该方法还包括：接收第三配置信息，第三配置信息用于配置N个AI编码器，和/或，N个AI编码器中的第一编码器，和/或，第二编码器。
可选地，接收第三配置信息是在发送第二信息之前的。
第三方面，本申请实施例提供了又一种信息编码的控制方法，应用于终端，该方法包括：接收第一配置信息，第一配置信息用于配置第一AI编码器的参数；向网络设备发送第一指示信息，第一指示信息用于指示第一编码器用于第一信息编码，第一编码器是根据第一AI编码器的参数和第一信息确定的，第一编码器是第一AI编码器，或者，第一编码器是和第一AI编码器不同的第二编码器。
其中，第一AI编码器的N组参数，可以替换为：第一AI解码器的N组参数，也可以替换为第一AI编解码器的N组参数，其中第一AI编码器对应第一AI解码器，第一AI编解码器包括第一AI编码器和第一AI解码器。
可选地，第二编码器为基于码本等传统编码方案的编码器，或者其他泛化性更好的AI编码器。
本申请中，用于第一信息编码的第一编码器是基于第一AI编码器的参数确定的，第一编 码器是第一AI编码器或者第二编码器，可以理解为是用于第一信息编码的第一编码器是经过判断得到的适用于第一信息编码的，从而避免通过第一信息不适用的AI编解码器(或者AI编码器和AI解码器)对第一信息进行编码和解码带来的信息失真，以及通信系统的性能恶化不可控的问题。
在一种可能的实现方式中，该方法还包括：发送第二信息，第二信息是基于第一编码器对第一信息的编码确定的。
在一种可能的实现方式中，第一编码器是根据第一判断参数和第一判断阈值的关系确定的，第一判断参数是根据第一AI编码器的参数和第一信息确定的，第一判断参数为余弦相似度CS、概率密度函数PDF、概率质量函数PMF或者欧式距离。
可选地，第一判断参数是根据第一AI编码器的参数、第一信息和第一预处理确定的，第一预处理包括以下至少一项：平移、缩放、傅里叶变换、压缩感知变换、截断、AI模型处理，以及对应的处理参数。
可选地，第一AI编码器的参数为第一AI编码器的训练数据集的统计信息，统计信息包括均值和/或数学分布的分布参数。
在一种可能的实现方式中，第一信息为信道状态信息CSI或上行数据。
可选地，第一信息为CSI，该方法还包括：接收CSI参考信号CSI-RS，第一信息是根据CSI-RS的测量结果确定的。
在一种可能的实现方式中，该方法还包括：向网络设备发送第二指示信息，第二指示信息用于指示第一编码器，或者用于指示第一AI编码器适用于第一信息编码，或者用于指示第一AI编码器不适用于第一信息编码；接收第二配置信息，第二配置信息用于配置第一编码器，第二配置信息是根据第二指示信息确定的。
可选地，发送第二指示信息是在发送第二信息之前的。
本申请中，终端使用的第一编码器可以是向网络设备实时请求获得的，终端可以无需提前存储第一编码器，减小终端的存储压力。
在一种可能的实现方式中，该方法还包括：接收第三配置信息，第三配置信息用于配置第一AI编码器，和/或，第二编码器。
可选地，接收第三配置信息是在发送第二信息之前的。
第四方面，本申请实施例提供了又一种信息编码的控制方法，应用于终端，该方法包括：接收第一配置信息，第一配置信息用于配置AI判决器，AI判决器用于确定第一AI编码器适用于第一信息编码，和/或用于确定第一AI编码器不适用于第一信息编码；向网络设备发送第一指示信息，第一指示信息用于指示第一编码器用于第一信息编码，第一编码器是根据AI判决器和第一信息确定的，第一编码器是第一AI编码器，或者，第一编码器是和第一AI编码器不同的第二编码器。
其中，AI判决器用于确定第一AI编码器适用于第一信息编码，和/或用于确定第一AI编码器不适用于第一信息编码，可以替换为：AI判决器用于确定第一AI编解码器适用于第一信息编解码，和/或用于确定第一AI编解码器不适用于第一信息编解码。也可以替换为：AI判决器用于确定第一AI解码器适用于第一信息解码，和/或用于确定第一AI解码器不适用于第一信息解码。其中第一AI编码器对应第一AI解码器，第一AI编解码器包括第一AI编码器和第一AI解码器。
可选地，第二编码器为基于码本等传统编码方案的编码器，或者其他泛化性更好的AI 编码器。
本申请中，用于第一信息编码的第一编码器是基于AI判决器确定的，第一编码器是第一AI编码器或者第二编码器，可以理解为是用于第一信息编码的第一编码器是经过判断得到的适用于第一信息编码的，从而避免通过第一信息不适用的AI编解码器(或者AI编码器和AI解码器)对第一信息进行编码和解码带来的信息失真，以及通信系统的性能恶化不可控的问题。
在一种可能的实现方式中，第一编码器是根据AI判决器的输出确定的，AI判决器的输出是根据第一信息和AI判决器得到的。
可选地，AI判决器的输出是根据第一输入信息和AI判决器得到的，第一输入信息是第一信息或第一信息经过第二预处理后得到的，第二预处理包括以下至少一项：平移、缩放、傅里叶变换、压缩感知变换、截断、AI模型处理，以及对应的处理参数。
可选地，AI判决器的输出是使用AI判决器对第一输入信息进行处理得到的。
可选地，AI判决器的输出是将第一输入信息作为AI判决器的输入得到的输出。
在一种可能的实现方式中，该方法还包括：第一信息为信道状态信息CSI或上行数据。
可选地，第一信息为CSI，该方法还包括：接收CSI参考信号CSI-RS，第一信息是根据CSI-RS的测量结果确定的。
在一种可能的实现方式中，该方法还包括：向网络设备发送第二指示信息，第二指示信息用于指示第一编码器，或者用于指示第一AI编码器适用于第一信息编码，或者用于指示第一AI编码器不适用于第一信息编码；接收第二配置信息，第二配置信息用于配置第一编码器，第二配置信息是根据第二指示信息确定的。
可选地，发送第二指示信息是在发送第二信息之前的。
本申请中，终端使用的第一编码器可以是向网络设备实时请求获得的，终端可以无需提前存储第一编码器，减小终端的存储压力。
在一种可能的实现方式中，该方法还包括：接收第三配置信息，第三配置信息用于配置第一AI编码器，和/或，第二编码器。
可选地，接收第三配置信息是在发送第二信息之前的。
第五方面，本申请实施例提供了又一种信息编码的控制方法，应用于终端，该方法包括：接收第一配置信息，第一配置信息用于配置N个AI编码器的N组参数或AI判决器，AI判决器用于确定N个AI编码器中第一信息适用的AI编码器，和/或用于确定N个AI编码器不适用于第一信息编码；向网络设备发送第一请求信息，第一请求信息用于指示N个AI编码器中的第一AI编码器，第一请求信息是根据第一配置信息和第一信息确定的；向网络设备发送第一指示信息，第一指示信息用于指示第二编码器，第二编码器和N个AI编码器不同。
可选地，第二编码器为基于码本等传统编码方案的编码器，或者其他泛化性更好的AI编码器。
可选地，该方法还包括：接收网络设备发送的第二配置信息，第二配置信息用于配置第一AI编码器，第二配置信息是根据第一请求信息确定的。
本申请中，虽然存在第一信息适用的第一AI编码器，但是第一AI编码器未被发送至终端，此时终端可以使用第二编码器进行编码，从而减小传输时延，保证信息编码反馈的过程不中断。
在一种可能的实现方式中，该方法还包括：发送第二信息，第二信息是基于第一信息和 第二编码器确定的。
在一种可能的实现方式中，第一请求信息是根据第一判断参数和第一判断阈值的关系确定的，第一判断参数是根据N组参数和第一信息确定的，第一判断参数为余弦相似度CS、概率密度函数PDF、概率质量函数PMF或者欧式距离。
在一种可能的实现方式中，第一请求信息是根据AI判决器的输出确定的，AI判决器的输出是根据第一信息和AI判决器得到的。
在一种可能的实现方式中，该方法还可以包括：发送第四信息和第三指示信息，第三指示信息用于指示第一AI编码器，第四信息是根据第一AI编码器和第三信息确定的。
可选地，第三信息为信道状态信息CSI或上行数据。
本申请中，虽然终端使用的是第二编码器，但是终端可以向网络设备发送信息，以请求配置第一AI编码器，以便后续直接对适用于第一AI编码器的第四信息进行编码，无需再次请求，减小传输开销和传输时延。
在一种可能的实现方式中，第一信息为信道状态信息CSI或上行数据。
第六方面，本申请实施例提供了又一种信息编码的控制方法，应用于终端，该方法包括：接收第一配置信息，第一配置信息用于配置N个AI编码器的N组参数或AI判决器，AI判决器用于确定N个AI编码器中第一信息适用的AI编码器，和/或用于确定N个AI编码器不适用于第一信息编码；向网络设备发送第四指示信息，第四指示信息用于指示第一信息是否适用于N个AI编码器的第一判断结果和/或第一判断参数，第四指示信息是根据第一配置信息和第一信息确定的；接收第五指示信息，第五指示信息用于指示第一编码器用于第一信息编码，第五指示信息是根据第四指示信息确定的，第一编码器是N个AI编码器中的编码器，或者，第一编码器是和N个AI编码器不同的第二编码器。
可选地，第二编码器为基于码本等传统编码方案的编码器，或者其他泛化性更好的AI编码器。
本申请中，用于第一信息编码的第一编码器是基于第一配置信息确定的，第一编码器是N个AI编码器中的一个AI编码器或者N个AI编码器中以外的第二编码器，可以理解为是用于第一信息编码的第一编码器是经过判断得到的适用于第一信息编码的，从而避免通过第一信息不适用的AI编解码器(或者AI编码器和AI解码器)对第一信息进行编码和解码带来的信息失真，以及通信系统的性能恶化不可控的问题。
在一种可能的实现方式中，第一判断结果用于指示第一信息适用于N个AI编码器中的第一AI编码器，或者用于指示N个AI编码器不适用于第一信息。
在一种可能的实现方式中，第一判断参数是根据N组参数和第一信息确定的，第一判断参数为余弦相似度CS、概率密度函数PDF、概率质量函数PMF或者欧式距离。
在一种可能的实现方式中，第一判断参数是AI判决器的输出，AI判决器的输出是根据第一信息和AI判决器得到的。
在一种可能的实现方式中，第一信息为信道状态信息CSI或上行数据。
第七方面，本申请实施例提供了又一种信息编码的控制方法，应用于终端，该方法包括：接收第一配置信息，第一配置信息用于配置N个AI编码器的N组参数或AI判决器，AI判决器用于确定N个AI编码器中第一信息适用的AI编码器，和/或用于确定N个AI编码器不适用于第一信息编码；向网络设备发送第一信息，第一信息未经过编码或经过高保真编码， 第一信息用于作为第一AI编码器的训练数据集训练第一AI编码器，第一信息用于作为第一AI编码器的训练数据集是根据第一配置信息和第一信息确定的。
本申请中，终端可以根据第一配置信息对待传输的信息进行筛选，以将对训练AI编解码器价值高的数据发送给网络设备，从而避免终端发送对训练AI编解码器价值低的数据给网络设备，由于训练数据不均衡带来的AI编解码器性能较差的情况，也避免了上行带宽被浪费。
在一种可能的实现方式中，第一信息用于作为第一AI编码器的训练数据集，是根据第一判断参数和第一判断阈值的关系确定的，第一判断参数是根据N组参数和第一信息确定的，第一判断参数为余弦相似度CS、概率密度函数PDF、概率质量函数PMF或者欧式距离。
在一种可能的实现方式中，第一信息用于作为第一AI编码器的训练数据集，是根据AI判决器的输出确定的，AI判决器的输出是根据第一信息和AI判决器得到的。
在一种可能的实现方式中，第一信息为信道状态信息CSI或上行数据。
第八方面，本申请实施例提供了又一种信息编码的控制方法，应用于网络设备，该方法包括：向终端发送第一配置信息；第一配置信息用于配置N个AI编码器的N组参数或AI判决器，AI判决器用于确定N个AI编码器中第一信息适用的AI编码器，和/或用于确定N个AI编码器不适用于第一信息编码；接收第一指示信息和第二信息，第二信息是终端基于第一信息和第一编码器确定的，第一指示信息用于指示第一编码器用于第一信息编码，第一编码器是根据第一配置信息和第一信息确定的，第一编码器是N个AI编码器中的编码器，或者，第一编码器是和N个AI编码器不同的第二编码器。
在一种可能的实现方式中，该方法还包括：使用与第一编码器对应的第一解码器对第二信息进行解码。
在一种可能的实现方式中，第一编码器是根据第一判断参数和第一判断阈值的关系确定的，第一判断参数是根据N组参数和第一信息确定的，第一判断参数为余弦相似度CS、概率密度函数PDF、概率质量函数PMF或者欧式距离。
在一种可能的实现方式中，第一编码器是根据AI判决器的输出确定的，AI判决器的输出是根据第一信息和AI判决器得到的。
在一种可能的实现方式中，第一信息为信道状态信息CSI或上行数据。
第九方面，本申请实施例提供了又一种信息编码的控制方法，应用于网络设备，该方法包括：向终端发送第一配置信息第一配置信息用于配置N个AI编码器的N组参数或AI判决器，AI判决器用于确定N个AI编码器中第一信息适用的AI编码器，和/或用于确定N个AI编码器不适用于第一信息编码；接收第一请求信息，第一请求信息用于指示N个AI编码器中的第一AI编码器，第一请求信息是根据第一配置信息和第一信息确定的；接收第一指示信息和第二信息，第一指示信息用于指示第二编码器用于第一信息编码，第二编码器和N个AI编码器不同，第二信息是基于第一信息和第二编码器确定的。
可选地，第二编码器为基于码本等传统编码方案的编码器，或者其他泛化性更好的AI编码器。
可选地，该方法还包括：发送第二配置信息，第二配置信息用于配置第一AI编码器，第二配置信息是根据第一请求信息确定的。
在一种可能的实现方式中，该方法还包括：使用与第二编码器对应的第二解码器对第二信息进行解码。
在一种可能的实现方式中，第一请求信息是根据第一判断参数和第一判断阈值的关系确定的，第一判断参数是根据N组参数和第一信息确定的，第一判断参数为余弦相似度CS、概率密度函数PDF、概率质量函数PMF或者欧式距离。
在一种可能的实现方式中，第一请求信息是根据AI判决器的输出确定的，AI判决器的输出是根据第一信息和AI判决器得到的。
在一种可能的实现方式中，该方法还可以包括：接收第四信息和第三指示信息，第三指示信息用于指示第一AI编码器，第四信息是根据第一AI编码器和第三信息确定的。
可选地，第四信息为信道状态信息CSI或上行数据。
在一种可能的实现方式中，第一信息为信道状态信息CSI或上行数据。
第十方面，本申请实施例提供了又一种信息编码的控制方法，应用于网络设备，该方法包括：向终端发送第一配置信息，第一配置信息用于配置N个AI编码器的N组参数或AI判决器，AI判决器用于确定N个AI编码器中第一信息适用的AI编码器，和/或用于确定N个AI编码器不适用于第一信息编码；接收第四指示信息，第四指示信息用于指示第一信息是否适用于N个AI编码器的第一判断结果和/或第一判断参数，第四指示信息是根据第一配置信息和第一信息确定的；发送第五指示信息，第五指示信息用于指示第一编码器用于第一信息编码，第五指示信息是根据第四指示信息确定的，第一编码器是N个AI编码器中的编码器，或者，第一编码器是和N个AI编码器不同的第二编码器。
可选地，第二编码器为基于码本等传统编码方案的编码器，或者其他泛化性更好的AI编码器。
在一种可能的实现方式中，第一判断结果用于指示第一信息适用于N个AI编码器中的第一AI编码器，或者用于指示N个AI编码器不适用于第一信息。
在一种可能的实现方式中，第一判断参数是根据N组参数和第一信息确定的，第一判断参数为余弦相似度CS、概率密度函数PDF、概率质量函数PMF或者欧式距离。
在一种可能的实现方式中，第一判断参数是AI判决器的输出，AI判决器的输出是根据第一信息和AI判决器得到的。
在一种可能的实现方式中，第一信息为信道状态信息CSI或上行数据。
第十一方面，本申请实施例提供了又一种信息编码的控制方法，应用于网络设备，该方法包括：向终端发送第一配置信息，第一配置信息用于配置N个AI编码器的N组参数或AI判决器，AI判决器用于确定N个AI编码器中第一信息适用的AI编码器，和/或用于确定N个AI编码器不适用于第一信息编码；接收第一信息，第一信息未经过编码或经过高保真编码，第一信息用于作为第一AI编码器的训练数据集训练第一AI编码器，第一信息用于作为第一AI编码器的训练数据集是根据第一配置信息和第一信息确定的。
在一种可能的实现方式中，第一信息用于作为第一AI编码器的训练数据集，是根据第一判断参数和第一判断阈值的关系确定的，第一判断参数是根据N组参数和第一信息确定的，第一判断参数为余弦相似度CS、概率密度函数PDF、概率质量函数PMF或者欧式距离。
在一种可能的实现方式中，第一信息用于作为第一AI编码器的训练数据集，是根据AI判决器的输出确定的，AI判决器的输出是根据第一信息和AI判决器得到的。
在一种可能的实现方式中，第一信息为信道状态信息CSI或上行数据。
第十二方面，本申请实施例提供了一种终端，包括收发器、处理器和存储器；上述存储 器用于存储计算机程序代码，上述计算机程序代码包括计算机指令，上述处理器调用上述计算机指令以使上述用户设备执行本申请实施例第一方面至第七方面，以及第一方面至第七方面的任意一种实现方式提供的信息编码的控制方法。
第十三方面，本申请实施例提供了一种网络设备，包括收发器、处理器和存储器；上述存储器用于存储计算机程序代码，上述计算机程序代码包括计算机指令，上述处理器调用上述计算机指令以使上述用户设备执行本申请实施例第八方面和第十一方面，以及第八方面和第十一方面的任意一种实现方式提供的信息编码的控制方法。
第十四方面，本申请实施例提供了又一种终端，用于执行本申请任一实施例终端所执行的方法。
第十五方面，本申请实施例提供了一种网络设备，用于执行本申请任一实施例网络设备所执行的方法。
第十六方面，本申请实施例提供了一种计算机存储介质，该计算机存储介质存储有计算机程序，该计算机程序被电子设备执行时，用于执行本申请实施例第一方面至第十一方面，以及第一方面至第十一方面的任意一种实现方式提供的信息编码的控制方法。
第十七方面，本申请实施例提供了一种计算机程序产品，当该计算机程序产品在电子设备上运行时，使得该电子设备执行本申请实施例第一方面至第十一方面，以及第一方面至第十一方面的任意一种实现方式提供的信息编码的控制方法。
第十八方面，本申请实施例提供一种电子设备，该电子设备包括执行本申请任一实施例所介绍的方法或装置。上述电子设备例如为芯片。
附图说明
以下对本申请实施例用到的附图进行介绍。
图1是本申请实施例提供的一种通信系统的架构示意图；
图2是本申请实施例提供的一种信息编解码的场景示意图；
图3是本申请实施例提供的一种终端100的结构示意图；
图4是本申请实施例提供的一种网络设备200的结构示意图；
图5-图10是本申请实施例提供的一些信息编码的控制方法的流程示意图；
图11是本申请实施例提供的一种筛选方法的流程示意图；
图12-图13是本申请实施例提供的一些人工智能AI判决器的生成过程的示意图；
图14是本申请实施例提供的一种AI编解码器的生成过程的示意图。
具体实施方式
下面将结合附图对本申请实施例中的技术方案进行清楚、详尽地描述。本申请实施例的实施方式部分使用的术语仅用于对本申请的具体实施例进行解释，而非旨在限定本申请。
本申请实施例提供了一种信息编码的控制方法，应用于包括网络设备和终端的通信系统。网络设备可以向终端发送第一配置信息，终端可以根据第一配置信息判断待传输的信息是否适用于用于信息编码的人工智能(artificial intelligence，AI)模型(也可以理解为是基于AI技术的编码器，简称AI编码器)。若待传输的信息适用于AI编码器，则终端可以使用该AI编码器对待传输的信息进行编码，若待传输的信息不适用AI编码器，终端可以使用备用编码器对待传输的信息进行编码。因此可以快速高效地选择合适的信息编码方案，避免待传输的 信息经过编码和解码后出现失真，从而避免影响通信系统的性能，提高传输质量。
需要说明的是，一个AI编码器可以对应一个用于信息解码的AI模型(也可以理解为是基于AI技术的解码器，简称AI解码器)，一个AI编码器和对应的AI解码器可简称为一组AI编解码器(也可以理解为是基于AI技术的编解码器)，不同组AI编解码器的结构不同或结构相同但模型参数(系数)不同。不同组AI编解码器所对应的标识不同。不同组AI编解码器中AI编码器可以相同但AI解码器不同，或者不同组AI编解码器中AI解码器可以相同但AI编码器可以不同。通过AI编码器编码后的信息需通过对应同组的AI解码器解码，其他解码器无法解码(或解码后的信息与原始待传输的信息偏差较大)。AI编码器通常是和对应的AI解码器一起作为一种自编码器(autoencoder)训练的，即在训练时将AI编码器的输入和AI解码器的输出都设置为待编码的信息(可以理解为是该AI编码器、该AI解码器或这组AI编解码器的训练数据集)，通过误差反向传播获得AI模型(包括AI编码器和/或AI解码器)中的参数(系数)，以实现在推理时，待编码的信息经过AI编码和AI解码之后得到的信息与待编码的信息尽可能一致。
可选地，待编码的信息适用于AI编码器即表征也适用于该AI编码器对应的AI解码器，也可以理解为是待编码的信息适用于对应的AI编解码器(即包括上述AI编码器和对应的AI解码器)。因此，终端根据第一配置信息判断待传输的信息是否适用于AI编码器，可以理解为是：终端根据第一配置信息判断待传输的信息是否适用于AI编码器和对应的AI解码器(即AI编解码器)。
相应地，一种备用编码器也可以对应一种解码器(简称备用解码器)，使用备用编码器编码后的信息也需要通过对应的备用解码器解码，其他解码器无法解码。备用解码器的解码方案取决于对应的备用编码器的编码方案，例如备用编码器是使用码本等传统编码方案的编码器，则对应的备用解码器是使用码本等传统解码方案的解码器。或者，备用编码器是其他泛化性更好的AI编码器，则对应的备用解码器是其他泛化性更好的AI解码器。
可选地，终端可以将编码后的信息反馈给网络设备，并向网络设备发送指示终端使用的编码器的通知，网络设备可以根据该通知确定对应的解码器，并使用该解码器解码获得终端上报的信息。其中，终端使用的编码器为AI编码器时，网络设备使用的解码器为和该AI编码器对应的AI解码器。终端使用的编码器为备用编码器时，网络设备使用的解码器为和该备用编码器对应的备用解码器。
本申请中，待传输的信息即为终端向网络设备发送的信息，可以是信道状态信息(channel state information，CSI)等测量信息、状态信息。不限于此，也可以是业务数据，例如音频数据、视频数据、文本数据等等。
可选地，终端使用的AI编码器或备用编码器可以是网络设备发送给终端的。可选地，终端使用的AI编码器或备用编码器可以是终端和网络设备预先协商配置的。
本申请中，通信系统可以是无线通信系统，例如但不限于全球移动通讯系统(global system for mobile communications，GSM)、码分多址接入(code division multiple access，CDMA)、宽带码分多址(wideband code division multiple access，WCDMA)、时分同步码分多址(time division synchronous code division multiple ac，TD-SCDMA)、长期演进(long term evolution，LTE)、新无线接入(new radio，NR)或者其他未来网络系统。
本申请中，网络设备可以是一种用于发送或接收信息的设备，可选地，网络设备为接入网设备，可选地，网络设备为核心网设备。例如但不限于：基站，用户设备(user equipment，UE)，无线接入点(access point，AP)，收发点(transmission and receiver point，TRP)，中继 设备，或者具备基站功能的其他网络设备等。其中，基站是一种部署在无线接入网(radio access network，RAN)中用于提供无线通信功能的设备。在不同的无线接入系统中，基站的名称可能不同。例如但不限于，GSM或CDMA中的基站收发台(base transceiver station，BTS)，WCDMA中的节点B(node B，NB)，LTE中的演进型基站(evolved node B，eNodeB)，还可以是NR中的下一代基站(g node B，gNB)，或者其他未来网络系统中的基站。示例性地，网络设备可以是下图1所示的基站110，或者核心网120。
本申请中，终端可以是具有无线通信功能的设备，可选地，终端为UE。在某些场景下，终端也可以被称为移动台、接入终端、用户代理等。例如，终端为手持设备、可穿戴设备、计算设备、便携式设备或车载设备等形式的终端。例如，终端具体为蜂窝电话、智能手机、智能眼镜、膝上型电脑、个人数字助理或无绳电话等设备。示例性地，终端可以是下图1所示的终端100。
请参见图1，图1是本申请实施例提供的一种通信系统的架构示意图。
如图1所示，该通信系统可以包括终端100、基站110和核心网120。其中，核心网120可以连接至少一个基站110，基站110可以为至少一个终端100提供无线通信服务，终端100可以通过空中接口连接至少一个基站110。核心网120为该通信系统中的关键控制节点，主要负责信令处理功能，例如但不限于用于实现接入控制、移动性管理、会话管理等功能。至少一个基站110可以构成一个RAN节点。在NR中，核心网120可以称为5G核心网(5G Core，5GC)120，基站110可以称为gNB110。NR-RAN节点可以包括至少一个通过NG接口连接至5GC120的gNB110，并且，NR-RAN节点中至少一个gNB110可以通过Xn-C接口连接和通信。终端100可以通过Uu接口连接gNB110。
核心网120可以通过基站110向终端100发送下行信息，终端100也可以通过连接的基站110向核心网120发送上行信息。其中，终端100处于基站110覆盖的范围内时，需通过随机接入等操作后才能和基站110连接，连接后终端100才可以和基站110进行信息交互，以及通过基站110和核心网120进行信息交互。
需要说明的是图1所示的终端100、基站110和核心网120的形态和数量仅用于示例，本申请实施例对此不作限定。
请参见图2，图2示例性示出一种信息编解码的场景示意图。图2所示的场景下包括终端100和网络设备200，其中终端100和网络设备200可以连接并通信。
如图2的(A)所示，网络设备200可以部署有AI编码器和对应的AI解码器。网络设备200可以向终端100发送AI编码器，以便后续终端100使用AI编码器对待传输的信息(即信息A)进行信息编码。如图2的(B)所示，终端100可以判断信息A是否适用于网络设备200发送的AI编码器，当适用时，终端100可以使用该AI编码器对信息A进行编码，并将编码后的信息发送给网络设备200。然后，网络设备200可以使用AI解码器对编码后的信息进行解码，以得到恢复后的信息(即信息A′)。
接下来示例性介绍本申请实施例中提供的终端和网络设备。
请参见图3，图3示出了一种终端100的结构示意图。终端100可以是图1所示的终端100，也可以是图2所示的终端100。终端100可以包括处理器110、存储器120和收发器130，处理器110、存储器120和收发器130通过总线相互连接。
处理器110可以是一个或多个中央处理器(central processing unit，CPU)，在处理器110是一个CPU的情况下，该CPU可以是单核CPU，也可以是多核CPU。在一些实施例中，处理器110可以包括多个处理单元，例如应用处理器(application processor，AP)、调制解调处理 器(modem)等。其中，不同的处理单元可以是独立的器件，也可以集成在一个或多个处理器中。存储器120可以包括但不限于是随机存储记忆体(random access memory，RAM)、只读存储器(read-only memory，ROM)、可擦除可编程只读存储器(erasable programmable read only memory，EPROM)、或便携式只读存储器(compact disc read-only memory，CD-ROM)。存储器120用于存储相关计算机程序及信息，可选地，存储器120用于存储网络设备发送的AI编码器，可选地，存储器120用于存储网络设备发送的用于判断是否存在待传输的信息适用的AI编码器的AI模型(简称AI判决器)。收发器130用于接收和发送信息。在一些实施例中，收发器130可以包括无线收发器和移动收发器。
在一些实施例中，终端100可以通过调制解调处理器和移动收发器实现GSM、CDMA、WCDMA、SCDMA、UMTS、LTE、NR等移动通信技术。终端100可以通过调制解调处理器和移动收发器和网络设备通信，例如传输CSI等测量信息、状态信息，AI编码器等。
终端100中的处理器110用于读取存储器120中存储的计算机程序代码，执行图5-图10所示的信息编码的控制方法、图11所示的筛选方法中终端执行的步骤。
请参见图4，图4示出了一种网络设备200的结构示意图。网络设备200可以是图1所示的基站110或核心网120，也可以是图2所示的网络设备200。网络设备200可以包括处理器210、存储器220和收发器230，处理器210、存储器220和收发器230通过总线相互连接。
处理器210可以是一个或多个CPU，在处理器210是一个CPU的情况下，该CPU可以是单核CPU，也可以是多核CPU。在一些实施例中，处理器210可以包括多个处理单元，其中，不同的处理单元可以是独立的器件，也可以集成在一个或多个处理器中。可选地，处理器210可以训练生成AI编解码器。可选地，处理器210可以训练生成AI判决器。存储器220可以包括但不限于是RAM、ROM、EPROM、或CD-ROM，存储器220用于存储相关计算机程序及信息，可选地，存储器220用于存储获取的AI编解码器，可选地，存储器220用于存储获取的AI判决器。收发器230用于接收和发送信息。
在一些实施例中，网络设备200可以通过处理器210和收发器230实现GSM、CDMA、WCDMA、SCDMA、UMTS、LTE、NR等移动通信技术。网络设备200可以通过处理器210和收发器230和终端通信，例如传输CSI等测量信息、状态信息，AI编码器等。
网络设备200中的处理器210用于读取存储器220中存储的计算机程序代码，执行图5-图10所示的信息编码的控制方法、图11所示的筛选方法中网络设备执行的步骤。
接下来介绍本申请实施例提供的信息编码的控制方法。该方法可以应用于图1所示的通信系统，也可以应用于图2所示的场景。该方法中的终端可以是图3所示的终端100。该方法中的网络设备可以是图4所示的网络设备200。
请参见图5，图5是本申请实施例提供的一种信息编码的控制方法的流程示意图。该方法可以包括但不限于如下步骤：
S101：网络设备向终端发送第一配置信息。
在一些实施例中，S101之前，该方法可以包括：网络设备和终端连接，连接方式可参见上图1-图4的说明。示例性地，假设网络设备为基站，终端进入基站覆盖的范围内向基站发起随机接入，随机接入成功后终端和基站连接，可以进行通信。
在一些实施例中，S101之前，该方法可以包括：网络设备和终端协商确定启用基于AI编码器的信息编码反馈机制。后续网络设备可以向终端发送AI编码器，终端可以在协商的时 刻使用AI编码器进行编码。协商的时刻例如但不限于为终端根据第一配置信息确定存在待传输的信息适用的AI编码器的时刻(例如S102之后)，或者接收到网络设备发送的指示使用的编码器的信息的时刻(例如接收到下图10所示的第六指示信息之后)。
具体地，S101为可选的步骤。可选地，第一配置信息可以包括N组AI编解码器的信息，也可以理解为是N个AI编码器的信息，也可以理解为是N个AI解码器的信息，其中这N个AI编码器和这N个AI解码器分别对应，N组AI编解码器可以分别包括这N个AI编码器和这N个AI解码器。可选地，第一配置信息包括的内容可以有三种情况，如下所示：
情况一，第一配置信息可以包括网络设备获取的N组AI编解码器(也可理解为N个AI编码器)的训练数据集的统计信息，其中，N为正整数。可选地，第一配置信息可以包括N组第二配置信息，这N组第二配置信息分别对应上述N组AI编解码器，每组第二配置信息可以包括对应的AI编解码器的训练数据集的统计信息。统计信息可以包括但不限于均值和/或数学分布的分布参数，其中，任意一个数学分布的分布参数可以根据该数学分布的特性决定。例如，一组AI编解码器的统计信息包括两个正态分布的均值和方差，这两个正态分布的均值和方差均不相同。或者，一组AI编解码器的统计信息包括一个正态分布的均值和方差，以及一个泊松分布的均值和λ。
在一些实施例中，统计信息可以是多个不同维度的。例如，假设训练数据集的维度为[L,K 1,K 2,…,K z]，其中，L,K 1,K 2,…,K z均为正整数，K 1,K 2,…,K z为训练数据集的特征维度。L为训练数据集的统计维度。那么训练数据集的均值、数学分布的分布参数可以分别为沿着K 1,K 2,…,K z的维度，或者也可以是K 1×K 2×…×K z的合并维度，不限于此，还可以是其他排列组合维度，本申请对此不作限定。
情况二和情况一类似，不同之处在于：第一配置信息中的统计信息是经过第一预处理的，第一预处理可以包括以下至少一项：平移、缩放、傅里叶变换、压缩感知变换(即乘以压缩感知的测量矩阵)、截断、AI模型处理等。可选地，第一预处理还可以包括对应的处理参数，例如平移的方向和数值。
其中，AI模型处理可以是通过AI模型或AI模型的浅层部分处理，例如进行特征提取，实现降维。需要说明的是，这里的AI模型和AI编码器的AI模型、AI解码器的AI模型可以不同。示例性地，假设待处理的信息是维度为R×W的CSI矩阵H，该CSI矩阵H可以经过某个AI模型的全连接线性层处理(即进行压缩感知编码)，具体包括乘以一个维度为E×R的测量矩阵U，其中E可以远小于R。处理后的矩阵O＝U×H，矩阵O的维度为E×W。假设对应压缩感知编码的解码算法为Learned AMP(LAMP)，则解码后的CSI矩阵H′＝LAMP(U×H)。矩阵U可以是作为一层神经网络训练得到的，即基于上述获取解码后的CSI矩阵的公式，使用已知的CSI矩阵H训练得到测量矩阵U。可选地，压缩感知变换中的测量矩阵可以是随机生成的，例如通过高斯分布随机生成的。
不限于此，还可以有其他类型的预处理，本申请对预处理的具体内容不作限定。
在一些实施例中，第一配置信息还可以包括第一预处理的信息，以便终端可以根据第一配置信息确定第一预处理的内容。可选地，具体可以是每组第二配置信息包括对应的第一预处理的信息。在另一些实施例中，S101之前，该方法还可以包括：终端接收网络设备发送的第一预处理的信息。在另一些实施例中，S101之前，该方法还可以包括：网络设备和终端协商确定第一预处理的内容，即第一预处理是预先定义的。不同AI编码器的训练数据集的统计信息所经过的第一预处理的内容可以不同，例如不同组第二配置信息指示的第一预处理可以不同。
可选地，终端在判断待传输的信息是否适用于待应用的AI编码器(即上述N组AI编解码器中任意一个AI编码器)时，需先对待传输的信息进行第一预处理，然后再基于第一预处理后的信息进行判断过程，具体如S102所示。相应地，终端在使用AI编码器(对应这里的 第一预处理)对待传输的信息进行编码时，需先对待传输的信息进行第一预处理，然后再使用该AI编码器对第一预处理后的信息进行编码，具体如S103所示。
示例性地，假设第一配置信息指示的第一预处理包括傅里叶变换，截断和压缩感知变换，则终端在进行判断过程，以及使用AI编码器进行编码时，需先对待传输的信息进行傅里叶变换、截断丢弃、以及乘以压缩感知的测量矩阵。
在上述情况一和上述情况二下，在一些实施例中，第一配置信息还可以包括统计信息的类型的指示。可选地，具体可以是每组第二配置信息包括统计信息的类型的指示，不同组第二配置信息包括的统计信息的类型可以不同。在另一些实施例中，S101之前，该方法还可以包括：网络设备和终端协商确定统计信息的类型，即统计信息的类型是预先定义的。
在上述情况一和上述情况二下，在一些实施例中，第一配置信息还可以包括一组判断阈值，可选地，还可以包括：根据这组判断阈值判断待传输的信息是否适用于待应用的AI编码器的判断方法。在另一些实施例中，S101之前，该方法还可以包括：网络设备和终端协商确定一组判断阈值，可选地以及上述判断方法，即判断阈值是预先定义的，可选地，判断方法是预先定义的。其中，上述一组判断阈值可以包括至少一个判断阈值。在另一些实施例中，S101之前，该方法还可以包括：终端接收网络设备发送的判断阈值和/或判断方法。可选地，对于不同终端，网络设备确定的判断阈值可以不同。
情况三，第一配置信息可以包括AI判决器，即用于判断待传输的信息是否适用于待应用的AI编码器的AI模型。可选地，AI判决器的输入为终端待传输的信息。可选地，AI判决器的输入为待传输的信息经过第二预处理后得到的。AI判决器的输出端口有至少一个，例如为N个，分别对应上述N组AI编解码器(也可理解为是对应N个AI编码器)。
示例性地，当N等于1时，AI判决器的一个输出值表征待传输的信息是否适用于这一个AI编码器。
示例性地，当N大于1时，AI判决器的N个输出值可以表征待传输的信息适用于这N个AI编码器的情况，AI判决器的任意一个输出端口的输出值可以表征待传输的信息适用于该输出端口对应的AI编码器的概率值，概率值越高表示待传输的信息越适用于该AI编码器，即应用该输出端口对应的AI编解码器对待传输的信息进行信息编解码的效果预期越好。
在一些实施例中，第一配置信息还可以包括AI判决器的输出的指示信息。该指示信息可以包括AI判决器的输出表征的内容，例如表征待传输的信息是否适用于待应用的AI编码器，或者表征待传输的信息适用于待应用的AI编码器的概率值。可选地，该指示信息还可以包括AI判决器的输出的阈值(简称输出阈值)，以及根据输出阈值和AI判决器的输出值判断待传输的信息是否适用于待应用的AI编码器的判断方法。在另一些实施例中，S101之前，该方法还可以包括：网络设备和终端协商确定AI判决器的输出的指示信息，即指示信息是预先定义的。在另一些实施例中，S101之前，该方法还可以包括：终端接收网络设备发送的AI判决器的输出的指示信息。
示例性地，当N等于1时，AI判决器的输出值分别为0或1，其中，输出为1时表征待传输的信息适用于这一个AI编码器，输出为0时表征待传输的信息不适用于这一个AI编码器。或者，AI判决器输出的取值范围为[0,1]，该输出值越高表示第一信息越适用于该AI编码器，即应用该输出端口对应的AI编解码器对待传输的信息进行信息编解码的效果预期越好。当该输出值大于输出阈值时，终端可以确定待传输的信息适用于该AI编码器，否则待传输的信息不适用于该AI编码器。
示例性地，当N大于1时，AI判决器的N个输出端口分别输出I 1、I 2、…、I N，其中， I 1+I 2+…+I N＝1。当I i>I t时表征待传输的信息适用于第i个输出端口对应的AI编码器，其中，i的取值范围为[1,N]，I t为AI判决器的输出阈值。当I i>I t以及I i大于其他N-1个输出端口的输出时，表征待传输的信息最适用于第i个输出端口对应的AI编码器，即应用第i个输出端口对应的AI编解码器对待传输的信息进行信息编解码的效果预期最好。
其他AI判决器的输出的示例可参见S102中AI判决器的输出的说明，暂不详述。
在一些实施例中，第一配置信息还可以包括第二预处理的信息，以便终端可以根据第一配置信息确定第二预处理的内容。在另一些实施例中，S101之前，该方法还可以包括：网络设备和终端协商确定第二预处理的内容，即第二预处理是预先定义的。在另一些实施例中，S101之前，该方法还可以包括：终端接收网络设备发送的第二预处理的信息。第一预处理和第二预处理可以相同，也可以不同，第二预处理的内容示例可参见上述第一预处理的示例。
在一些实施例中，在情况三下，第一配置信息也可以包括第一预处理的信息或预先定义了第一预处理，以便终端使用AI编码器对待传输的信息进行编码时，先对待传输的信息进行第一预处理，然后使用AI编码器对第一预处理后的信息进行编码，具体如S103所示。并且，若第一配置信息包括第二预处理的信息或预先定义了第二预处理，则终端判断待传输的信息是否适用于待应用的AI编码器时，需先对待传输的信息进行第二预处理，然后再基于第二预处理后的信息进行判断过程，具体如S102所示。
在上述三种情况下，在一些实施例中，第一配置信息还可以包括N个AI编码器的性能要求，例如但不限于包括存储空间的容量要求，计算能力的要求，时延的要求等。可选地，具体可以是每组第二配置信息包括对应的AI编码器的性能要求。在另一些实施例中，S101之前，该方法还可以包括：网络设备向终端发送N个AI编码器的性能要求。在另一些实施例中，S101之前，该方法还可以包括：网络设备和终端协商确定N个AI编码器的性能要求，即性能要求是预先定义的。
在上述三种情况下，在一些实施例中，第一配置信息还可以包括备用编码方案(即通过备用编码器进行编码的方案)，可选地，以及备用编码方案的使用方式。在另一些实施例中，S101之前，该方法还可以包括：网络设备和终端协商确定备用编码方案，可选地，以及备用编码方案的使用方式，即备用编码方案是预先定义的，可选地，使用方式是预先定义的。在另一些实施例中，S101之前，该方法还可以包括：终端接收网络设备发送的备用编码方案，可选地，以及使用方式。其中，备用编码器可以但不限于是使用码本等传统编码方案的编码器，或其他泛化性更好的AI编码器。
示例性地，使用方式可以包括：当终端确定不存在待传输的信息适用的AI编码器时，可以使用备用编码器对待传输的信息进行编码。和/或，虽然终端确定存在待传输的信息适用的AI编码器，但终端还未接收到该AI编码器，则使用备用编码器对待传输的信息进行编码。和/或，虽然终端确定存在待传输的信息适用的AI编码器，但终端性能不满足该AI编码器的性能要求，则使用备用编码器对待传输的信息进行编码。
S102：终端根据第一配置信息判断是否存在第一信息适用的AI编码器。
具体地，S102为可选的步骤。可选地，终端根据第一配置信息判断待传输的第一信息是否适用于待应用的AI编码器(即上述N组AI编解码器中任意一个AI编码器)。第一信息即为待传输的信息，可以但不限于是CSI等测量信息、状态信息，以及音频数据、文本数据等业务数据。可选地，当N等于1时，终端可以根据第一配置信息判断第一信息是否适用于这一个AI编码器，可选地，当N大于1时，终端可以根据第一配置信息判断上述N组AI编解码器(也可理解为N个AI编码器)中是否存在第一信息适用的AI编码器，可选地，以及第 一信息适用的AI编码器。
其中，上述N个AI编码器可以由网络设备发送给终端，可选地，可以是在S102之前网络设备向终端发送N个AI编码器，可选地，可以是S102(终端确定待传输的信息适用的AI编码器)之后，终端向网络设备发送信息，以请求配置该AI编码器，网络设备响应于终端的请求向终端发送该AI编码器。
在一些实施例中，终端可以根据第一配置信息判断是否满足N个AI编码器的性能要求，若均不满足，终端可以确定使用的编码器为备用编码器；若终端确定满足M个AI编码器的性能要求，则可以根据第一配置信息判断这M个AI编码器中是否存在第一信息适用的AI编码器，M小于或等于N。示例性地，若这M个AI编码器中存在第一信息适用的AI编码器，终端可以确定使用的编码器为该AI编码器，若不存在，终端可以确定使用的编码器为备用编码器。在另一些实施例中，终端也可以先判断得到第一信息适用的AI编码器，然后再根据第一配置信息判断是否满足该AI编码器的性能要求。示例性地，当不满足时，终端可以确定使用的编码器为备用编码器；当满足时，终端可以确定使用的编码器为该AI编码器。
其中，终端满足AI编码器的性能要求可以包括以下至少一项：终端满足AI编码器的存储能力要求，终端满足AI编码器的计算能力要求，终端满足AI编码器的时延要求。示例性地，当终端本地可用的存储空间大于AI编码器对存储空间的容量要求，终端可以确定满足AI编码器的存储能力要求。当终端表征计算能力的参数(例如单位时间内的计算量等)大于AI编码器对计算能力的要求，终端可以确定满足AI编码器的计算能力要求。当终端基于AI编码器完成信息编码的时延小于该AI编码器的时延要求，终端可以确定满足AI编码器的时延要求。
需要说明的是，满足AI编码器的性能要求仅为使用AI编码器的前置条件，即若不满足AI编码器的性能要求则无法使用AI编码器，若满足AI编码器的性能要求则要根据实际情况(例如终端是否接收到AI编码器、网络设备的指示)等确定是否使用AI编码器。
后续实施例以终端满足N个AI编码器的性能要求为例进行说明。
对于不同情况的第一配置信息，终端判断是否存在第一信息适用的AI编码器的方式可以不同，具体可以包括以下三种情况：
情况一，当第一配置信息为S101的情况一所示，以及包括训练数据集的均值时，终端可以计算第一信息和该均值的余弦相似度(cosine similarity，CS)。当第一配置信息为S101的情况二所示，以及包括经过第一预处理的训练数据集的均值，则终端可以先对第一信息进行第一预处理，然后计算第一信息经过第一预处理后得到的信息和该均值的余弦相似度。最后，终端可以根据计算得到的余弦相似度和对应的判断阈值判断第一信息是否适用于待应用的AI编码器。
可选地，当N等于1时，终端可以计算这一个AI编码器对应的余弦相似度，并根据该余弦相似度和判断阈值的关系判断第一信息是否适用于该AI编码器。例如，预先定义或第一配置信息中指示的一组判断阈值包括一个判断阈值CS t，计算得到的余弦相似度CS 0>CS t时表征第一信息适用于该AI编码器(即适用于这组AI编解码器)，否则表征第一信息不适用于该AI编码器。或者，该组判断阈值包括两个判断阈值CS t1和CS t2，计算得到的余弦相似度CS 0<CS t1时表征第一信息不适用于该AI编码器，CS t2≥CS 0≥CS t1时表征第一信息适用于该AI编码器，但应用该组AI编解码器对第一信息进行编解码的效果预期较差，CS 0>CS t2时表征第一信息适用于该AI编码器，以及应用该组AI编解码器对第一信息进行编解码的效果预期较好。预先定义或第一配置信息中指示的判断方法可以包括：CS 0>CS t2时才确定应用的编码器为该AI编码器，或者，CS 0≥CS t1时就确定应用的编码器为该AI编码器。
可选地，当N大于1时，终端可以分别计算N个AI编码器对应的余弦相似度，并根据这N个余弦相似度和判断阈值的关系判断是否存在第一信息适用的AI编码器，可选地，以及第一信息适用的AI编码器。余弦相似度越大则表示第一信息越适用于对应的AI编码器，即应用该组AI编解码器对第一信息进行编解码的效果预期越好。当余弦相似度和判断阈值满足预设阈值条件时，终端可以确定第一信息适用于该余弦相似度对应的AI编码器，预设阈值条件的示例可参见上述N等于1的判断阈值和判断方法。可选地，终端可以将余弦相似度和判断阈值满足预设阈值条件，且余弦相似度最大的AI编码器确定为第一信息最适用的AI编码器，即应用该组AI编解码器对第一信息进行编解码的效果预期最好。
情况二，当第一配置信息为S101的情况一所示，以及包括数学分布的完整分布参数(例如包括可以共同构成多维高斯分布的均值和方差矩阵)时，终端可以计算第一信息在该数学分布中的概率参数，例如概率密度函数(probability density function，PDF)或概率质量函数(probability mass function，PMF)。当第一配置信息为S101的情况二所示，以及包括经过第一预处理的数学分布的完整分布参数时，终端可以先对第一信息进行第一预处理以得到第一处理信息，然后计算第一处理信息在该数学分布中的概率参数。最后，终端可以根据计算得到的概率参数和对应的判断阈值判断第一信息是否适用于待应用的AI编码器。
可选地，当N等于1时，终端可以计算这一个AI编码器对应的概率参数，并根据该概率参数和判断阈值的关系判断第一信息是否适用于该AI编码器。可选地，当N大于1时，终端可以分别计算N个AI编码器对应的概率参数，并根据N个概率参数和判断阈值的关系判断是否存在第一信息适用的AI编码器，可选地，以及第一信息适用的AI编码器。概率参数越大则表示第一信息越适用于对应的AI编码器，即应用该组AI编解码器对第一信息进行编解码的效果预期越好。当概率参数和判断阈值满足预设阈值条件时，终端可以确定第一信息适用于该概率参数对应的AI编码器。可选地，终端可以将概率参数和判断阈值满足预设阈值条件，且概率参数最大的AI编码器确定为第一信息最适用的AI编码器，即应用该组AI编解码器对第一信息进行编解码的效果预期最好。具体示例和上述情况一所示的示例类似，不再赘述。
情况三，第一配置信息为S101的情况三所示。可选地，终端可以将AI判决器的输入设置为第一信息以获取AI判决器的输出值。可选地，终端可以对第一信息进行第二预处理以得到第二处理信息，然后将AI判决器的输入设置为经过第二处理信息，以获取AI判决器的输出值。
可选地，终端可以根据AI判决器的输出值判断是否存在第一信息适用的AI编码器。
可选地，AI判决器为单输出端口，即有一个输出值。可选地，AI判决器的输出值的取值有N个(例如[1,N]，AI判决器的输出值为正整数)，这N个取值分别对应N个AI编码器，当AI判决器的输出值为第一AI编码器对应的取值时，表征第一AI编码器为AI判决器确定的N个AI编码器中第一信息适用的AI编码器。可选地，AI判决器的输出值的取值有N+1个(例如[0,N]，AI判决器的输出值为正整数)，其中N个取值(例如[1,N])分别对应N个AI编码器，这N个取值和上述取值有N个的N个取值的说明一致，不再赘述，当AI判决器的输出值为这N个取值外的1个取值(例如0)时，表征AI判决器确定第一信息不适用于N个AI编码器。
可选地，AI判决器为N输出端口，即有N个输出值，这N个输出值分别对应N个AI编码器。可选地，N个输出值中任意一个输出值的取值有2个(例如0或1)，其中1个(例如0)表征第一信息不适用于该输出值对应的AI编码器，另外1个(例如1)表征第一信息 适用于该输出值对应的AI编码器。
可选地，AI判决器为N+1输出端口，即有N+1个输出值，其中N个输出值和N输出端口的AI判决器的N个输出值的说明一致，不再赘述。可选地，另外1个输出值的取值有2个(例如0或1)，其中1个(例如0)表征第一信息不适用于N个AI编码器，另外1个(例如1)表征第一信息适用于N个AI编码器中的至少一个编码器。
可选地，终端也可以根据AI判决器的输出值和输出阈值的关系判断是否存在第一信息适用的AI编码器。
可选地，N等于1，AI判决器为单输出端口，终端可以根据AI判决器的输出值判断第一信息是否适用于这一个AI编码器。例如，AI判决器的输出值分别为0或1，当输出为1时终端可以确定第一信息适用于该AI编码器，当输出为0时终端可以确定第一信息不适用于该AI编码器。或者，AI判决器输出的取值范围为[0,1]，该输出值越高表示第一信息越适用于该AI编码器，即应用该组AI编解码器对第一信息进行编解码的效果预期越好。当该输出值大于输出阈值时，终端可以确定第一信息适用于该AI编码器，否则第一信息不适用于该AI编码器。不限于此，输出阈值也可以有多个，判断示例和上述情况一和情况二所示的多个判断阈值的情况类似，不再赘述。
可选地，N大于1，AI判决器为多输出端口，N个输出端口可以分别输出一个概率值，可以表示为I 1、I 2、…、I N。第i个输出端口输出的概率值I i越高，表征第一信息越适用于第i个输出端口对应的AI编码器，即应用第i个输出端口对应的AI编解码器对第一信息进行编解码的效果预期越好，其中，i的取值范围为[1,N]。当概率值大于输出阈值，即I i>I t时，终端可以确定第一信息适用于第i个输出端口对应的AI编码器。当概率值大于输出阈值且大于其他N-1个输出端口输出的概率值时，终端可以确定第一信息最适用于第i个输出端口对应的AI编码器，即应用第i个输出端口对应的AI编解码器对第一信息进行编解码的效果预期最好。
S103：终端使用第一编码器对第一信息进行编码并得到第二信息。
在一些实施例中，第一编码器可以是终端确定的第一信息(最)适用的AI编码器。可选地，当终端确定第一信息(最)适用于N个AI编码器中的第一编码器，以及终端在S103之前已接收第一编码器时，终端可以使用第一编码器对第一信息进行编码，具体流程示例可参见下图6。可选地，当终端确定第一信息(最)适用于N个AI编码器中的第一编码器时，终端可以向网络设备发送信息，以请求网络设备配置第一编码器，接收到第一编码器后可以使用第一编码器对第一信息进行编码，具体流程示例可参见下图7。可选地，终端可以直接将第一信息作为第一编码器的输入以获取输出的第二信息。可选地，终端也可以先对第一信息进行第一预处理以得到第一处理信息，然后将第一处理信息作为第一编码器的输入以获取输出的第二信息。
在一些实施例中，第一编码器可以是备用编码器。可选地，若终端确定N个AI编码器中存在第一信息适用的AI编码器，但终端还未接收到该AI编码器，则终端可以将第一信息作为备用编码器的输入以获取输出，具体流程示例可参见下图8所示。可选地，若终端确定N个AI编码器中不存在第一信息适用的AI编码器，则终端可以将第一信息作为备用编码器的输入以获取输出，具体流程示例下图9所示。
接下来示出两个上述S101-S103的过程的示例：
示例性一：假设N等于1，终端获取的第一配置信息为S101的情况二所示，具体可以包括：经过第一预处理的该组AI编解码器的训练数据集(假设为CSI)的均值H TRN，第一预处理的信息(即第一预处理包括：傅里叶变换(可以转换至角度域和时延域的)，针对部分高时 延数据的截断)，判断阈值H Th，使用码本进行编码的备用编码器，备用编码器的使用方式(即不存在待传输的信息适用的AI编码器时使用备用编码器)。
相应地，终端根据第一配置信息判断第一信息是否适用于待应用的AI编码器的方式为上图5的S102的情况一所示，具体为：终端对第一信息进行上述第一预处理，即对第一信息进行傅里叶变换以转换至角度域和时延域，然后再对部分高时延数据进行截断，以得到第一处理信息H 1。然后，终端计算第一处理信息H 1和上述均值H TRN的余弦相似度CS H，如果CS H>H Th，则终端可以确定第一信息适用于该AI编码器，并确定使用的编码器(即第一编码器)为该AI编码器。此时，在S103中，终端可以将第一处理信息H 1作为该AI编码器的输入，以得到输出H FB。如果CS H≤H Th，则终端可以确定第一信息不适用该AI编码器，并确定使用的编码器为备用编码器。此时，在S103中，终端可以使用备用编码器对第一信息进行编码。
可选地，若N大于1，终端判断N个AI编码器中是否存在待传输的信息适用的AI编码器的示例和上述示例类似，不同之处在于：终端还可以在余弦相似度大于判断阈值的至少一个AI编码器中确定出余弦相似度最大的AI编码器，该AI编码器即为终端确定的第一信息最适用的AI编码器，也是使用的编码器。
示例性二：假设N且大于1，终端获取的第一配置信息为S101的情况三所示，具体可以包括：有N个输出端口的AI判决器(其中N个输出端口分别对应N组AI编解码器，也可以理解为是对应N个AI编码器)，输出阈值Th，根据输出阈值和AI判决器的输出值判断N个AI编码器中是否存在待传输的信息(假设为CSI)适用的AI编码器的判断方法(即当AI判决器的某个输出端口的输出值大于输出阈值Th，且大于其他N-1个输出端口的输出值时，待传输的信息适用于该端口对应的AI编码器)，第一预处理的信息(即第一预处理包括：傅里叶变换，针对部分高时延数据的截断)，第二预处理的信息(第二预处理和第一预处理一致)，使用码本进行编码的备用编码器，备用编码器的使用方式(即不存在待传输的信息适用的AI编码器时使用备用编码器)。
相应地，终端根据第一配置信息判断N个AI编码器中是否存在第一信息适用的AI编码器的方式为S102的情况三所示，具体为：终端对第一信息进行上述第二预处理，即对第一信息进行傅里叶变换以转换至角度域和时延域，然后再对部分高时延数据进行截断，以得到第二处理信息H 2。然后终端将第二处理信息H 2作为AI判决器的输入，以得到N个输出端口的输出值I 1、I 2、…、I N。如果I i>Th以及I i>I j，其中i的取值范围为[1,N]，j的取值范围为[1,N]且j和i不等，则终端可以确定第一信息适用于第i个输出端口对应的AI编码器，并确定使用的编码器为该AI编码器。此时，在S103中，终端可以对第一信息进行上述第一预处理，以得到第一处理信息H 1，其中，第一预处理和第二预处理一致，因此第一处理信息H 1和第二处理信息H 2相同。然后，终端可以将第一处理信息作为该AI编码器的输入，以得到输出H FB。如果任意一个输出端口的输出值都小于或等于Th，则终端可以确定第一信息不适用于这N个AI编码器，并确定使用的编码器为备用编码器。此时，在S103中，终端可以使用备用编码器对第一信息进行编码。
S104：终端向网络设备发送第二信息和第一指示信息。
可选地，终端也可以不向网络设备发送第一指示信息。
可选地，第一指示信息可以用于指示终端是否使用了AI编码器。当第一指示信息用于指示终端使用了AI编码器时，第一指示信息还用于指示终端具体使用的AI编码器，例如包括终端确定的第一信息适用的一组AI编解码器的标识。此时，第一编码器为该AI编码器。当第一指示信息用于指示终端未使用AI编码器时，第一指示信息还用于指示终端使用的是备用编码器，即第一编码器为备用编码器。网络设备还可以根据第一指示信息得到：第二信息是终端使用第一编码器编码得到的，可选地，第一编码器为终端确定的第一信息(最)适用的 AI编码器。
S105：网络设备根据第一指示信息确定和第一编码器对应的第一解码器。
具体地，S105为可选的步骤。可选地，网络设备可以根据第一指示信息确定终端使用的第一编码器，可选地，确定和第一编码器对应的第一解码器。
例如，第一编码器为AI编码器，第一解码器为AI解码器。第一指示信息包括一组AI编解码器的标识，网络设备可以根据该标识确定第一编码器为该组AI编码器中的AI编码器，并根据该标识确定和该AI编码器对应的AI解码器为第一解码器。或者，第一编码器为备用编码器，第一解码器为对应的备用解码器。第一指示信息包括第一编码器的标识，网络设备可以根据该标识确定第一编码器为使用码本编码方案的编码器，并确定和该编码器对应的使用码本解码方案的解码器为第一解码器。
S106：网络设备使用第一解码器对第二信息进行解码。
具体地，S106为可选的步骤。上述S102-S106是第一信息的传输过程，即一次信息传输过程，其中，S102可以是终端获取到第一配置信息后进行的第一次判断。
S107：终端根据第一配置信息判断是否存在第三信息适用的AI编码器。
S108：终端使用第二编码器对第三信息进行编码并得到第四信息。
S109：终端向网络设备发送第四信息和第二指示信息。
S110：网络设备根据第二指示信息确定和第二编码器对应的第二解码器。
S111：网络设备使用第二解码器对第四信息进行解码。
可选地，上述S107-S111是第二信息的传输过程，也为一次信息传输过程，具体和S102-S106所示的传输过程类似，只是传输时刻和上述第一信息的传输时刻不同，其中，S107可以是终端获取到第一配置信息后进行的第二次判断。可选地，在具体实现中，终端可以持续获取到待传输的信息，周期性或触发式地进行判断过程，即根据第一配置信息判断是否存在待传输的信息适用的AI编码器，其中触发式是指每次获取到待传输的信息时就进行判断。可选地，终端可以根据判断结果选择使用的编码器，也就是说，终端可以进行多次信息传输过程。其中，每次信息传输过程和上述一次传输过程类似，但由于待传输的信息可能不同，因此每次判断的判断结果和使用的编码器可以不同。例如，第一信息和第三信息可以不是同分布的，则S102和S107的判断结果可以不同，第一编码器和第二编码器可以不同(第一解码器和第二解码器也不同)。每次信息传输过程的示例可参见下图6的S203-S207、图7的S302-S308、图8的S402-S408、图9的S502-S506。
在图5所示的方法中，终端可以判断是否存在待传输的信息适用的AI编码器，即判断待传输的信息和AI编解码器的训练数据集是否同分布，从而避免通过AI编解码器对和训练数据集不是同分布的信息进行编码和解码，而带来的信息失真，通信系统的性能恶化不可控的情况。并且，待传输的信息通常较大，终端可以根据不同的待传输的信息快速灵活地切换使用的编码器，相比终端向网络设备发送未经编码或经过高保真编码的信息，由网络设备判断并指示终端使用的编码器，终端自行判断的传输开销更小，时延也更小。
在一些实施例中，网络设备可以在终端根据第一配置信息判断是否存在待传输的信息适用的AI编码器之前，向终端发送D个AI编码器，D为正整数，D小于或等于N，此时一次信息传输过程(即上图5的S102-S106)的示例可参见下图6。
请参见图6，图6是本申请实施例提供的又一种信息编码的控制方法的流程示意图。该方法可以包括但不限于如下步骤：
S201：网络设备向终端发送D个AI编码器。
具体地，S201为可选的步骤。
在一些实施例中，S201之前，该方法还可以包括：终端向网络设备上报终端的性能参数，网络设备根据终端的性能参数判断终端是否满足N个AI编码器的性能要求。D个AI编码器为N个AI编码器中网络设备确定的终端满足性能要求的AI编码器，D小于或等于N。不限于此，网络设备也可以直接向终端发送网络设备获取的N个AI编码器，此时D等于N。
S202：网络设备向终端发送第一配置信息。
具体地，S202为可选的步骤。可选地，第一配置信息可以包括N个AI编码器的统计信息，也可以包括D个AI编码器的统计信息。第一配置信息的说明可参见上图5的S101的说明，不再赘述。
S203：终端根据第一配置信息确定第一信息适用于D个AI编码器中的第一AI编码器。
具体地，S203为可选的步骤。S203所示的判断过程和上图5的S102类似，不同之处在于，终端可以基于D个AI编码器进行判断，而不是基于N个AI编码器进行判断，并且判断结果为终端确定存在第一信息适用的AI编码器，且该AI编码器为第一AI编码器。
S204：终端使用第一AI编码器对第一信息进行编码并得到第五信息。
具体地，S204和上图5的S103一致，只是S103中的第一编码器为S204中的第一AI编码器。
S205：终端向网络设备发送第五信息和第三指示信息。
可选地，终端也可以不向网络设备发送第三指示信息。
可选地，第三指示信息可以用于指示终端使用了AI编码器，且终端使用的AI编码器为第一AI编码器。可选地网络设备可以根据第三指示信息得到：第五信息是终端使用第一AI编码器得到的，可选地，第一AI编码器为终端确定的第一信息(最)适用的AI编码器。
S206：网络设备根据第三指示信息确定和第一AI编码器对应的第一AI解码器。
具体地，S206为可选的步骤。可选地，网络设备可以根据第一指示信息确定终端使用的第一AI编码器，可选地，确定和第一AI编码器对应同组的第一AI解码器。
S207：网络设备使用第一AI解码器对第五信息进行解码。
具体地，S207为可选的步骤。
可选地，S207之后，终端还可以进行其他信息传输过程。示例性地，假设终端获取到待传输的第三信息，终端可以根据第一配置信息判断D个AI编码器中是否存在第三信息适用的AI编码器。可选地，若存在第三信息适用的AI编码器，例如也为第一AI编码器(终端已接收到第一AI编码器)，则终端可以使用该AI编码器对第三信息进行编码，并向网络设备发送使用了该AI编码器的指示信息。可选地，若存在第三信息适用的AI编码器且终端未接收到该AI编码器，则终端可以向网络设备发送信息，以请求网络设备配置该AI编码器，然后使用该AI编码器对第三信息进行编码，并向网络设备发送使用了该AI编码器的指示信息，具体过程和下图7的S302-S308类似。可选地，虽然存在第三信息适用的AI编码器，但终端未接收到该AI编码器，则终端可以使用备用编码器对第三信息进行编码并向网络设备发送使用了备用编码器的指示信息，具体过程和下图8的S402-S408类似。可选地，若不存在第三信息适用的AI编码器，则终端可以按照第一配置信息回退至备用编码方案，即使用备用编码器对第三信息进行编码，并向网络设备发送使用了备用编码器的指示信息，具体过程和下图9的S502-S506类似。
不限于上述列举的情况，在具体实现中，若D小于N，终端可以仍然基于N个AI编码 器判断是否存在待传输的信息适用的AI编码器。当终端确定待传输的信息适用于AI编码器X(属于N个AI编码器中除D个AI编码器之外的一个AI编码器)，终端可以向网络设备发送信息，以请求网络设备配置AI编码器X，然后使用AI编码器X进行编码，具体过程和下下图7的S302-S308类似，或者，终端可以使用备用编码器进行编码，具体过程和下图8的S402-S408类似。
在一些实施例中，当终端确定待传输的信息适用于第一AI编码器，以及终端未接收到第一AI编码器时，终端可以向网络设备发送信息，以请求网络设备配置第一AI编码器，然后使用第一AI编码器对待传输的信息进行编码，此时一次信息传输过程(即上图5的S102-S106)的示例可参见下图7。
请参见图7，图7是本申请实施例提供的又一种信息编码的控制方法的流程示意图。该方法可以包括但不限于如下步骤：
S301：网络设备向终端发送第一配置信息。
具体地，S301为可选的步骤。第一配置信息的说明可参见上图5的S101的说明，不再赘述。
S302：终端根据第一配置信息确定第一信息适用于第一AI编码器。
具体地，S302为可选的步骤。S302所示的判断过程和上图5的S102一致，只是判断结果为终端确定存在第一信息适用的AI编码器，且该AI编码器为第一AI编码器。由于终端未接收到第一AI编码器，因此终端可以向网络设备发送信息，以请求网络设备配置第一AI编码器，即执行S303。
S303：终端向网络设备发送第四指示信息。
具体地，S303为可选的步骤。
示例性地，第四指示信息可以包括第一AI编码器所在的AI编解码器组的标识，网络设备可以根据该标识确定终端请求配置的编码器为该组AI编解码器中的AI编码器。
S304：响应于第四指示信息，网络设备向终端发送第一AI编码器。
具体地，S304为可选的步骤。
在一些实施例中，若终端在S304之前上报给终端的性能参数，则网络设备接收到第四指示信息后，可以先根据终端的性能参数判断终端是否满足第一AI编码器的性能要求，满足时可以直接发送第一AI编码器，否则可以向终端发送相应的指示信息(例如不满足性能要求的指示等)。
S305：终端使用第一AI编码器对第一信息进行编码并得到第五信息。
S306：终端向网络设备发送第五信息和第三指示信息。
可选地，终端也可以不向网络设备发送第三指示信息。
S307：网络设备根据第三指示信息确定和第一AI编码器对应的第一AI解码器。
具体地，S307为可选的步骤。
S308：网络设备使用第一AI解码器对第五信息进行解码。
具体地，S308为可选的步骤。S305-S308和上图6的S204-S207一致，不再赘述。
不限于上述列举的情况，在具体实现中，第四指示信息还可以包括S302的判断结果，即用于指示：终端确定第一信息(最)适用于第一AI编码器。终端在S306中也可以不发送第三指示信息，网络设备可以根据终端发送的第四指示信息确定终端所需的AI编码器。
可选地，S308之后，终端还可以进行其他信息传输过程。示例性地，假设终端获取到待 传输的第三信息，终端可以根据第一配置信息判断N个AI编码器中是否存在第三信息适用的AI编码器。可选地，若存在第三信息适用的AI编码器且为第一AI编码器(终端已接收到第一AI编码器)，则终端可以无需向网络设备发送请求配置第一AI编码器的信息，而是直接使用第一AI编码器对第三信息进行编码，并向网络设备发送使用了第一AI编码器的指示信息，减小了传输时延，具体过程和上图6的S203-S207类似。可选地，若存在第三信息适用的AI编码器且终端未接收到该AI编码器，则终端可以再次向网络设备发送信息，以请求网络设备配置该AI编码器，然后使用该AI编码器对第三信息进行编码，并向网络设备发送使用了该AI编码器的指示信息。可选地，虽然存在第三信息适用的AI编码器，但终端未接收到该AI编码器，则终端可以使用备用编码器对第三信息进行编码并向网络设备发送使用了备用编码器的指示信息，具体过程和下图8的S402-S408类似。可选地，若不存在第三信息适用的AI编码器，则终端可以按照第一配置信息回退至备用编码方案，即使用备用编码器对第三信息进行编码，并向网络设备发送使用了备用编码器的指示信息，具体过程和下图9的S502-S506类似。
在一些实施例中，当终端确定待传输的信息适用于第一AI编码器，以及终端未接收到第一AI编码器时，终端可以先使用备用编码器对待传输的信息进行编码，从而减小传输时延，避免影响后续的信息传输过程，此时一次信息传输过程(即上图5的S102-S106)的示例可参见下图8。
请参见图8，图8是本申请实施例提供的又一种信息编码的控制方法的流程示意图。该方法可以包括但不限于如下步骤：
S401：网络设备向终端发送第一配置信息。
具体地，S401为可选的步骤。第一配置信息的说明可参见上图5的S101的说明，不再赘述。
S402：终端根据第一配置信息确定第一信息适用于第一AI编码器。
具体地，S402为可选的步骤。S402所示的判断过程和上图5的S102一致，只是判断结果为终端确定存在第一信息适用的AI编码器，且该AI编码器为第一AI编码器。由于终端未接收到第一AI编码器，因此终端可以向网络设备信息，以请求网络设备配置第一AI编码器，即执行S405。
S403：终端使用备用编码器对第一信息进行编码并得到第六信息。
S404：终端向网络设备发送第六信息和第五指示信息。
可选地，第五指示信息可以用于指示终端使用了备用编码器，可选地，没有使用AI编码器。可选地网络设备可以根据第五指示信息得到：第六信息是终端使用备用编码器得到的。
可选地，终端也可以不向网络设备发送第五指示信息。
S405：终端向网络设备发送第四指示信息。
具体地，S405为可选的步骤。
示例性地，第四指示信息可以包括第一AI编码器所在的AI编解码器组的标识，网络设备可以根据该标识确定终端请求配置的编码器为该组AI编解码器中的AI编码器。
示例性地，终端识别到近期待传输的信息相差不大，例如10分钟内都处于体育馆的室内，用户在体育馆的室内移动时，终端待传输的CSI是同分布的。因此终端可以通过第四指示信息向网络设备请求配置第一AI编码器，便于后续可以直接通过第一AI编码器对适用于第一AI编码器的信息进行编码，无需再次请求，减小传输时延。
S406：响应于第四指示信息，网络设备向终端发送第一AI编码器。
具体地，S406为可选的步骤。
在一些实施例中，若终端在S304之前上报给终端的性能参数，则网络设备接收到第四指示信息后，可以先根据终端的性能参数判断终端是否满足第一AI编码器的性能要求，满足时可以直接发送第一AI编码器，否则可以向终端发送相应的指示信息(例如不满足性能要求的指示等)。
可选地，虽然终端确定第一信息(最)适用于第一AI编码器，但由于终端未接收到第一AI编码器，因此终端可以使用备用编码器对第一信息进行编码(即执行S403)，可选地可以向网络设备发送信息，以请求配置第一AI编码器(即执行S405)。需要说明的是，S406在S403和S405之后，但网络设备发送第一AI编码器的具体时刻不作限定，例如S406和S404、S407、S408的顺序不作限定。
其中，S403和S405的顺序不作限定。在一些实施例中，终端可以先进行编码再请求配置第一AI编码器，即S403在S405之前，但此时终端发送编码后的第六信息和请求配置第一AI编码器的顺序不作限定，即S404和S405的顺序不作限定。在另一些实施例中，终端可以先向网络设备发送信息，以请求配置第一AI编码器，并在接收到第一AI编码器之前就进行编码，即S405在S403之前，S406在S403之后。
S407：网络设备根据第五指示信息确定和备用编码器对应的备用解码器。
具体地，S407为可选的步骤。
S408：网络设备使用备用解码器对第六信息进行解码。
具体地，S408为可选的步骤。
可选地，S408之后，终端还可以进行其他信息传输过程。示例性地，假设终端获取到待传输的第三信息，终端可以根据第一配置信息判断N个AI编码器中是否存在第三信息适用的AI编码器。可选地，若存在第三信息适用的AI编码器且为第一AI编码器(终端已接收到该AI编码器)，则终端可以无需向网络设备发送请求配置第一AI编码器的信息，而是直接使用第一AI编码器对第三信息进行编码，并向网络设备发送使用了第一AI编码器的指示信息，减小了传输时延，具体过程和上图6的S203-S207类似。可选地，若存在第三信息适用的AI编码器且终端未接收到该AI编码器，则终端可以向网络设备发送信息，以请求配置该AI编码器，然后使用该AI编码器对第三信息进行编码，并向网络设备发送使用了该AI编码器的指示信息，具体过程和上图7的S302-S308类似。可选地，虽然存在第三信息适用的AI编码器，但终端未接收到该AI编码器，则终端可以使用备用编码器对第三信息进行编码并向网络设备发送使用了备用编码器的指示信息，减小传输时延，保证信息编码反馈的过程不中断。可选地，若不存在第三信息适用的AI编码器，则终端可以按照第一配置信息回退至备用编码方案，即使用备用编码器对第三信息进行编码，并向网络设备发送使用了备用编码器的指示信息，具体过程和下图9的S502-S506类似。
在一些实施例中，当终端确定N个AI编码器中不存在待传输的信息适用的AI编码器时，终端可以使用备用编码器对待传输的信息进行编码，从而避免误使用AI编码器带来的传输失真，通信系统的性能恶化的情况，此时一次信息传输过程(即上图5的S102-S106)的示例可参见下图9。
请参见图9，图9是本申请实施例提供的又一种信息编码的控制方法的流程示意图。该方法可以包括但不限于如下步骤：
S501：网络设备向终端发送第一配置信息。
具体地，S501为可选的步骤。第一配置信息的说明可参见上图5的S101的说明，不再赘述。
S502：终端根据第一配置信息确定不存在第一信息适用的AI编码器。
具体地，S502为可选的步骤。S502所示的判断过程和上图5的S102一致，只是判断结果为终端确定不存在第一信息适用的AI编码器。
S503：终端使用备用编码器对第一信息进行编码并得到第六信息。
S504：终端向网络设备发送第六信息和第五指示信息。
可选地，第五指示信息可以用于指示终端使用了备用编码器，可选地，没有使用AI编码器。可选地，网络设备可以根据第五指示信息得到：第六信息是终端使用备用编码器得到的。
可选地，终端也可以不向网络设备发送第五指示信息。
S505：网络设备根据第五指示信息确定和备用编码器对应的备用解码器。
具体地，S505为可选的步骤。
S506：网络设备使用备用解码器对第六信息进行解码。
具体地，S506为可选的步骤。
可选地，备用编码器可以是使用码本等传统编码方案的编码器，也可以是其他泛化性更好的AI编码器。若备用编码器为其他泛化性更好的AI编码器，则备用编码器可以是终端判断是否存在待传输的信息适用的AI编码器之前网络设备发送的，也可以是终端判断是否存在待传输的信息适用的AI编码器之后网络设备发送的。
可选地，S506之后，终端还可以进行其他信息传输过程。示例性地，假设终端获取到待传输的第三信息，终端可以根据第一配置信息判断N个AI编码器中是否存在第三信息适用的AI编码器。可选地，若存在第三信息适用的AI编码器且终端已接收该AI编码器，则终端可以无需向网络设备发送请求配置该AI编码器的信息，而是直接使用该AI编码器对第三信息进行编码，并向网络设备发送使用了该AI编码器的指示信息，具体过程和上图6的S203-S207类似。可选地，若存在第三信息适用的AI编码器且终端未接收到该AI编码器，则终端可以向网络设备发送信息，以请求配置该AI编码器，然后使用该AI编码器对第三信息进行编码，并向网络设备发送使用了该AI编码器的指示信息，具体过程和上图7的S302-S308类似。可选地，虽然存在第三信息适用的AI编码器，但终端未接收到该AI编码器，则终端可以使用备用编码器对第三信息进行编码，并向网络设备发送使用了备用编码器的指示信息，具体过程和上图8的S402-S408类似。可选地，若不存在第三信息适用的AI编码器，则终端继续使用备用编码器对第三信息进行编码，并向网络设备发送使用了备用编码器的指示信息。
在一种可能的实现方式中，终端可以在判断是否存在待传输的信息适用的AI编码器之后(例如上图5的S102之后)，将此次判断过程得到的判断结果，可选地以及相关参数，上报给网络设备，由网络设备决策终端使用的编码器，具体示例如下图10所示。
请参见图10，图10是本申请实施例提供的又一种信息编码的控制方法的流程示意图。该方法可以包括但不限于如下步骤：
S601：网络设备向终端发送第一配置信息。
具体地，S601为可选的步骤。第一配置信息的说明可参见上图5的S101的说明。在一些实施例中，第一配置信息还可以包括判断准则。在另一些实施例中，S601之前，该方法还 可以包括：网络设备和终端协商确定判断准则，即判断准则是预先定义的。在另一些实施例中，S601之前，该方法还可以包括：终端接收网络设备发送的判断准则。其中，判断准则是针对终端根据判断阈值判断是否存在待传输的信息适用的AI编码器的判断过程(例如上图5的S102，也即S602)。判断准则可以为软判决准则或者硬判决准则。
S602：终端根据第一配置信息判断是否存在第一信息适用的AI编码器。
具体地，S602为可选的步骤。可选地，S602和上图5的S102类似，不同之处在于：上图5的S102中终端自行判断以及自行决策最终使用的编码器，因此判断准则默认为硬判决准则，最终的判断结果即为决策结果(一个编码器，即使用的编码器)。而S602中终端只自行判断但不进行决策，判断准则为硬判决准则时判断结果也为一个编码器，但当判断准则为软判决准则时判断结果可以为多个，其中任意一个判断结果对应一个AI编码器，表征待传输的信息是否适用于该AI编码器。
S603：终端向网络设备发送第一判断信息。
具体地，S603为可选的步骤。可选地，第一判断信息可以包括判断结果，可选地以及判断过程(即S602)得到的相关参数，例如获取的余弦相似度、概率参数、AI判决器的输出值等。
可选地，当终端使用硬判决准则进行判断时，若终端确定存在待传输的信息适用的AI编码器，则第一判断信息可以包括：该AI编码器所在的AI编解码器组的标识以及表征待传输的信息适用于该AI编码器的指示(即判断结果)。若终端确定不存在待传输的信息适用的AI编码器，则第一判断信息可以包括：表征待传输的信息不适用任意一个AI编码器的指示(即判断结果)。
可选地，当终端使用软判决准则进行判断时，可选地，第一判断信息可以包括：待传输的信息是否适用于N个AI编码器的指示或者待传输的信息适用于N个AI编码器的概率值(即N个判断结果)以及表征这N个判断结果的可信程度的指标。可选地，第一判断信息可以包括：待传输的信息适用的S个AI编码器的指示(即S个判断结果，S小于或等于N)以及表征待传输的信息适用于这S个AI编码器的程度的指标。例如，第一判断信息可以包括：待传输的信息适用于2个AI编码器(分别为AI编码器1和AI编码器2)，其中待传输的信息适用于AI编码器1、AI编码器2的程度的指标分别为：1、2，其中指标值越大表征适用程度越高，因此待传输的信息更适用于AI编码器2。
可选地，终端判断是否存在第一信息适用的AI编码器的方法为上图5的S102的情况一所示，第一判断信息还可以包括获取的余弦相似度，例如第一信息和N个AI编码器的统计信息的余弦相似度中最大的S个，其中S为正整数，且1≤S≤N。可选地，终端判断是否存在第一信息适用的AI编码器的方法为上图5的S102的情况二所示，第一判断信息还可以包括获取的概率参数，例如第一信息通过N个AI编码器的统计信息计算所得的N个PDF中最大的T个，其中T为正整数，且1≤T≤N。可选地，终端判断是否存在第一信息适用的AI编码器的方法为上图5的S102的情况三所示，第一判断信息还可以包括AI判断器的至少一个输出端口的输出值，例如N个输出端口的输出值，或者大于输出阈值的输出值，或者N个输出值中最大的Y个输出值，其中Y为正整数，且1≤Y≤N。
S604：网络设备根据第一判断信息确定第三编码器。
具体地，S604为可选的步骤。可选地，第三编码器为AI编码器或者备用编码器。可选地，当终端使用硬判决准则进行判断时，可选地，若终端确定存在待传输的信息适用的AI编码器，网络设备可以根据实际情况(例如终端是否满足该AI编码器的性能要求)决策终端使用的编码器是否为该AI编码器或备用编码器。可选地，若终端确定不存在待传输的信息适 用的AI编码器，网络设备可以决策终端使用的编码器为备用编码器。可选地，当终端使用软判决准则进行判断时，网络设备可以结合实际情况(例如终端是否满足AI编码器的性能要求)以及第一判断信息决策终端使用的编码器，决策方式和上图5的S102中终端决策使用的编码器的方式一致，不再赘述。
S605：网络设备向终端发送第六指示信息和第三编码器。
具体地，S605为可选的步骤。可选地，第六指示信息用于指示终端使用第三编码器进行编码。
S606：终端使用第三编码器对第一信息进行编码并得到第七信息。
在一些实施例中，第三编码器为AI编码器，终端可以先对第一信息进行第一预处理，然后再将第一预处理后的信息输入该AI编码器以获取输出的第七信息。
在一些实施例中，第三编码器为AI编码器，第六指示信息中可以包括该AI编码器对应的第一预处理的信息，则终端可以先对第一信息进行第一预处理，然后再将第一预处理后的信息输入该AI编码器以获取输出的第七信息。
S607：终端向网络设备发送第七信息。
S608：网络设备使用第三编码器对应的第三解码器对第七信息进行解码。
具体地，S608为可选的步骤。可选地，由于网络设备向终端发送了第六指示信息，因此网络设备可以默认第六指示信息中指示的第三编码器即为终端使用的编码器，可选地，网络设备可以直接使用第三编码器对应的第三解码器对第七信息进行解码。
在一些实施例中，S607中还可以包括指示终端对第七信息进行编码时使用的编码器为第三编码器的信息，网络设备可以根据该信息确定第三编码器对应的第三解码器，并使用第三解码器对第七信息进行解码。
在一些实施例中，网络设备可以在终端根据第一配置信息判断是否存在待传输的信息适用的AI编码器之前向终端发送AI编码器。例如在上述S601之前，网络设备向终端发送D个AI编码器，那么S605中网络设备可以无需发送第三编码器(AI编码器)，具体过程和上图6类似，不再赘述。
在一些实施例中，网络设备在S605中未向终端发送第三编码器且第三编码器为AI编码器，则终端可以在接收到第六指示信息之后向网络设备发送信息，以请求配置第三编码器，例如在S605之后S606之前向网络设备发送请求配置第三编码器的指示信息，具体过程和上图7类似，不再赘述。
在一些实施例中，第三编码器为备用编码器，则网络设备可以在S605中不发送第三编码器。
可选地，S608之后，终端还可以进行其他信息传输过程。其他信息传输过程中任意一个信息传输过程可以和上述S602-S608一样由终端判断由网络设备决策终端使用的编码器，也可以和上图5的S102-S106一样由终端判断和决策终端使用的编码器，例如为上图6的S203-S207、图7的S302-S308、图8的S402-S408、图9的S502-S506所示。
在图10所示的方法中，也可以由网络设备决策终端使用的编码器，减小终端的处理压力。而且终端可以无需发送指示使用的编码器的信息，减小传输开销。若决策的终端使用的编码器是网络设备未发送的AI编码器，则可以由网络设备直接发送该AI编码器，无需终端发送请求配置该AI编码器的信息，减小传输开销。
在一些实施例中，网络设备可以训练生成AI编解码器。其中，网络设备训练AI编解码 器时所使用的训练数据集可以是终端向网络设备发送的待传输信息，终端可以对待传输信息进行筛选，以将对训练AI编解码器价值高的数据(未经编码或经过高保真编码)发送给网络设备，从而提高AI编解码器的泛化性，具体示例如下图11所示。
请参见图11，图11是本申请实施例提供的一种筛选方法的流程示意图。该方法可以应用于图1所示的通信系统，也可以应用于图2所示的场景。该方法中的终端可以是图3所示的终端100。该方法中的网络设备可以是图4所示的网络设备200。该方法可以包括但不限于如下步骤：
S701：网络设备向终端发送第一配置信息。
具体地，S701为可选的步骤。第一配置信息的说明可参见上图5的S101的说明。可选地，第一配置信息还可以包括终端判断是否发送待传输信息的判断准则，该判断准则可以是硬判决准则或软判决准则。可选地，如果是硬判决准则，则终端可以直接根据待传输的信息是否适用于AI编解码器确定是否发送待传输信息。可选地，如果是软判决准则，则终端可以按照预设规则确定待传输的信息的发送概率，然后根据发送概率发送待传输的信息。可选地，第一配置信息还可以包括上述预设规则，例如判断过程得到的相关参数(例如余弦相似度、概率参数或AI判决器的输出值)和发送概率的映射关系(例如成正比，该映射关系可以通过函数和/或参数来表示)。可选地，上述预设规则由终端和网络设备预先协商配置。
S702：终端根据第一配置信息确定发送第八信息。
具体地，S702为可选的步骤。可选地，终端可以根据第一配置信息判断待传输信息是否可以被发送，即待传输信息是否对训练AI编解码器价值高。可选地，若想提高AI编解码器的泛化性，需在训练AI编解码器时将输入设置为和已有的训练数据集不是同分布的数据，也就是说不适用AI编解码器的数据对训练AI编解码器的价值更高。
可选地，如果第一配置信息中终端判断是否发送待传输的信息的判断准则是硬判决准则，则终端可以根据第一配置信息判断待传输信息是否适用于AI编解码器，若不适用则可以将待传输信息发送至网络设备，判断方法和上图5的S102所示的判断方法类似，但所需的判断结果相反。可选地，如果第一配置信息中终端判断是否发送待传输信息的判断准则是软判决准则，则终端可以基于预设规则，根据判断过程得到的相关参数确定待传输的信息的发送概率，其中，上述相关参数越大，发送概率越低。具体判断示例如下所示：
示例一：假设第一配置信息为S101的情况二所示，以及包括经过第一预处理的训练数据集的均值。N等于1，预先定义或第一配置信息中指示的一组判断阈值包括一个判断阈值CS t。终端可以先对待传输的信息进行第一预处理，然后计算第一预处理后的信息和该均值的余弦相似度CS 0。如果第一配置信息中的上述判断准则是硬判决准则，则在CS 0≤CS t的情况下，终端可以确定待传输信息不适用于这组AI编解码器，即对训练这组AI编解码器的价值高，因此可以确定发送待传输信息，否则确定不发送待传输信息。如果第一配置信息中的上述判断准则是软判决准则，则终端可以根据上述计算得到的余弦相似度CS 0确定待传输信息的发送概率。例如，若CS 0≤CS t，则以概率P1发送待传输信息，若CS 0>CS t，则以概率P2发送待传输信息，其中P2<P1。或者，根据预设规则计算得到的发送概率P＝(1-CS 0)/2，其中，由于CS 0的取值范围为[-1,1]，则P的取值范围为[0,1]。然后，终端可以根据发送概率P的抽样确定是否发送待传输的信息。或者，终端也可以将发送概率P发送给网络设备，由网络设备根据发送概率P的抽样确定是否允许终端发送第八信息。例如当抽样值表征发送时即发送待传输的信息，否则不发送待传输的信息。
示例二：假设第一配置信息为S101的情况一所示，以及包括训练数据集的数学分布的完整分布参数。N等于1，预先定义或第一配置信息中指示的一组判断阈值包括两个判断阈值PDF t1和PDF t2。终端可以计算待传输的信息在该数学分布中的PDF，即为PDF 0。如果第一配 置信息中的上述判断准则是硬判决准则，则在PDF t1≤PDF 0≤PDF t2情况下，终端可以确定待传输信息适用于这组AI编解码器，但应用这组AI编解码器对待传输的信息进行编解码的效果预期较差；若PDF 0>PDF t2，则终端可以确定待传输信息适用于这组AI编解码器，以及应用这组AI编解码器对待传输的信息进行编解码的效果预期较好，在这两种情况下，终端均可以确定不发送待传输的信息。若PDF 0<PDF t1，则终端可以确定待传输信息不适用于这组AI编解码器，即待传输的信息是对训练这组AI编解码器价值高的信息，因此可以确定发送待传输的信息。如果第一配置信息中的上述判断准则是软判决准则，则终端可以根据预设规则确定待传输信息的发送概率，例如，若PDF 0<PDF t1，则以概率P1发送待传输信息；若PDF t1≤PDF 0≤PDF t2，则以概率P2发送待传输信息；若PDF 0>PDF t2，则以概率P3发送待传输信息，其中P3<P2<P1。或者，根据预设规则计算得到的发送概率为P＝(1-PDF 0/PDF max)，其中，PDF 0的取值范围为[0,PDF max]，PDF max为该数学分布的PDF的最大值，相应地，P的取值范围为[0,1]。然后，终端可以根据发送概率P的抽样确定是否发送待传输的信息。或者，终端也可以将发送概率P发送给网络设备，由网络设备根据发送概率P的抽样确定是否允许终端发送待传输的信息。例如当抽样值表征发送时即发送待传输的信息，否则不发送待传输的信息。
示例三：假设第一配置信息为S101的情况三所示，N大于1。终端可以将AI判决器的输入设置为待传输信息，以获取AI判决器的N个输出端口的输出I 1、I 2、…、I N。如果第一配置信息中的上述判断准则是硬判决准则，则对于第i个输出端口对应的一组AI编解码器，若I i≤输出阈值I t，i的取值范围为[1,N]，则终端可以确定待传输信息不适用于该组AI编解码器，即对训练该组AI编解码器的价值高，因此可以确定发送待传输信息。如果第一配置信息中的上述判断准则是软判决准则，则对于第i个输出端口对应的一组AI编解码器，I i为待传输信息适用该组AI编解码器的概率，则根据预设规则计算得到的待传输的信息的发送概率为P＝1-I i，P的取值范围为[0,1]。然后，终端可以根据发送概率P的抽样确定是否发送待传输的信息。或者，终端也可以将发送概率P发送给网络设备，由网络设备根据发送概率P的抽样确定是否允许终端发送待传输的信息。例如当抽样值表征发送时即发送待传输的信息，否则不发送待传输的信息。
可选地，假设终端确定第八信息为对训练AI编解码器价值高的数据，因此终端可以向网络设备请求发送第八信息，即执行S702。
S703：终端向网络设备发送第一通知信息。
具体地，S703为可选的步骤。可选地，第一通知信息可以用于向网络设备请求发送数据，例如请求传输资源。可选地，第一通知信息具体可以包括上述判断过程(即S702)得到的判断结果，即待传输的第八信息对训练AI编解码器价值高。可选地，第一通知信息具体可以包括上述判断过程(即S702)得到的发送概率。可选地，第一通知信息具体可以包括上述判断过程(即S702)得到的相关参数，例如：计算出的余弦相似度、概率参数(如上述PDF 0或PMF 0)、AI判决器的输出值。
S704：响应于第一通知信息，网络设备向终端发送第二通知信息。
具体地，S704为可选的步骤。可选地，网络设备可以结合第一通知信息(例如包括S702得到的判断结果、发送概率、和/或相关参数)和资源利用情况(例如是否存在可用的上行传输资源)等实际情况判断是否允许终端发送数据。例如，第一通知信息包括S702得到的判断结果，网络设备可以在确定传输资源可用时允许终端发送数据。或者，第一通知信息包括S702得到的发送概率，网络设备可以根据发送概率的抽样确定是否允许终端发送第八信息。或者，第一通知信息包括S702得到的相关参数，网络设备可以根据终端上报的相关参数判断第八信息是否为对训练AI编解码器价值高的数据，即是否可以被发送，判断方式可以和S702所示的判断过程一致，只是此时S702中终端可以仅计算出相关参数，但不判断第八信息是否为对 训练AI编解码器价值高的数据，即S702可以无需得到判断结果。在上述基础上，网络设备可以在上行传输资源可用的情况下允许终端发送第八信息。可选地，第二通知信息可以用于指示允许终端发送数据，可选地以及网络设备为终端分配的传输资源的相关信息，例如传输频段的标识。可选地，第二通知信息也可以用于指示拒绝终端发送数据，可选地，终端接收到第二通知信息时可以取消发送数据。
S705：终端向网络设备发送第八信息。
可选地，假设第二通知信息用于指示允许终端发送数据，则终端可以向网络设备发送未经过编码的第八信息或者经过高保真编码的第八信息，保证数据的完整性，提高训练AI编解码器的准确性。
S706：终端根据第一配置信息确定不发送第九信息。
具体地，S706为可选的步骤。可选地，上述S702-S705可以是终端获取到第一配置信息后第一次判断是否发送待传输的信息，判断结果为确定发送，上述S706可以是终端获取到第一配置信息后第二次判断是否发送待传输的信息，判断结果为确定不发送。可选地，在具体实现中，终端可以持续获取到待传输的信息，周期性或触发式地进行多次判断，即根据第一配置信息判断是否上环待传输的信息。其中，每次判断和上述第一次判断、第二次判断类似，但由于待传输的信息可能不同，因此每次得到的判断结果可以不同，例如为确定发送或确定不发送。
在图11所示的方法中，待传输的信息通常较大，终端可以对待传输的信息进行筛选，以将对训练AI编解码器价值高的数据发送给网络设备，从而避免终端发送对训练AI编解码器价值低的数据给网络设备，由于训练数据集不均衡带来的AI编解码器性能较差的情况，也避免了上行带宽被浪费。
可选地，上述AI编解码器的信息(例如AI编解码器的标识)，也可以理解为是该AI编解码器中AI编码器的信息(例如AI编码器的标识)，或者也可以理解为是该AI编解码器中AI解码器的信息(例如AI解码器的标识)，也可以反过来理解。
在一些实施例中，网络设备向终端发送的AI判决器可以是独立训练生成的，具体示例如下图12所示。
请参见图12，图12示例性示出一种AI判决器的生成过程。
如图12的(A)所示，训练生成用于判断待传输的信息是否适用于一组AI编解码器的AI判决器时，可以将训练数据集分为两个集合：R(1)和R(2)，其中，R(1)可以是和这组AI编解码器的训练数据集同分布的数据，例如就为这组AI编解码器的训练数据集的集合，R(2)可以是和这组AI编解码器的训练数据集不是同分布的数据，例如为其他组AI编解码器的训练数据集的集合。可选地，AI判决器的AI模型可以设置为至少一层神经网络，该AI模型的最后一层假设设置为Sigmoid函数，则输出值的取值范围为[0,1]。可选地，AI判决器的输入的数据类型可以设置为CSI，输出端口可以为一个。可选地，在训练AI判决器时，可以在将输入设置为R(1)中的数据时，将输出设置为1，以及在将输入设置为R(2)中的数据时，将输出设置为0。可选地，后续使用该AI判决器进行推理时，可以将待传输的信息设置为该AI判决器的输入，以获取该AI判决器的输出值。输出值越大，表征待传输的信息越适用于该AI编解码器，即应用该AI编解码器的效果越好。可选地，当输出值大于输出阈值(例如为0.5)时，可以表征待传输的信息适用于该AI编码器，否则表征待传输的信息不适用于该AI编码器。
如图12的(B)所示，训练生成用于判断待传输的信息是否适用于N组AI编解码器的 AI判决器时，可以将训练数据集分为N个集合：H(1)、H(2)、…、H(N)，其中，H(i)可以是和第i组AI编解码器的训练数据集同分布的数据，例如就为第i组AI编解码器的训练数据集的集合，i为正整数，i的取值范围为[1,N]。可选地，AI判决器的AI模型可以设置为至少一层神经网络，该AI模型的最后一层假设设置为Softmax函数，则每个输出端口的输出的取值范围均为[0,1]，且所有端口输出值的和为1。可选地，AI判决器的输入的数据类型可以设置为CSI，输出端口可以为N个：端口1、端口2、…端口N，其中端口i可以对应第i组AI编解码器，即输出端口i的输出值表征待传输的信息是否适用于第i组AI编解码器。可选地，在训练AI判决器时，可以在将输入设置为H(i)中的数据时，将输出端口i的输出值设置为1，其他输出端口的输出值设置为0。例如，输入为H(1)中的数据时，输出端口1的输出值设置为1，其他N-1个输出端口的输出值设置为0。这样训练得到的AI判决器也可理解为是N分类判决器。可选地，后续使用该AI判决器进行推理时，可以将待传输的信息设置为该AI判决器的输入，以获取该AI判决器的N个输出端口的输出值。可以通过比较N个输出端口的输出值确定出待传输的信息最适用的AI编解码器，例如当输出端口i的输出值大于输出阈值(例如为0.5)，以及大于其他N-1个输出端口的输出值时，可以表征待传输的信息最适用于第i组AI编解码器。也就是说，输出端口i的输出值越大，表征待传输的信息越适用于第i组AI编解码器，即应用第i组AI编解码器对待传输的信息进行编解码的效果预期越好。
可选地，上述A“大于”B，也可以替换为(-A)“小于”(-B)。并且，上述“大于”某个阈值，也可以替换为“大于或等于”，也可以反过来替换。类似地，上述“小于”某个阈值，也可以替换为“小于或等于”，也可以反过来替换。
在一些实施例中，网络设备向终端发送的AI判决器可以是迁移学习得到的，具体示例如下图13所示。
请参见图13，图13示例性示出又一种AI判决器的生成过程。
如图13的(A)所示，可选地，N个AI编码器(AI编码器1、AI编码器2、…、AI编码器N)可以和AI判决器共用一个共享骨干网络，可选地，可以先用所有数据统一训练一个AI编解码器，然后保留该AI编解码器中AI编码器部分的浅层网络(例如前F层网络，F为正整数，且F小于AI编码器的总层数)作为共享骨干网络。可选地，保持共享骨干网络不变(即在后续的训练中参数不再更新)，后面分别对接不同的AI编码器子网络及AI解码器，使用各组AI编解码器对应的训练数据集训练生成最终的各组AI编解码器。可选地，保持共享骨干网络不变，后面对接AI判决器子网络，使用所有数据训练生成N分类判决器(即AI判决器)。即共享骨干网络的训练需要在各子网络训练之前完成。可选地，训练生成AI判决器时，可以保持共享骨干网络不变训练子网络k，具体示例可参见图13的(B)。训练生成AI编解码器i时，可以保持共享骨干网络不变训练(AI编码器i对应的)子网络i和AI解码器i，具体示例可参见图13的(C)所示，其中i为正整数，i的取值范围为[1,N]。
示例性地，共享骨干网络可以包括至少一层(可选地，任意一层也可以为一个包括多层的模块)，子网络可以是一层(可选地，这一层也可以为一个包括多层的模块)。
如图13的(B)所示，可选地，训练AI判决器时，可以将子网络k的最后一层设置为Softmax函数。训练AI判决器时，可以在将输入设置为H(i)中的数据时，将输出端口i的输出值设置为1，其他输出端口的输出值设置为0。例如，输入为H(N)中的数据时，输出端口N的输出值设置为1，其他N-1个输出端口的输出值设置为0。可选地，在训练过程中只更新子网络k的参数，不更新共享骨干网络的参数。可选地，后续使用该AI判决器进行推理时， 可以将待传输的信息设置为该AI判决器的输入，以获取该AI判决器的N个输出端口的输出值。可以通过比较N个输出端口的输出值确定出待传输的信息最适用的AI编码器，具体示例可参见上图12的(B)的说明。
如图13的(C)所示，可选地，AI编码器i和同组对应的AI解码器i一起进行训练，可以构成一个大的AI模型。可选地，训练该AI模型时，将输入设置为H(i)，将输出同样设置为上述H(i)，即输入数据和输出数据相同。可选地，在训练过程中只更新子网络i和AI解码器i的参数，不更新共享骨干网络的参数。该AI模型训练完成后即可得到对应的AI编码器i和AI解码器i。
在一些实施例中，在训练AI判决器时，可以对训练数据集进行第二预处理，然后将第二预处理后的数据作为AI判决器的输入。可选地，在进行推理时，也可以先对待传输的信息进行第二预处理，然后将第二预处理后的信息作为AI判决器的输入以获取输出。
本申请中，网络设备向终端发送AI判决器或任意一个AI编码器后，后续若向终端发送AI判决器或其他AI编码器时，可以只发送对应的子网络，无需发送整个AI模型，减小传输资源。并且，若终端使用AI判决器判断是否存在任意一个待传输的信息(简称为信息B)适用的AI编码器，即信息B已经过共享骨干网络处理(处理后的信息简称为信息C)，则后续终端使用任意一个AI编码器对信息B进行编码时，无需再将信息B经过该AI编码器的整个AI模型进行处理，直接将信息C输入该AI编码器的子网络即可，大大减小了计算量。
不限于上图13的(C)所示实施例，在具体实现时，也可以独立训练生成AI编解码器，具体示例如下图14所示。
请参见图14，图14示例性示出一种AI编解码器的生成过程。
如图14所示，可选地，AI编码器i和同组对应的AI解码器i一起进行训练，可以构成一个大的AI模型，例如该AI模型为用于实现CSI编解码的CsiNet。可选地，训练该AI模型时，将输入(也是AI编码器i的输入)设置为H(i)，将输出(也是AI解码器i)同样设置为上述H(i)，即输入数据和输出数据相同。可选地，H(i)可以是CSI或经过预处理的CSI，AI编码器的输出H′(i)可以是CSI的编码信息或者该编码信息对应的索引，相比H(i)，H′(i)的维度被大大降低。可选地，AI编码器的输出H′(i)可以再作为AI解码器的输入，输出即为恢复的CSI。训练过程中，AI编码器i和AI解码器i的参数都可以被更新，也可以根据需要只更新AI编码器i或AI解码器i的参数。该AI模型训练完成后即可得到对应的AI编码器i和AI解码器i。
示例性地，假设网络设备为基站，终端进入基站覆盖的范围内后和基站建立连接，连接后基站可以向终端发送第一配置信息。在保持和基站的连接的情况下，假设第一时刻终端处于地理位置A且位于室内，此时终端需要向基站发送第一CSI。终端可以根据第一配置信息判断是否存在第一CSI适用的AI编码器，假设第一CSI最适用于第一AI编码器，则终端可以向基站发送请求配置第一AI编码器的信息，然后使用第一AI编码器对第一CSI进行编码。然后终端可以将编码后的CSI和指示使用了第一AI编码器的信息发送给基站，基站可以使用和第一AI编码器对应的AI解码器解码得到终端上报的第一CSI。终端可以发生移动，假设第二时刻处于地理位置B且位于室外，此时终端需要向基站发送第二CSI。终端可以根据第一配置信息判断是否存在第二CSI适用的AI编码器，假设不不存在第二CSI适用的AI编码器，则终端可以使用备用编码器对第二CSI进行编码。然后终端可以将编码后的CSI和指示使用了备用编码器的信息发送给基站，基站可以使用和备用编码器对应的备用解码器解码得到终端上报的第二CSI。
不限于上述列举的情况，在具体实现中，终端接收的AI编解码器的信息(例如上述第一 配置信息)和AI编码器也可以不是网络设备发送的，而是其他设备发送的，例如网络设备为核心网设备时，可以先将AI编解码器的信息和AI编码器发送给基站，然后由基站根据实际情况发送给终端。
可选地，AI编码器的输入可以是终端待传输的信息，也可以是终端待传输的信息经过第一预处理后得到的信息，还可以是终端待传输的信息经过其他处理得到的，例如CSI经过矩阵分解后得到的预编码矩阵指示(precoding matrix indicator，PMI)，或CSI的相关矩阵等，本申请对此不作限定。
可选地，AI判决器的输入可以是终端待传输的信息，也可以是终端待传输的信息经过第二预处理后得到的信息，还可以是终端待传输的信息经过其他处理得到的，例如CSI经过矩阵分解后得到的PMI，或CSI的相关矩阵等，本申请对此不作限定。
不限于上述说明的AI判决器，在具体实现中，AI判决器中的部分模块可以是AI模型，其他模块可以是其他处理模块，即AI判决器中的部分模块或全部模块为AI模型。
不限于上述列举的情况，在具体实现中，本申请实施例中终端也可以是其他设备，例如为基站、核心网设备等，执行操作等相关说明可参见终端的说明。类似地，网络设备也可以是其他设备，例如为终端等，执行操作等相关说明可参见网络设备的说明。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，该流程可以由计算机程序来计算机程序相关的硬件完成，该计算机程序可存储于计算机可读取存储介质中，该计算机程序在执行时，可包括如上述各方法实施例的流程。而前述的存储介质包括：只读存储器(read-only memory，ROM)或随机存取存储器(random access memory，RAM)、磁碟或者光盘等各种可存储计算机程序代码的介质。

Claims (28)

  1. 一种信息编码的控制方法，其特征在于，应用于终端，所述方法包括：
    接收第一配置信息，所述第一配置信息用于配置N个AI编码器的N组参数，N为大于1的正整数；
    向网络设备发送第一指示信息，所述第一指示信息用于指示第一编码器用于第一信息编码，所述第一编码器是根据所述N组参数和所述第一信息确定的，所述第一编码器是所述N个AI编码器中的编码器，或者，所述第一编码器是和所述N个AI编码器不同的第二编码器。
  2. 如权利要求1所述的方法，其特征在于，所述方法还包括：
    发送第二信息，所述第二信息是基于所述第一编码器对所述第一信息的编码确定的。
  3. 如权利要求1或2所述的方法，其特征在于，所述第一编码器是根据第一判断参数和第一判断阈值的关系确定的，所述第一判断参数是根据所述N组参数和所述第一信息确定的，所述第一判断参数为余弦相似度CS、概率密度函数PDF、概率质量函数PMF或者欧式距离。
  4. 如权利要求1-3任一项所述的方法，其特征在于，所述第一信息为信道状态信息CSI或上行数据。
  5. 一种信息编码的控制方法，其特征在于，应用于终端，所述方法包括：
    接收第一配置信息，所述第一配置信息用于配置AI判决器，所述AI判决器用于确定N个AI编码器中第一信息适用的AI编码器，和/或用于确定所述N个AI编码器不适用于所述第一信息编码，N为大于1的正整数；
    向网络设备发送第一指示信息，所述第一指示信息用于指示第一编码器用于所述第一信息编码，所述第一编码器是根据所述AI判决器和所述第一信息确定的，所述第一编码器是所述N个AI编码器中的编码器，或者，所述第一编码器是和所述N个AI编码器不同的第二编码器。
  6. 如权利要求5所述的方法，其特征在于，所述方法还包括：
    发送第二信息，所述第二信息是基于所述第一编码器对所述第一信息的编码确定的。
  7. 如权利要求5或6所述的方法，其特征在于，所述第一编码器是根据所述AI判决器的输出确定的，所述AI判决器的输出是根据所述第一信息和所述AI判决器得到的。
  8. 如权利要求5-7任一项所述的方法，其特征在于，所述第一信息为信道状态信息CSI或上行数据。
  9. 一种信息编码的控制方法，其特征在于，应用于终端，所述方法包括：
    接收第一配置信息，所述第一配置信息用于配置第一AI编码器的参数；
    向网络设备发送第一指示信息，所述第一指示信息用于指示第一编码器用于第一信息编码，所述第一编码器是根据所述第一AI编码器的参数和所述第一信息确定的，所述第一编码器是所述第一AI编码器，或者，所述第一编码器是和所述第一AI编码器不同的第二编码器。
  10. 如权利要求9所述的方法，其特征在于，所述方法还包括：
    发送第二信息，所述第二信息是基于所述第一编码器对所述第一信息的编码确定的。
  11. 如权利要求8或9所述的方法，其特征在于，所述第一编码器是根据第一判断参数和第一判断阈值的关系确定的，所述第一判断参数是根据所述第一AI编码器的参数和所述第一信息确定的，所述第一判断参数为余弦相似度CS、概率密度函数PDF、概率质量函数PMF或者欧式距离。
  12. 如权利要求9-11任一项所述的方法，其特征在于，所述第一信息为信道状态信息CSI或上行数据。
  13. 一种信息编码的控制方法，其特征在于，应用于终端，所述方法包括：
    接收第一配置信息，所述第一配置信息用于配置AI判决器，所述AI判决器用于确定第一AI编码器适用于第一信息编码，和/或用于确定所述第一AI编码器不适用于所述第一信息编码；
    向网络设备发送第一指示信息，所述第一指示信息用于指示第一编码器用于所述第一信息编码，所述第一编码器是根据所述AI判决器和所述第一信息确定的，所述第一编码器是所述第一AI编码器，或者，所述第一编码器是和所述第一AI编码器不同的第二编码器。
  14. 如权利要求13所述的方法，其特征在于，所述方法还包括：
    发送第二信息，所述第二信息是基于所述第一编码器对所述第一信息的编码确定的。
  15. 如权利要求13或14所述的方法，其特征在于，所述第一编码器是根据所述AI判决器的输出确定的，所述AI判决器的输出是根据所述第一信息和所述AI判决器得到的。
  16. 如权利要求13-15任一项所述的方法，其特征在于，所述第一信息为信道状态信息CSI或上行数据。
  17. 一种信息编码的控制方法，其特征在于，应用于网络设备，所述方法包括：
    向终端发送第一配置信息；所述第一配置信息用于配置N个AI编码器的N组参数；
    接收第一指示信息和第二信息，所述第二信息是所述终端基于第一信息和第一编码器确定的，所述第一指示信息用于指示所述第一编码器用于所述第一信息编码，所述第一编码器是根据所述N组参数和所述第一信息确定的，所述第一编码器是所述N个AI编码器中的编码器，或者，所述第一编码器是和所述N个AI编码器不同的第二编码器。
  18. 如权利要求17所述的方法，其特征在于，所述方法还包括：
    使用与所述第一编码器对应的第一解码器对所述第二信息进行解码。
  19. 如权利要求17或18所述的方法，其特征在于，所述第一编码器是根据第一判断参数和第一判断阈值的关系确定的，所述第一判断参数是根据所述N组参数和所述第一信息确定 的，所述第一判断参数为余弦相似度CS、概率密度函数PDF、概率质量函数PMF或者欧式距离。
  20. 如权利要求17-19任一项所述的方法，其特征在于，所述第一信息为信道状态信息CSI或上行数据。
  21. 一种信息编码的控制方法，其特征在于，应用于网络设备，所述方法包括：
    向终端发送第一配置信息；所述第一配置信息用于配置AI判决器，所述AI判决器用于确定N个AI编码器中第一信息适用的AI编码器，和/或用于确定所述N个AI编码器不适用于所述第一信息编码；
    接收第一指示信息和第二信息，所述第二信息是所述终端基于第一信息和第一编码器确定的，所述第一指示信息用于指示所述第一编码器用于所述第一信息编码，所述第一编码器是根据所述AI判决器和所述第一信息确定的，所述第一编码器是所述N个AI编码器中的编码器，或者，所述第一编码器是和所述N个AI编码器不同的第二编码器。
  22. 如权利要求21所述的方法，其特征在于，所述方法还包括：
    使用与所述第一编码器对应的第一解码器对所述第二信息进行解码。
  23. 如权利要求21或22所述的方法，其特征在于，所述第一编码器是根据所述AI判决器的输出确定的，所述AI判决器的输出是根据所述第一信息和所述AI判决器得到的。
  24. 如权利要求21-23任一项所述的方法，其特征在于，所述第一信息为信道状态信息CSI或上行数据。
  25. 一种终端，其特征在于，包括收发器、处理器和存储器，所述存储器用于存储计算机程序，所述处理器调用所述计算机程序，用于执行如权利要求1-16任一项所述的方法。
  26. 一种网络设备，其特征在于，包括收发器、处理器和存储器，所述存储器用于存储计算机程序，所述处理器调用所述计算机程序，用于执行如权利要求17-24任一项所述的方法。
  27. 一种计算机存储介质，其特征在于，所述计算机存储介质存储有计算机程序，所述计算机程序被处理器执行时，实现权利要求1-16任一项或权利要求17-24任一项所述的方法。
  28. 一种计算机程序产品，其特征在于，所述计算机程序产品在电子设备上运行时，使得所述电子设备执行权利要求1-16任一项或权利要求17-24任一项所述的方法。
PCT/CN2022/075987 2021-02-24 2022-02-11 信息编码的控制方法及相关装置 WO2022179402A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US18/547,449 US20240146582A1 (en) 2021-02-24 2022-02-11 Information encoding control method and related apparatus
EP22758767.2A EP4283894A1 (en) 2021-02-24 2022-02-11 Information encoding control method and related apparatus

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
CN202110206769 2021-02-24
CN202110206769.9 2021-02-24
CN202110376449.8 2021-04-06
CN202110376449.8A CN114978413B (zh) 2021-02-24 2021-04-06 信息编码的控制方法及相关装置

Publications (1)

Publication Number Publication Date
WO2022179402A1 true WO2022179402A1 (zh) 2022-09-01

Family

ID=82973922

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/075987 WO2022179402A1 (zh) 2021-02-24 2022-02-11 信息编码的控制方法及相关装置

Country Status (4)

Country Link
US (1) US20240146582A1 (zh)
EP (1) EP4283894A1 (zh)
CN (1) CN114978413B (zh)
WO (1) WO2022179402A1 (zh)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1237073A (zh) * 1998-05-26 1999-12-01 皇家菲利浦电子有限公司 选择源编码器的收发器和在此收发器中进行的过程
US20190303766A1 (en) * 2018-03-28 2019-10-03 Fujitsu Limited Adaptability calculation method, adaptability calculation device, identification method, identification device, and computer-readable recording medium
WO2020035684A1 (en) * 2018-08-15 2020-02-20 Imperial College Of Science, Technology And Medicine Joint source channel coding of information sources using neural networks
CN111656697A (zh) * 2018-01-26 2020-09-11 高通股份有限公司 预编码配置
CN112367675A (zh) * 2020-11-11 2021-02-12 内蒙古大学 基于自编码器的无线传感器网络数据融合方法及网络系统

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10763925B2 (en) * 2015-11-11 2020-09-01 Samsung Electronics Co., Ltd. Method and apparatus for reduced feedback MIMO
CN109474315B (zh) * 2017-09-07 2021-06-01 华为技术有限公司 一种指示及确定预编码矩阵的方法和设备
US11205121B2 (en) * 2018-06-20 2021-12-21 Disney Enterprises, Inc. Efficient encoding and decoding sequences using variational autoencoders

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1237073A (zh) * 1998-05-26 1999-12-01 皇家菲利浦电子有限公司 选择源编码器的收发器和在此收发器中进行的过程
CN111656697A (zh) * 2018-01-26 2020-09-11 高通股份有限公司 预编码配置
US20190303766A1 (en) * 2018-03-28 2019-10-03 Fujitsu Limited Adaptability calculation method, adaptability calculation device, identification method, identification device, and computer-readable recording medium
WO2020035684A1 (en) * 2018-08-15 2020-02-20 Imperial College Of Science, Technology And Medicine Joint source channel coding of information sources using neural networks
CN112367675A (zh) * 2020-11-11 2021-02-12 内蒙古大学 基于自编码器的无线传感器网络数据融合方法及网络系统

Also Published As

Publication number Publication date
EP4283894A1 (en) 2023-11-29
CN114978413B (zh) 2024-05-03
CN114978413A (zh) 2022-08-30
US20240146582A1 (en) 2024-05-02

Similar Documents

Publication Publication Date Title
US11811431B2 (en) Enhanced frequency compression for overhead reduction for CSI reporting and usage
TW201911904A (zh) 資料傳輸方法、基地台和終端
WO2019114690A1 (zh) 多天线系统中预编码的方法及装置
KR20100081333A (ko) 자가적응 코드북의 처리방법
CN109889247B (zh) 适用于窄带物联网的低开销动态反馈安全传输方法及系统
WO2022179402A1 (zh) 信息编码的控制方法及相关装置
WO2020143446A1 (zh) 一种信道状态信息发送及接收方法、终端设备和网络设备
CN114303322A (zh) 用于上行链路控制信令的装置、方法和计算机程序
US20210328708A1 (en) ACQI Decoding Confidence Detection
CN112205049B (zh) 信道状态信息反馈
CN114492784A (zh) 神经网络的测试方法和装置
WO2024065800A1 (zh) 信道状态信息的反馈方法和装置
WO2024008004A1 (zh) 一种通信方法及装置
CN111934734A (zh) Csi上报方法、csi接收方法、终端和网络侧设备
He et al. Deep CSI Feedback for FDD MIMO Systems
US20240106498A1 (en) System and method for uplink coordinated transmission in an open radio access network
WO2024078074A1 (zh) 信道状态信息的发送方法、接收方法及装置、存储介质
WO2024026793A1 (zh) 数据传输方法、装置、设备、存储介质及系统
WO2024031456A1 (zh) 通信方法、装置、设备、存储介质、芯片及程序产品
WO2022206470A1 (zh) 数据处理方法、装置、电子设备以及存储介质
WO2023116407A1 (zh) 信息处理方法、装置、终端及网络设备
WO2024051594A1 (zh) 信息传输方法、ai网络模型训练方法、装置和通信设备
WO2023115254A1 (zh) 处理数据的方法及装置
WO2023071683A1 (zh) 用于反馈信道状态的方法和装置
WO2023207920A1 (zh) 信道信息反馈方法、终端及网络侧设备

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: 22758767

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 18547449

Country of ref document: US

ENP Entry into the national phase

Ref document number: 2022758767

Country of ref document: EP

Effective date: 20230823

NENP Non-entry into the national phase

Ref country code: DE