WO2024027682A1 - 特征信息传输方法、转换信息确定方法、装置和通信设备 - Google Patents

特征信息传输方法、转换信息确定方法、装置和通信设备 Download PDF

Info

Publication number
WO2024027682A1
WO2024027682A1 PCT/CN2023/110480 CN2023110480W WO2024027682A1 WO 2024027682 A1 WO2024027682 A1 WO 2024027682A1 CN 2023110480 W CN2023110480 W CN 2023110480W WO 2024027682 A1 WO2024027682 A1 WO 2024027682A1
Authority
WO
WIPO (PCT)
Prior art keywords
node
information
network model
feature information
conversion
Prior art date
Application number
PCT/CN2023/110480
Other languages
English (en)
French (fr)
Inventor
任千尧
Original Assignee
维沃移动通信有限公司
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 维沃移动通信有限公司 filed Critical 维沃移动通信有限公司
Publication of WO2024027682A1 publication Critical patent/WO2024027682A1/zh

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00Baseband systems
    • H04L25/02Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/0202Channel estimation
    • H04L25/024Channel estimation channel estimation algorithms
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00Machine learning
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00Baseband systems
    • H04L25/02Details ; arrangements for supplying electrical power along data transmission lines
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L69/00Network arrangements, protocols or services independent of the application payload and not provided for in the other groups of this subclass
    • H04L69/04Protocols for data compression, e.g. ROHC

Definitions

  • the present application belongs to the field of communication technology, and specifically relates to a characteristic information transmission method, conversion information determination method, device and communication equipment.
  • the AI network model may include an encoding part (i.e., encoding AI network model) and a decoding part (i.e., decoding AI network model).
  • the encoding AI network model is used to encode channel information into channel feature information
  • the decoding AI network model is used to convert the encoding AI network model into The channel characteristic information output by the network model is restored to channel information.
  • the encoding AI network model and the decoding AI network model need to be jointly trained in the same device, and then the joint training
  • the obtained encoding AI network model is transmitted to the terminal, and the decoding AI network model obtained by joint training is transmitted to the base station.
  • Embodiments of the present application provide a feature information transmission method, conversion information determination method, device and communication equipment, which can reduce the transmission overhead of the AI network model.
  • a feature information transmission method which method includes:
  • the first node processes the first information into first feature information based on the first AI network model
  • the first node sends second information to the second node, where the second information includes the first feature information or second feature information, where the second feature information is a converted version of the first feature information.
  • Feature information includes the first feature information or second feature information, where the second feature information is a converted version of the first feature information.
  • a characteristic information transmission device applied to the first node, and the device includes:
  • the first processing module is used to process the first information into first feature information based on the first AI network model
  • a first sending module configured to send second information to a second node, where the second information includes the first feature information or second feature information, where the second feature information is a conversion of the first feature information. subsequent feature information.
  • a feature information transmission method including:
  • the second node receives second information from the first node, wherein the second information includes the first characteristic information or the first Two feature information, wherein the first feature information is feature information obtained by processing the first information based on the first AI network model of the first node, and the second feature information is the first feature Characteristic information after information conversion;
  • the second node performs restoration processing on the second feature information based on the fourth AI network model to obtain the first information, wherein, when the second information includes the first feature information, the The second feature information is obtained by the second node converting the first feature information.
  • a characteristic information transmission device applied to the second node, and the device includes:
  • a first receiving module configured to receive second information from a first node, where the second information includes first feature information or second feature information, where the first feature information is based on the first node Feature information obtained after the first AI network model processes the first information, and the second feature information is feature information converted from the first feature information;
  • the second processing module is configured to perform restoration processing on the second feature information based on the fourth AI network model to obtain the first information, wherein, when the second information includes the first feature information, The second feature information is obtained by the second node converting the first feature information.
  • the fifth aspect provides a method for determining conversion information, including:
  • the third node obtains the third characteristic information obtained by processing the training sample data by the first AI network model, and obtains the fourth characteristic information obtained by processing the training sample data by the second AI network model, wherein the first The AI network model is the AI network model owned by the first node, and the second AI network model is the AI network model owned by the second node;
  • the third node determines conversion information based on the third feature information and the fourth feature information, wherein the conversion information is used to convert the third feature information corresponding to the target training sample into the target training sample.
  • the target training sample is any sample in the training sample data;
  • the third node sends the conversion information to at least one of the first node and the second node.
  • a device for determining conversion information is provided, applied to a third node, and the device includes:
  • the first acquisition module is used to acquire the third characteristic information obtained by processing the training sample data by the first AI network model, and obtain the fourth characteristic information obtained by processing the training sample data by the second AI network model, wherein,
  • the first AI network model is the AI network model owned by the first node, and the second AI network model is the AI network model owned by the second node;
  • a first determination module configured to determine conversion information according to the third feature information and the fourth feature information, wherein the conversion information is used to convert the third feature information corresponding to the target training sample into the target training sample.
  • the fourth characteristic information corresponding to the training sample, the target training sample is any sample in the training sample data;
  • a second sending module is configured to send the conversion information to at least one of the first node and the second node.
  • a communication device in a seventh aspect, includes a processor and a memory.
  • the memory stores programs or instructions that can be run on the processor.
  • the program or instructions are implemented when executed by the processor. The steps of the method described in the first aspect, the third aspect or the fifth aspect.
  • a communication device including a processor and a communication interface, wherein the processor is used to process first information into first feature information based on the first AI network model; the communication interface is used to provide The second node sends second information, where the second information includes the first feature information or second feature information, where the second feature information is feature information converted from the first feature information; or,
  • the communication interface is used to receive second information from a first node, wherein the second information includes first characteristic information or second characteristic information, wherein the first characteristic information is based on the first node having
  • the first AI network model processes the first information to obtain the characteristic information, and the second characteristic information is the characteristic information after the first characteristic information is converted; the processor is configured to process the first characteristic information based on the fourth AI network model.
  • the second characteristic information is restored to obtain the first information, wherein, in the case where the second information includes the first characteristic information, the second characteristic information is configured by the second node to the The first feature information is obtained by conversion processing; or,
  • the communication interface is used to obtain the third characteristic information obtained by processing the training sample data by the first AI network model, and obtain the fourth characteristic information obtained by processing the training sample data by the second AI network model, wherein:
  • the first AI network model is an AI network model owned by the first node, and the second AI network model is an AI network model owned by the second node;
  • the processor is configured to use the third characteristic information and the fourth AI network model according to the third feature information and the fourth AI network model.
  • the communication interface is further configured to send the conversion information to at least one of the first node and the second node.
  • a ninth aspect provides a communication system, including: a first node, a second node and a third node.
  • the first node can be used to perform the steps of the characteristic information transmission method as described in the first aspect.
  • the third node The second node may be used to perform the steps of the characteristic information transmission method described in the third aspect, and the third node may be used to perform the steps of the conversion information determination method described in the fifth aspect.
  • a readable storage medium is provided. Programs or instructions are stored on the readable storage medium. When the programs or instructions are executed by a processor, the steps of the method described in the first aspect are implemented, or the steps of the method are implemented as described in the first aspect. The steps of the method described in the third aspect, or the steps of implementing the method described in the fifth aspect.
  • a chip in an eleventh aspect, includes a processor and a communication interface.
  • the communication interface is coupled to the processor.
  • the processor is used to run programs or instructions to implement the method described in the first aspect. method, or implement the method as described in the third aspect, or implement the method as described in the fifth aspect.
  • a computer program/program product is provided, the computer program/program product is stored in a storage medium, and the computer program/program product is executed by at least one processor to implement as described in the first aspect
  • the steps of the characteristic information transmission method, or the computer program/program product is executed by at least one processor to implement the steps of the characteristic information transmission method as described in the third aspect, or the computer program/program product is processed by at least one
  • the processor executes the steps to implement the conversion information determining method described in the fifth aspect.
  • the first node processes the first information into first feature information based on the first AI network model; the first node sends second information to the second node, and the second information includes the First characteristic information or second characteristic information, wherein the second characteristic information is characteristic information converted from the first characteristic information.
  • the AI network model used by the first node and the second node may not match, resulting in the first feature information compressed by the first node being unable to
  • the first node may convert the first feature information to send the second feature information that it can use or restore to the second node, or the first node may also send the second feature information to the second node.
  • First feature information and after the second node converts the first feature information into second feature information that can be used or restored by the AI network model of the second node, the AI network model is used to restore the second feature information. .
  • the first node and the second node can independently use or train their respective AI network models, thereby reducing the overhead caused by transmitting the AI network model.
  • the first node and the second node can interact with their respective AI networks. network model, thus improving the information security of the first node and the second node.
  • Figure 1 is a schematic structural diagram of a wireless communication system to which embodiments of the present application can be applied;
  • Figure 2 is a flow chart of a feature information transmission method provided by an embodiment of the present application.
  • Figure 3 is a schematic diagram of the AI network model of the first node and the second node in the embodiment of the present application;
  • Figure 4 is a flow chart of another feature information transmission method provided by an embodiment of the present application.
  • Figure 5 is a flow chart of a method for determining conversion information provided by an embodiment of the present application.
  • Figure 6 is a schematic structural diagram of a feature information transmission device provided by an embodiment of the present application.
  • Figure 7 is a schematic structural diagram of another feature information transmission device provided by an embodiment of the present application.
  • Figure 8 is a schematic structural diagram of a conversion information determining device provided by an embodiment of the present application.
  • Figure 9 is a schematic structural diagram of a communication device provided by an embodiment of the present application.
  • first, second, etc. in the description and claims of this application are used to distinguish similar objects and are not used to describe a specific order or sequence. It is to be understood that the terms so used are interchangeable under appropriate circumstances so that the embodiments of the present application can be practiced in sequences other than those illustrated or described herein, and that "first" and “second” are distinguished objects It is usually one type, and the number of objects is not limited.
  • the first object can be one or multiple.
  • “and/or” in the description and claims indicates at least one of the connected objects, and the character “/" generally indicates that the related objects are in an "or” relationship.
  • LTE Long Term Evolution
  • LTE-Advanced, LTE-A Long Term Evolution
  • CDMA Code Division Multiple Access
  • TDMA Time Division Multiple Access
  • FDMA Frequency Division Multiple Access
  • OFDMA Orthogonal Frequency Division Multiple Access
  • SC-FDMA Single-carrier Frequency Division Multiple Access
  • NR New Radio
  • FIG. 1 shows a block diagram of a wireless communication system to which embodiments of the present application are applicable.
  • the wireless communication system includes a terminal 11 and a network side device 12.
  • the terminal 11 can be a mobile phone, a tablet computer (Tablet Personal Computer), a laptop computer (Laptop Computer), or a notebook computer, a personal digital assistant (Personal Digital Assistant, PDA), a handheld computer, a netbook, or a super mobile personal computer.
  • Tablet Personal Computer Tablet Personal Computer
  • laptop computer laptop computer
  • PDA Personal Digital Assistant
  • PDA Personal Digital Assistant
  • UMPC ultra-mobile personal computer
  • UMPC mobile Internet device
  • MID mobile Internet device
  • augmented reality augmented reality, AR
  • VR virtual reality
  • robots wearable devices
  • Vehicle user equipment VUE
  • pedestrian terminal pedestrian terminal
  • PUE pedestrian terminal
  • smart home home equipment with wireless communication functions, such as refrigerators, TVs, washing machines or furniture, etc.
  • game consoles personal computers (personal computer, PC), teller machine or self-service machine and other terminal-side devices.
  • Wearable devices include: smart watches, smart bracelets, smart headphones, smart glasses, smart jewelry (smart bracelets, smart bracelets, smart rings, smart necklaces, smart anklets) bracelets, smart anklets, etc.), smart wristbands, smart clothing, etc.
  • the network side equipment 12 may include access network equipment or core network equipment, where the access network equipment may also be called wireless access network equipment, radio access network (Radio Access Network, RAN), radio access network function or wireless access network unit.
  • Access network equipment can include base stations, Wireless Local Area Network (WLAN) access points or Wireless Fidelity (WiFi) nodes, etc.
  • the base station can be called Node B, Evolved Node B (Evolved Node B).
  • the base station is not limited to specific technical terms. It needs to be explained that , in the embodiment of this application, only the base station in the NR system is taken as an example for introduction, and the specific type of the base station is not limited.
  • AI network models such as neural networks, decision trees, support vector machines, Bayesian classifiers, etc. This application takes a neural network as an example for explanation, but does not limit the specific type of AI network model.
  • the AI algorithm selected and the network model used are also different.
  • the main way to improve 5G network performance with the help of AI network models is to enhance or replace existing algorithms or processing modules with neural network-based algorithms and models.
  • neural network-based algorithms and models can achieve better performance than deterministic-based algorithms.
  • the more commonly used neural networks include deep neural networks, convolutional neural networks, and recurrent neural networks.
  • the construction, training and verification of neural networks can be achieved Work.
  • an AI network model includes an encoding part and a decoding part.
  • the encoding part (the encoding part can also be called an encoding AI network model or an encoder, that is, the AI network model used in the encoder) and the decoding part (
  • the decoding part can also be called the decoding AI network model or decoder (that is, the AI network model used in the decoder) is usually jointly trained, and after the joint training is completed, the encoding part is sent to the first node for use, Send the decoding part to the second node for use, where the first node may be a terminal or network side device in the wireless communication network, and the first node encodes the information that needs to be transmitted based on the encoding part to obtain the characteristic information of the information, The characteristic information is sent to the second node, and the second node can restore the received characteristic information based on the decoded part.
  • the second node can be a terminal or network side device different from the first node in the wireless communication network.
  • the encoding AI network model used by the first node and the decoding AI network model used by the second node can be trained independently. In this way, there is no need to transmit the encoding AI network between the first node and the second node. Model and/or decode AI network models.
  • the encoding AI network model used by the first node and the decoding AI network model used by the second node may not match. For example, the feature information output by the encoding AI network model used by the first node cannot be used by the second node.
  • the decoding AI network model used by the node decodes, or the dimension and size of the feature information output by the encoding AI network model used by the first node is equal to the input information of the decoding AI network model used by the second node, which does not match, etc.
  • the feature information output by the encoding AI network model used by the first node can be converted at the first node or the second node, so that the converted feature information can be used or processed by the second node. .
  • An embodiment of the present application provides a method for transmitting characteristic information.
  • the execution subject is a first node.
  • the method for transmitting characteristic information executed by the first node may include the following steps:
  • Step 201 The first node processes the first information into first feature information based on the first AI network model.
  • Step 202 The first node sends second information to a second node, where the second information includes the first feature information or second feature information, where the second feature information is the first feature information. Converted feature information.
  • the first node may be an information sending end, and the second node may be an information receiving end.
  • the first information is information that the first node needs to transmit to the second node, such as channel information.
  • the channel information may specifically include channel matrix information. or precoding matrix information.
  • the channel information is usually channel matrix information as an example, and no specific limitation is constituted here.
  • the first node may be a terminal or a network-side device, wherein the network-side device may be an access network device or a core network device, and the second node may be a terminal or network-side device different from the first node.
  • the first node is the terminal and the second node is the access network equipment; or the first node is the access network equipment and the second node is the core network equipment. equipment; or, the first node is a core network equipment, and the second node is another core network equipment, etc.
  • the first node is a terminal and the second node is an access network device, as an example, which does not constitute a specific limitation.
  • the above-mentioned first AI network model may be a coded AI network model.
  • the coded AI network model is used to compress and/or code the first information that needs to be transmitted by the first node to obtain the first feature information, and to obtain the first feature information.
  • the information is transmitted. In this way, compared with transmitting the first information, transmitting the first characteristic information can reduce the transmission overhead and improve the security of the transmitted information.
  • the second node may use the decoding AI network model (the decoding AI network model of the second node is referred to as the fourth AI network model in the following embodiments) to decompress and/or decode the feature information to restore
  • the first information however, the encoding AI network model used by the first node and the decoding AI network model used by the second node are not jointly trained. Therefore, the encoding AI network model used by the first node and the decoding AI network model used by the second node are not jointly trained.
  • the decoding AI network model may not match. In this case, the first feature information output by the encoding AI network model used by the first node needs to be converted to obtain a decoding AI network model that can be used or restored by the second node. the second characteristic information.
  • the first feature information can be converted into the second feature information at the first node, and the second feature information can be transmitted to the second node.
  • the second node can directly convert the second feature information into Input it to its own decoding AI network model to restore the above-mentioned first information.
  • the above-mentioned first AI network model and the conversion AI network model can be integrated into one AI network model, for example: the first AI network model and the conversion AI network model can be jointly trained, so that the fused AI When the network model inputs the first information, it can output the second feature information, wherein the conversion AI network model can convert the first feature information into the second feature information.
  • the first node may also use an independent conversion AI network model or a non-AI conversion method to convert the first feature information output by the first AI network model into the second feature information.
  • the first node can directly send the first feature information to the second node, and the second node converts and restores the first feature information.
  • the second node After the first feature information is converted into the second feature information, the second feature information can be input into its own decoding AI network model to restore the above-mentioned first information.
  • the decoding AI network model and the conversion AI network model of the above-mentioned second node can be integrated into one AI network model.
  • the fourth AI network model and the conversion AI network model can be jointly trained, so that the fusion
  • the resulting AI network model can output the first information when the first feature information is input, wherein the conversion AI network model can convert the first feature information into the second feature information.
  • the second node can also use an independent conversion AI network model or a non-AI conversion method to convert the received first feature information into second feature information that matches the decoding AI network model it has. .
  • the method further includes:
  • the first node trains to obtain the first AI network model.
  • the first node independently trains the first AI network model. For example, the first node performs joint training to obtain the first AI network model and the decoding AI network model that matches the first AI network model (hereinafter In the embodiment, the decoding AI network model that matches the first AI network model is called the third AI network model). In use, the first node can only retain the first AI network model and delete the data related to the first AI network model. Matching decoding AI network model.
  • the second node can also independently train the decoding AI network model.
  • the second node performs joint training to obtain the encoding AI network model and the decoding AI network model that matches the encoding AI network model.
  • the encoding AI network model of the second node is called the second AI network model
  • the decoding AI network model of the second node is called the fourth AI network model).
  • the second node can only retain the fourth AI network model, delete the encoded AI network model that matches the fourth AI network model.
  • the conversion module can be used to convert the feature information output by encoder 1 to obtain the same feature information as the feature information output by encoder 2.
  • the conversion module may be configured with conversion information for converting the feature information output by the encoder 1 into the feature information output by the encoder 2 .
  • the network-side device can obtain the closest sounding reference signal (Sounding Reference Signal, SRS) triggered by a channel state information (Channel State Information, CSI) report (report). ) is used as the uplink channel matrix, and the decoding AI network model is trained based on the uplink channel matrix.
  • SRS Sounding Reference Signal
  • CSI Channel State Information
  • the network side device can obtain the uplink channel matrices of at least two terminals in the same cell, so that the trained
  • the decoding AI network model is applicable to the uplink channels of at least two or all terminals in the cell.
  • the decoding AI network model is also applicable to the downlink of at least two or all terminals in the cell. channel.
  • the network side device can use the same decoding AI network model to decode the channel characteristic information reported by different terminals in the same cell, but each terminal can correspond to different conversion information.
  • the cell of the same base station includes terminal A and terminal B, and terminal A performs compression coding on the channel matrix based on the coding AI network model a to obtain the first channel characteristic information a, and terminal B performs compression coding on the channel based on the coding AI network model b.
  • the matrix is compressed and encoded to obtain the first channel characteristic information b, then the conversion AI network model a can be used to convert the first channel characteristic information a, and the second channel characteristic information that can be processed by the decoding AI network model of the base station is obtained, and The first channel characteristic information b is converted using the conversion AI network model b to obtain the second channel characteristic information that can be processed by the decoding AI network model of the base station.
  • the first node may also use other methods to obtain the first AI network model, such as obtaining the first AI network model issued by the core network device, etc.
  • the decoding AI network model of the second node can be independently trained by the second node, or the second node can also obtain the decoding AI network model from the core network device.
  • the decoding AI obtained by the second node The network model may be different from the encoded AI network model obtained by the first node. Matching, for example: the core network device can deliver the encoding AI network model to the first node according to the capability of the first node, and deliver the decoding AI network model to the second node according to the capability of the second node. At this time, the first node Neither the node nor the second node knows the AI network model used by the other party.
  • the method before the first node sends the first information to the second node, the method further includes:
  • the first node obtains conversion information, wherein the conversion information is used to convert feature information output by the first AI network model into feature information matched by the AI network model of the second node;
  • the first node converts the first feature information into the second feature information according to the conversion information.
  • the conversion information may be used to convert the first feature information into the second feature information.
  • the conversion information may include at least one of the following:
  • the first indication information is used to indicate the unit array
  • the second indication information is used to indicate the sparse matrix
  • the third instruction information is used to indicate relevant information of the conversion AI network model, and the conversion AI network model is used to convert the first feature information into the second feature information.
  • first feature information and the second feature information are feature matrices with the same bit
  • "1" and/or "1" in the first feature information can be adjusted through a unit matrix or a sparse matrix. or "0" position to obtain the second feature information.
  • an AI network model can be used to realize the conversion of the first feature information to the second feature information.
  • the AI network model is called a conversion AI network model.
  • the number of iterations required to train the conversion AI network model is less than the number of iterations to train the encoding and/or decoding AI network model. That is to say, the training sample data required to train the conversion AI network model is less and the training is complex. To a lesser extent.
  • the model size of the converted AI network model is small, and the overhead required when transmitting the converted AI network model is small, and transmitting the converted AI network model does not involve the encoding and decoding implementation process of each node, even in the first node, Transferring between the second node or even the third node will not reduce the confidentiality between the first node and the second node.
  • the first node implements conversion between the first feature information and the second feature information according to the obtained conversion information.
  • the above conversion information may be determined by the first node, or the first node may receive the conversion information from other nodes (the second node or the third node).
  • the first node receives the conversion information from the third node, where the third node may be a device other than the first node and the second node. For example, the third node sends the information to the first node and the second node.
  • the third node can know the structure, parameters, and parameters of the AI network model used by the first node and the second node respectively.
  • the conversion AI network model can be trained by the third node.
  • the training sample data required to train the conversion AI network model can come from the first node, the second node, a node dedicated to collecting training sample data, and a node that does not have the first AI network model and the second AI network model. At least one of the nodes.
  • the first node sends the first AI network model to the third node
  • the second node sends the second AI network model to the third node.
  • the third node can input the acquired training sample data into the first AI network model respectively. and the second AI network model, and analyze the processing results of the first AI network model and the second AI network model on the same training sample data to obtain the converted AI network model.
  • the processing result of the first AI network model on the training sample data is the third feature information
  • the processing result of the second AI network model on the training sample data is the fourth feature information
  • Fourth feature information corresponding to the same training sample data as the third feature information can be output.
  • the first node may directly send the third feature information to the third node
  • the second node may also directly send the fourth feature information to the third node
  • one of the first node and the second node may send the third feature information directly to the third node.
  • the three nodes send the training sample data and the characteristic information after processing the training sample data.
  • the other of the first node and the second node sends the encoded AI network model to the third node, which can also enable the third node to obtain the third feature. information and fourth feature information, where the third feature information corresponds one-to-one to the fourth feature obtained by processing the same training sample data.
  • the third node can use the third feature information as the input of the conversion AI network model, and use the fourth feature information corresponding to the third feature information as the output target of the conversion AI network model.
  • the first node determines the conversion information based on auxiliary information related to the first AI network model and the second AI network model.
  • the conversion AI network model can be trained by the first node.
  • the first node can also train and convert the AI network model based on the third feature information and the fourth feature information according to the same process as the third node training and conversion AI network model.
  • the first node can jointly train the first AI network model and the conversion AI network model based on the fourth feature information and the training sample data, wherein during the joint training process, the input of the AI network model The information is training sample data, and the output target is the fourth feature information.
  • the first node trains to obtain the conversion AI network model, including:
  • the first node acquires third feature information and receives fourth feature information from the second node, where the third feature information is obtained by processing training sample data based on the first AI network model.
  • Feature information is feature information obtained by processing the training sample data based on the second AI network model of the second node;
  • the first node trains to obtain the conversion AI network model based on the third feature information and the fourth feature information, wherein the third feature information serves as the input of the conversion AI network model, and the third feature information is used as the input of the conversion AI network model.
  • Four feature information are used as the output targets of the conversion AI network model.
  • the above-mentioned third feature information may be the output information of the first AI network model when the first node inputs the training sample data into the first AI network model.
  • the first node also needs to obtain fourth feature information when training the conversion AI network model.
  • the fourth feature information can come from the second node.
  • the second node and the first node have learned the training samples in advance. data and the order or identification of the training sample data.
  • the second node processes the training sample data into fourth feature information based on its own encoding AI network model, and sends the fourth feature information to the first node.
  • the first node can process the above training sample data into the third feature information based on the first AI network model, it can process the above training sample data into the third feature information based on the same training sample data, training sample data arranged at the same position, and training samples with the same identification. Data is used to determine the correspondence between the third characteristic information and the fourth characteristic information.
  • the first node can also learn the encoded AI network model of the second node in advance, and process the training sample data into the fourth feature information based on the encoded AI network model of the second node, where the fourth feature is obtained for the first node
  • the form of information is not specifically limited.
  • the first node receives the conversion information from the second node, wherein the second node may determine the conversion information based on auxiliary information related to the first AI network model and the second AI network model;
  • the conversion AI network model can be trained by the second node.
  • the process of training the conversion AI network model by the second node in this embodiment is similar to the process of training the AI network model by the first node in the above scenario 2.
  • the second node is training When converting the AI network model, you need to obtain at least one of the third feature information, training sample data, and the first AI network model from the first node.
  • Scenario 2 The process of training and converting the AI network model on the first node will not be described in detail here.
  • the first node needs to provide auxiliary information to the node training the conversion AI network model, and the auxiliary information is used to assist in training the conversion AI network model.
  • the node obtains the third feature information.
  • the method further includes:
  • the first node sends third information to the third node or the second node, where the third information includes relevant information of the first AI network model, or the third information includes training Sample data and third characteristic information, or the third information includes the third characteristic information;
  • the third feature information is feature information obtained after processing the training sample data based on the first AI network model, and the conversion AI network model is based on the third feature information and the fourth feature information. Obtained from at least one item of training, the fourth feature information is feature information obtained by processing the training sample data based on the second AI network model of the second node.
  • the relevant information of the first AI network model may include at least one of the model structure information, model parameter information and model file of the first AI network model.
  • the node that trains and converts the AI network model can be based on the first AI network model.
  • the relevant information of the network model implements model reasoning of the first AI network model. For example, based on the relevant information of the first AI network model, the training sample data is processed based on the first AI network model to obtain the third feature information.
  • training sample data in the process of training and converting the AI network model may be pre-stored by the third node or come from the fourth node, where the fourth node may include at least one of the following:
  • the training sample data when the training sample data comes from the first node, the training sample data may be channel information detected by the first node.
  • the characteristic information transmission method also includes:
  • the first node estimates first channel information of a first channel, wherein the training sample data includes the first channel information.
  • the conversion AI network model can be trained based on the channel information of the actual channel estimated by the first node. In this way, the matching degree between the conversion AI network model and the actual channel state of the first node can be improved.
  • the set of training sample data at this time can be fixed, and both the first node and the node training the conversion AI network model know the training sample data set.
  • a fixed training sample data set is set through protocol agreement and regular update.
  • the training sample data set is in the The order of the training sample data is fixed.
  • the first third feature information reported by the first node corresponds to the first training sample data in the training sample data set, and so on.
  • the node that trains the conversion AI network model can train the conversion AI network model based on the training sample data corresponding to the third feature information, for example: based on the correspondence between the third feature information and the training sample data, and the corresponding relationship between the fourth feature information and the training sample data.
  • the corresponding relationship between the training sample data is to correlate the third feature information and the fourth feature information corresponding to the same training sample data, so that the correlated third feature information and the fourth feature information are used as training conversion AI networks respectively. Model input information and labels.
  • the first node when reporting the third feature information, can also report the identifier (Identifier, ID) of the training sample data corresponding to the third feature information.
  • ID can indicate that the training sample data used is in ID in the training sample data set. In this way, the node that trains the conversion AI network model can find the corresponding training sample data based on this ID, thereby completing the training of the conversion AI network model.
  • the training sample data may also be channel information detected by the second node.
  • the training sample data may include the channel information of the downlink channel estimated by the terminal, or the training sample data may include the channel information of the uplink channel estimated by the base station.
  • the channel information of the uplink channel is close to the channel information of the corresponding downlink channel, so that the conversion AI network model trained based on the channel information of the uplink channel matches the downlink channel status of the terminal.
  • the node that trains the converted AI network model can also obtain training sample data through a node dedicated to collecting training sample data or a node that does not have the first AI network model and the second AI network model, which can reduce the cost of collecting training samples.
  • the complexity of the data can be obtained by the node that trains the converted AI network model.
  • the node that converts the AI network model can obtain the training sample data corresponding to the third feature information or the third feature information.
  • the feature information corresponds to the fourth feature information of the same training sample data. For example: assuming that the third node trains and converts the AI network model, the third node can obtain the third feature information from the first node and the fourth feature information from the second node.
  • the third feature information and the fourth feature information are based on The same training sample data is obtained; or, assuming that the second node jointly trains the converted AI network model and the fourth AI network model, the second AI node can obtain the third feature information and training sample data, and based on this joint training, the The third feature information is restored into the AI network model of the training sample data.
  • the first node processes the first information into first feature information based on the first AI network model; the first node sends second information to the second node, and the second information includes the First characteristic information or second characteristic information, wherein the second characteristic information is characteristic information converted from the first characteristic information.
  • the AI network model used by the first node and the second node may not match, causing the first node to compress the first information.
  • the first node can convert the first feature information to send the second feature information that it can use or restore to the second node, or the first node can also send The second node sends the first feature information, and after the second node converts the first feature information into the second feature information that can be used or restored by the AI network model of the second node, the AI network model is used to calculate the second feature information. Information is restored.
  • the first node and the second node can independently use or train their respective AI network models, thereby reducing the overhead caused by transmitting the AI network model.
  • FIG. 4 Another feature information transmission method provided by an embodiment of the present application is executed by a second node. As shown in Figure 4, the feature information transmission method executed by the second node may include the following steps:
  • Step 401 The second node receives second information from the first node, where the second information includes first feature information or second feature information, where the first feature information is based on the first node having The first AI network model processes the first information to obtain feature information, and the second feature information is the feature information converted from the first feature information.
  • Step 402 The second node restores the second feature information based on the fourth AI network model, The first information is obtained, wherein, when the second information includes the first feature information, the second feature information is obtained by the second node converting the first feature information.
  • the method embodiment shown in Figure 4 corresponds to the method embodiment shown in Figure 2.
  • the difference is that the execution subject of the method embodiment shown in Figure 2 is the first node, while the method embodiment shown in Figure 4
  • the execution subject of is the second node, and the specific meanings of the first node, the second node, the second information, the first feature information, the second feature information, the first AI network model, and the fourth AI network model in this embodiment Reference may be made to the explanations in the method embodiment shown in Figure 2, which will not be described again here.
  • the feature information transmission method before the second node restores the second feature information based on the fourth AI network model, the feature information transmission method further includes:
  • the second node trains to obtain the fourth AI network model.
  • the second node can jointly train the second AI network model and the fourth AI network model, and retain the fourth AI network model for decoding the received feature information; or, the second node can independently train the fourth AI network model.
  • the second node can jointly train the conversion AI network model and the fourth AI network model.
  • the first AI network model and the third AI network model are jointly trained, and the third AI network model matches the first AI network model and is used to process the first AI network model.
  • the subsequent feature information is restored;
  • the second AI network model is jointly trained with the fourth AI network model.
  • the second AI network model is the AI network model of the second node that matches the second feature information.
  • the fourth AI The network model matches the second AI network model and is used to restore the feature information processed by the second AI network model.
  • the feature information transmission method before the second node restores the second feature information based on the fourth AI network model, the feature information transmission method further includes:
  • the second node obtains conversion information, wherein the conversion information is used to convert the feature information output by the first AI network model into feature information matched by the AI network model of the second node;
  • the second node converts the first feature information into the second feature information according to the conversion information.
  • the second node implements conversion between the first feature information and the second feature information.
  • the conversion information includes at least one of the following:
  • the first indication information is used to indicate the unit array
  • the second indication information is used to indicate the sparse matrix
  • the third instruction information is used to indicate relevant information of the conversion AI network model, and the conversion AI network model is used to convert the first feature information into the second feature information.
  • the second node obtains the conversion information, including:
  • the second node trains to obtain the conversion AI network model; or,
  • the second node receives the third indication information from a third node or the first node, where the third node is a device other than the first node and the second node.
  • the second node trains to obtain the conversion AI network model, including:
  • the second node acquires fourth feature information and receives third feature information from the first node, where the third feature information is obtained by processing training sample data based on the first AI network model.
  • Feature information the fourth feature information is feature information obtained by processing the training sample data based on the second AI network model of the second node;
  • the second node trains and obtains the conversion AI network model based on the third feature information and the fourth feature information, wherein the third feature information is used as the input of the conversion AI network model, and the third feature information is used as the input of the conversion AI network model.
  • Four feature information are used as the output targets of the conversion AI network model.
  • the second node serves as the node for training the conversion AI network model
  • the training sample data required for training the conversion AI network model can come from at least one of the first node, the second node, and the third node
  • the specific process of the second node training and converting the AI network model based on the third feature information and the fourth feature information is the same as that in the method embodiment shown in Figure 2, the first node trains the conversion AI based on the third feature information and the fourth feature information.
  • the specific process of the network model is similar and will not be described again here.
  • the characteristic information transmission method before the second node receives the third indication information from the third node or the first node, the characteristic information transmission method further includes:
  • the second node sends fourth information to the third node or the first node, wherein the fourth information includes relevant information of the second AI network model owned by the second node, or, the The fourth information includes training sample data and fourth feature information, or the fourth information includes the fourth feature information;
  • the fourth feature information is feature information obtained after processing the training sample data based on the second AI network model, and the conversion AI network model is based on the third feature information and the fourth feature information. Obtained from at least one piece of training, the third feature information is feature information obtained after processing the training sample data based on the first AI network model.
  • the first node or the third node is used as a node for training and converting the AI network model.
  • the specific process please refer to the corresponding description in the method embodiment as shown in Figure 2, which will not be described again here.
  • the characteristic information transmission method also includes:
  • the second node estimates second channel information of a second channel, wherein the training sample data includes the second channel information.
  • the second node when the second node is a network-side device, the second node receives second information from the first node, including:
  • the second node receives a target channel state information CSI report from at least one terminal, and the first characteristic information includes channel characteristic information of the target channel carried in the target CSI report, wherein the at least one terminal includes the first characteristic information.
  • the method also includes:
  • the second node obtains the third channel information of the SRS related to the target channel
  • the second node trains the fourth AI network model according to the third channel information.
  • the fourth AI network model trained based on the third channel information can recover the channel characteristic information reported by the terminal, wherein each terminal uses its own first AI network model to encode the detected channel information, Obtain channel characteristic information.
  • the network side device can obtain the uplink channel estimation result of the closest sounding reference signal (Sounding Reference Signal, SRS) triggered by the channel state information (CSI) report (report) as the uplink channel matrix, and
  • the decoding AI network model is trained based on the uplink channel matrix.
  • the network side device can obtain the uplink channel matrices of at least two terminals in the same cell, so that the trained decoding AI network model is suitable for at least two terminals in the cell.
  • the decoding AI network model is also applicable to the downlink channels of at least two or all terminals in the cell.
  • the same decoding AI network model of the network side device is suitable for the characteristic information reported by different terminals, the characteristic information reported by the terminal can still be processed in advance using a conversion AI network model corresponding to each terminal one-to-one. Conversion processing to obtain feature information that the same decoding AI network model can process.
  • the characteristic information transmission method executed by the second node provided by the embodiment of the present application cooperates with the characteristic information transmission method executed by the first node provided by the method embodiment shown in Figure 2, and can jointly achieve the realization of the first node and the second node.
  • Nodes can independently use or train their own AI network models, thereby reducing the beneficial effect of overhead caused by transmitting AI network models.
  • the execution subject may be a characteristic information transmission device.
  • the characteristic information transmission device performing the characteristic information transmission method is taken as an example to illustrate the characteristic information transmission device provided by the embodiment of the present application.
  • a feature information transmission device provided by an embodiment of the present application may be a device in the first node. As shown in Figure 5, the feature information transmission device 500 may include the following modules:
  • the first processing module 501 is used to process the first information into first feature information based on the first AI network model;
  • the first sending module 502 is configured to send second information to a second node, where the second information includes the first feature information or second feature information, where the second feature information is the first feature information. Converted feature information.
  • the feature information transmission device 500 also includes:
  • the first training module is used to train and obtain the first AI network model.
  • the first AI network model and the third AI network model are jointly trained, and the third AI network model matches the first AI network model and is used to process the first AI network model.
  • the subsequent feature information is restored;
  • the second AI network model and the fourth AI network model are jointly trained.
  • the second AI network model is the AI network model of the second node that matches the second feature information.
  • the fourth AI network model Match the second AI network model, and be used to restore the feature information processed by the second AI network model.
  • the feature information transmission device 500 also includes:
  • the second acquisition module is used to acquire conversion information, wherein the conversion information is used to convert the feature information output by the first AI network model into feature information matched by the AI network model of the second node;
  • a first conversion module configured to convert the first feature information into the second feature information according to the conversion information.
  • the conversion information includes at least one of the following:
  • the first indication information is used to indicate the unit array
  • the second indication information is used to indicate the sparse matrix
  • the third instruction information is used to indicate relevant information of the conversion AI network model, and the conversion AI network model is used to convert the first feature information into the second feature information.
  • the second acquisition module is used to:
  • the third indication information is received from a third node or the second node, where the third node is a device other than the first node and the second node.
  • the second acquisition module includes:
  • a first acquisition unit configured to acquire third feature information and receive fourth feature information from the second node, where the third feature information is obtained after processing training sample data based on the first AI network model. Obtained feature information, the fourth feature information is feature information obtained after processing the training sample data based on the second AI network model of the second node;
  • a first training unit configured to train and obtain the conversion AI network model based on the third feature information and the fourth feature information, wherein the third feature information serves as the input of the conversion AI network model, so The fourth feature information is used as the output target of the conversion AI network model.
  • the feature information transmission device 500 also includes:
  • a third sending module configured to send third information to the third node or the second node, where the third information includes relevant information of the first AI network model, or the third information Including training sample data and third feature information, or the third information includes the third feature information;
  • the third feature information is feature information obtained after processing the training sample data based on the first AI network model, and the conversion AI network model is based on the third feature information and the fourth feature information. Obtained from at least one item of training, the fourth feature information is feature information obtained by processing the training sample data based on the second AI network model of the second node.
  • the feature information transmission device 500 also includes:
  • a first channel estimation module configured to estimate first channel information of a first channel, wherein the training sample data includes the first channel information.
  • the characteristic information transmission device 500 provided by the embodiment of the present application can implement each process implemented by the first node in the method embodiment as shown in Figure 2, and can achieve the same beneficial effects. To avoid duplication, details will not be described here.
  • Another feature information transmission device provided by the embodiment of the present application can be a device in the second node.
  • the feature information transmission device 600 can include the following modules:
  • the first receiving module 601 is configured to receive second information from the first node, where the second information includes first feature information or second feature information, where the first feature information is based on the first Feature information obtained after the first AI network model of the node processes the first information, and the second feature information is feature information converted from the first feature information;
  • the second processing module 602 is configured to perform restoration processing on the second feature information based on the fourth AI network model to obtain the first information, where, when the second information includes the first feature information , the second feature information is obtained by the second node converting the first feature information.
  • the feature information transmission device 600 also includes:
  • the second training module is used to train and obtain the fourth AI network model.
  • the first AI network model and the third AI network model are jointly trained, and the third AI network model matches the first AI network model and is used to process the first AI network model.
  • the subsequent feature information is restored;
  • the second AI network model is jointly trained with the fourth AI network model.
  • the second AI network model is the AI network model of the second node that matches the second feature information.
  • the fourth AI The network model matches the second AI network model and is used to restore the feature information processed by the second AI network model.
  • the feature information transmission device 600 also includes:
  • the third acquisition module is used to acquire conversion information, wherein the conversion information is used to convert the feature information output by the first AI network model into feature information matched by the AI network model of the second node;
  • a second conversion module configured to convert the first feature information into the second feature information according to the conversion information.
  • the conversion information includes at least one of the following:
  • the first indication information is used to indicate the unit array
  • the second indication information is used to indicate the sparse matrix
  • the third instruction information is used to indicate relevant information of the conversion AI network model, and the conversion AI network model is used to convert the first feature information into the second feature information.
  • the third acquisition module is used to:
  • the third indication information is received from a third node or the first node, where the third node is a device other than the first node and the second node.
  • the third acquisition module includes:
  • a second acquisition unit configured to acquire fourth feature information and receive third feature information from the first node
  • the third feature information is the feature information obtained after processing the training sample data based on the first AI network model
  • the fourth feature information is based on the second AI network model of the second node. Characteristic information obtained after processing the training sample data;
  • a second training unit configured to train and obtain the conversion AI network model according to the third feature information and the fourth feature information, wherein the third feature information serves as the input of the conversion AI network model, so The fourth feature information is used as the output target of the conversion AI network model.
  • the feature information transmission device 600 also includes:
  • a fourth sending module configured to send fourth information to the third node or the first node, where the fourth information includes information related to the second AI network model of the second node, or, The fourth information includes training sample data and fourth feature information, or the fourth information includes the fourth feature information;
  • the fourth feature information is feature information obtained after processing the training sample data based on the second AI network model, and the conversion AI network model is based on the third feature information and the fourth feature information. Obtained from at least one piece of training, the third feature information is feature information obtained after processing the training sample data based on the first AI network model.
  • the feature information transmission device 600 also includes:
  • the second channel estimation module is configured to estimate second channel information of the second channel, wherein the training sample data includes the second channel information.
  • the first receiving module 601 is used to:
  • the feature information transmission device 600 also includes:
  • the fourth acquisition module is used to acquire the third channel information of the SRS related to the target channel
  • a third training module configured to train the fourth AI network model according to the third channel information.
  • the characteristic information transmission device 600 provided by the embodiment of the present application can implement each process implemented by the second node in the method embodiment as shown in Figure 4, and can achieve the same beneficial effects. To avoid duplication, details will not be described here.
  • the characteristic information transmission device in the embodiment of the present application may be an electronic device, such as an electronic device with an operating system, or may be a component in the electronic device, such as an integrated circuit or chip.
  • the electronic device may be a terminal or a network-side device, or may be other devices besides the terminal and the network-side device.
  • the terminal may include, but is not limited to, the type of terminal 11 listed above
  • the network side device may include, but is not limited to, the type of network side device 12 listed above.
  • Other devices may be servers, network attached storage (Network Attached Storage, NAS), etc., which are not specifically limited in the embodiments of this application.
  • the feature information transmission device provided by the embodiment of the present application can implement each process implemented by the method embodiment shown in Figure 2 or Figure 4, and achieve the same technical effect. To avoid duplication, the details will not be described here.
  • An embodiment of the present application provides a method for determining conversion information.
  • the execution subject is a third node.
  • the method for determining conversion information executed by the third node may include the following steps:
  • Step 701 The third node obtains the third characteristic information obtained by processing the training sample data by the first AI network model, and obtains the fourth characteristic information obtained by processing the training sample data by the second AI network model, wherein:
  • the first AI network model is an AI network model owned by the first node
  • the second AI network model is an AI network model owned by the second node.
  • Step 702 The third node determines conversion information based on the third feature information and the fourth feature information, where the conversion information is used to convert the third feature information corresponding to the target training sample into the third feature information corresponding to the target training sample.
  • the fourth feature information corresponding to the target training sample which is any sample in the training sample data.
  • Step 703 The third node sends the conversion information to at least one of the first node and the second node.
  • the first node, the second node, the third node, the first AI network model, the second AI network model, the third feature information, the fourth feature information, the conversion information, and the target training sample The meaning and function of can be referred to the explanation in the method embodiment shown in Figure 2.
  • the implementation method of determining the conversion information is mainly aimed at the third node.
  • the third node updates the third feature information and the fourth feature information to train the conversion AI network model, and sends it to the first node or the second node.
  • node where the process of the third node determining the conversion information may refer to the explanation of the third node determining the conversion information in the method embodiment shown in Figure 2, which will not be described again.
  • the conversion information includes at least one of the following:
  • the first indication information is used to indicate the unit array
  • the second indication information is used to indicate the sparse matrix
  • the third instruction information is used to indicate the relevant information of the converted AI network model.
  • the converted AI network model is used to convert the first feature information into the second feature information, wherein the first feature information is the same as the first AI network model. Network model matching, the second feature information matches the second AI network model.
  • the third node obtains the third feature information obtained after the first AI network model processes the training sample data, including:
  • the third node receives fifth information from the first node, where the fifth information includes relevant information of the first AI network model;
  • the third node inputs training sample data into the first AI network model, and obtains the third feature information output by the first AI network model;
  • the third node obtains fourth feature information obtained by processing the training sample data by the second AI network model, including:
  • the third node receives sixth information from the second node, where the sixth information includes relevant information of the second AI network model;
  • the third node inputs training sample data into the second AI network model, and obtains the fourth feature information output by the second AI network model.
  • the third node can obtain the coding AI network models of the first node and the second node, and process the training sample data based on the coding AI network models of the first node and the second node respectively to obtain the determined transformation.
  • the third characteristic information and the fourth characteristic information required for the information are included in the third characteristic information and the fourth characteristic information required for the information.
  • the training sample data is pre-stored in the third node, or the training sample data comes from a fourth node, and the fourth node includes at least one of the following:
  • the third node obtains the third feature information obtained after the first AI network model processes the training sample data, including:
  • the third node receives the training sample data and the third feature information from the first node, wherein the third feature information is obtained by inputting the training sample data into the first AI network model.
  • the third node obtains the fourth feature information obtained by processing the training sample data by the second AI network model, including:
  • the third node inputs the training sample data into the second AI network model, and obtains the fourth feature information output by the second AI network model.
  • the third node obtains the fourth feature information obtained after the second AI network model processes the training sample data, including:
  • the third node receives the training sample data and the fourth feature information from the second node, wherein the fourth feature information is obtained by inputting the training sample data into the second AI network model.
  • the third node obtains the third feature information obtained after the first AI network model processes the training sample data, including:
  • the third node inputs the training sample data into the first AI network model, and obtains the third feature information output by the first AI network model.
  • the third node can be a device that is mutually trusted by the terminal and the base station.
  • the third node's solution for training the AI network model can be:
  • the terminal and the base station send the relevant information of their respective encoded AI network models to the third node, and the third node uses its own saved channel information to obtain the encoded third feature information and the fourth feature based on the encoded AI network models of the terminal and the base station respectively.
  • the terminal sends the actual estimated channel information and the third characteristic information processed by the terminal's encoded AI network model to the third node, and the base station sends relevant information of its own encoded AI network model to the third node.
  • the third node processes the channel information reported by the terminal through the base station's encoding AI network model to obtain the fourth feature information, and uses the third feature information as input to train the conversion AI network model using the so-called label of the fourth feature information.
  • the third node can obtain the third feature information obtained after the first AI network model of the first node processes the training sample data, and obtain the second AI network model of the second node to process the training sample data.
  • the fourth characteristic information obtained after processing is used to determine conversion information based on the difference between the two, so that based on the conversion information, the output result of the first AI network model can be converted into the output result of the second AI network model.
  • the execution subject may be a characteristic information transmission device.
  • the characteristic information transmission device performing the characteristic information transmission method is taken as an example to illustrate the characteristic information transmission device provided by the embodiment of the present application.
  • a conversion information determination device provided by an embodiment of the present application may be a device in a third node. As shown in Figure 8, the conversion information determination device 800 may include the following modules:
  • the first acquisition module 801 is used to acquire the third feature information obtained by processing the training sample data by the first AI network model, and obtain the fourth feature information obtained by processing the training sample data by the second AI network model, wherein , the first AI network model is the AI network model owned by the first node, and the second AI network model is the AI network model owned by the second node;
  • the first determination module 802 is configured to determine conversion information according to the third feature information and the fourth feature information, wherein the conversion information is used to convert the third feature information corresponding to the target training sample into the third feature information corresponding to the target training sample.
  • the fourth feature information corresponding to the target training sample where the target training sample is any sample in the training sample data;
  • the second sending module 803 is configured to send the conversion information to at least one of the first node and the second node.
  • the conversion information includes at least one of the following:
  • the first indication information is used to indicate the unit array
  • the second indication information is used to indicate the sparse matrix
  • the third instruction information is used to indicate the relevant information of the converted AI network model.
  • the converted AI network model is used to convert the first feature information into the second feature information, wherein the first feature information is the same as the first AI network model. Network model matching, the second feature information matches the second AI network model.
  • the first acquisition module 801 includes:
  • a first receiving unit configured to receive fifth information from the first node, where the fifth information includes relevant information of the first AI network model
  • a first processing unit configured to input training sample data into the first AI network model and obtain the third feature information output by the first AI network model
  • the first acquisition module 801 also includes:
  • a second receiving unit configured to receive sixth information from the second node, where the sixth information includes the second Information related to AI network models;
  • the second processing unit is used to input training sample data into the second AI network model, and obtain the fourth feature information output by the second AI network model.
  • the training sample data is pre-stored in the third node, or the training sample data comes from a fourth node, and the fourth node includes at least one of the following:
  • the first acquisition module 801 includes:
  • a third receiving unit configured to receive training sample data from the first node and the third feature information, where the third feature information is obtained after inputting the training sample data into the first AI network model. , the feature information output by the first AI network model;
  • a third processing unit configured to input the training sample data into the second AI network model, and obtain the fourth feature information output by the second AI network model.
  • the first acquisition module 801 includes:
  • a fourth receiving unit configured to receive training sample data from the second node and the fourth feature information, where the fourth feature information is obtained after inputting the training sample data into the second AI network model. , the feature information output by the second AI network model;
  • a fourth processing unit configured to input the training sample data into the first AI network model, and obtain the third feature information output by the first AI network model.
  • the conversion information determination device 800 provided by the embodiment of the present application can implement each process implemented by the third node in the method embodiment as shown in Figure 7, and can achieve the same beneficial effects. To avoid duplication, details will not be described here.
  • this embodiment of the present application also provides a communication device 900, which includes a processor 901 and a memory 902.
  • the memory 902 stores programs or instructions that can be run on the processor 901, such as , when the communication device 900 serves as the first node, when the program or instruction is executed by the processor 901, each step of the method embodiment shown in Figure 2 is implemented, and the same technical effect can be achieved.
  • the communication device 900 serves as the second node, when the program or instruction is executed by the processor 901, each step of the method embodiment shown in Figure 4 is implemented, and the same technical effect can be achieved.
  • the communication device 900 serves as the third node, when the program or instruction is executed by the processor 901, each step of the method embodiment shown in Figure 7 is implemented, and the same technical effect can be achieved. To avoid duplication, the details will not be described here.
  • An embodiment of the present application also provides a communication device, including a processor and a communication interface.
  • the processor when the communication device serves as the first node, the processor is used to process the first information into the first feature information based on the first AI network model; the communication interface is used to provide the communication device to the second node.
  • Send second information where the second information includes the first feature information or second feature information, wherein the second feature information is feature information converted from the first feature information; or,
  • the communication interface is used to receive second information from the first node, wherein the second information includes first feature information or second features.
  • Information wherein the first characteristic information is characteristic information obtained by processing the first information based on the first AI network model of the first node, and the second characteristic information is the conversion of the first characteristic information The resulting feature information; the processor is configured to restore the second feature information based on the fourth AI network model to obtain the first information, wherein the second information includes the first feature information
  • the second feature information is obtained by the second node converting the first feature information.
  • the communication interface is used to obtain the third feature information obtained by processing the training sample data by the first AI network model, and obtain the second AI network model.
  • the fourth characteristic information obtained after processing the training sample data wherein the first AI network model is the AI network model of the first node, and the second AI network model is the AI network model of the second node.
  • the processor is configured to determine conversion information according to the third feature information and the fourth feature information, wherein the conversion information is used to convert the third feature information corresponding to the target training sample into the target training sample.
  • the fourth characteristic information corresponding to the sample, the target training sample is any sample in the training sample data; the communication interface is also used to send the data to at least one of the first node and the second node. Conversion information.
  • This communication device embodiment corresponds to the method embodiment shown in Figure 2, Figure 4, or Figure 7.
  • Each implementation process and implementation manner of the method embodiment shown in Figure 2, Figure 4, or Figure 7 can be applied to this communication equipment implementation. example, and can achieve the same technical effect.
  • Embodiments of the present application also provide a readable storage medium, with programs or instructions stored on the readable storage medium.
  • the program or instructions are executed by the processor, the method embodiments shown in Figure 2, Figure 4, or Figure 7 are implemented. Each process can achieve the same technical effect. To avoid repetition, we will not go into details here.
  • the processor is the processor in the terminal described in the above embodiment.
  • the readable storage medium includes computer readable storage media, such as computer read-only memory ROM, random access memory RAM, magnetic disk or optical disk, etc.
  • An embodiment of the present application further provides a chip.
  • the chip includes a processor and a communication interface.
  • the communication interface is coupled to the processor.
  • the processor is used to run programs or instructions.
  • the implementation is as shown in Figure 2 or Figure 4 or Each process of the method embodiment shown in Figure 7 can achieve the same technical effect. To avoid repetition, it will not be described again here.
  • chips mentioned in the embodiments of this application may also be called system-on-chip, system-on-a-chip, system-on-chip or system-on-chip, etc.
  • Embodiments of the present application further provide a computer program/program product, the computer program/program product is stored in a storage medium, and the computer program/program product is executed by at least one processor to implement Figure 2 or Figure 4 Or each process of the method embodiment shown in Figure 7, and can achieve the same technical effect, so to avoid repetition, they will not be described again here.
  • An embodiment of the present application also provides a communication system, including: a first node, a second node and a third node.
  • the first node can be used to perform the steps of the characteristic information transmission method as shown in Figure 2.
  • the third node The second node can be used to perform the steps of the characteristic information transmission method as shown in Figure 4, and the third node can be used to perform the conversion information confirmation as shown in Figure 7. Determine the steps of the method.
  • the methods of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better. implementation.
  • the technical solution of the present application can be embodied in the form of a computer software product that is essentially or contributes to the existing technology.
  • the computer software product is stored in a storage medium (such as ROM/RAM, disk , CD), including several instructions to cause a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in various embodiments of this application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Software Systems (AREA)
  • Power Engineering (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Medical Informatics (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Computer Security & Cryptography (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本申请公开了一种特征信息传输方法、转换信息确定方法、装置和通信设备,属于通信技术领域,本申请实施例的特征信息传输方法包括:第一节点基于第一AI网络模型,将第一信息处理成第一特征信息;所述第一节点向第二节点发送第二信息,所述第二信息包括所述第一特征信息或第二特征信息,其中,所述第二特征信息为所述第一特征信息转换后的特征信息。

Description

特征信息传输方法、转换信息确定方法、装置和通信设备
相关申请的交叉引用
本申请主张在2022年08月04日在中国提交的中国专利申请No.202210933644.0的优先权,其全部内容通过引用包含于此。
技术领域
本申请属于通信技术领域,具体涉及一种特征信息传输方法、转换信息确定方法、装置和通信设备。
背景技术
在相关技术中,对借助人工智能(Artificial Intelligence,AI)网络模型来传输信道特征信息的方法进行了研究。
该AI网络模型可以包括编码部分(即编码AI网络模型)和解码部分(即解码AI网络模型),编码AI网络模型用于将信道信息编码成信道特征信息,解码AI网络模型用于将编码AI网络模型输出的信道特征信息恢复成信道信息,这样,为了使编码AI网络模型与解码AI网络模型匹配,该编码AI网络模型和解码AI网络模型需要在同一设备中进行联合训练,然后将联合训练得到的编码AI网络模型传输至终端,将联合训练得到的解码AI网络模型传输至基站。
由上可知,在相关技术中,存在由于AI网络模型的传输而增大终端和/或基站的传输开销的问题。
发明内容
本申请实施例提供一种特征信息传输方法、转换信息确定方法、装置和通信设备,能够降低AI网络模型的传输开销。
第一方面,提供了一种特征信息传输方法,该方法包括:
第一节点基于第一AI网络模型,将第一信息处理成第一特征信息;
所述第一节点向第二节点发送第二信息,所述第二信息包括所述第一特征信息或第二特征信息,其中,所述第二特征信息为所述第一特征信息转换后的特征信息。
第二方面,提供了一种特征信息传输装置,应用于第一节点,该装置包括:
第一处理模块,用于基于第一AI网络模型,将第一信息处理成第一特征信息;
第一发送模块,用于向第二节点发送第二信息,所述第二信息包括所述第一特征信息或第二特征信息,其中,所述第二特征信息为所述第一特征信息转换后的特征信息。
第三方面,提供了一种特征信息传输方法,包括:
第二节点接收来自第一节点的第二信息,其中,所述第二信息包括第一特征信息或第 二特征信息,其中,所述第一特征信息为基于所述第一节点具有的第一AI网络模型对第一信息进行处理后得到的特征信息,所述第二特征信息为所述第一特征信息转换后的特征信息;
所述第二节点基于第四AI网络模型对所述第二特征信息进行恢复处理,得到所述第一信息,其中,在所述第二信息包括所述第一特征信息的情况下,所述第二特征信息由所述第二节点对所述第一特征信息进行转换处理得到。
第四方面,提供了一种特征信息传输装置,应用于第二节点,该装置包括:
第一接收模块,用于接收来自第一节点的第二信息,其中,所述第二信息包括第一特征信息或第二特征信息,其中,所述第一特征信息为基于所述第一节点具有的第一AI网络模型对第一信息进行处理后得到的特征信息,所述第二特征信息为所述第一特征信息转换后的特征信息;
第二处理模块,用于基于第四AI网络模型对所述第二特征信息进行恢复处理,得到所述第一信息,其中,在所述第二信息包括所述第一特征信息的情况下,所述第二特征信息由所述第二节点对所述第一特征信息进行转换处理得到。
第五方面,提供了一种转换信息确定方法,包括:
第三节点获取第一AI网络模型对训练样本数据处理后得到的第三特征信息,以及获取第二AI网络模型对所述训练样本数据处理后得到的第四特征信息,其中,所述第一AI网络模型为第一节点具有的AI网络模型,所述第二AI网络模型为第二节点具有的AI网络模型;
所述第三节点根据所述第三特征信息和所述第四特征信息,确定转换信息,其中,所述转换信息用于将目标训练样本对应的第三特征信息转换为与所述目标训练样本对应的第四特征信息,所述目标训练样本为所述训练样本数据中的任一样本;
所述第三节点向所述第一节点和所述第二节点中的至少一个发送所述转换信息。
第六方面,提供了一种转换信息确定装置,应用于第三节点,该装置包括:
第一获取模块,用于获取第一AI网络模型对训练样本数据处理后得到的第三特征信息,以及获取第二AI网络模型对所述训练样本数据处理后得到的第四特征信息,其中,所述第一AI网络模型为第一节点具有的AI网络模型,所述第二AI网络模型为第二节点具有的AI网络模型;
第一确定模块,用于根据所述第三特征信息和所述第四特征信息,确定转换信息,其中,所述转换信息用于将目标训练样本对应的第三特征信息转换为与所述目标训练样本对应的第四特征信息,所述目标训练样本为所述训练样本数据中的任一样本;
第二发送模块,用于向所述第一节点和所述第二节点中的至少一个发送所述转换信息。
第七方面,提供了一种通信设备,该通信设备包括处理器和存储器,所述存储器存储可在所述处理器上运行的程序或指令,所述程序或指令被所述处理器执行时实现如第一方面或第三方面或第五方面所述的方法的步骤。
第八方面,提供了一种通信设备,包括处理器及通信接口,其中,所述处理器用于基于第一AI网络模型,将第一信息处理成第一特征信息;所述通信接口用于向第二节点发送第二信息,所述第二信息包括所述第一特征信息或第二特征信息,其中,所述第二特征信息为所述第一特征信息转换后的特征信息;或者,
所述通信接口用于接收来自第一节点的第二信息,其中,所述第二信息包括第一特征信息或第二特征信息,其中,所述第一特征信息为基于所述第一节点具有的第一AI网络模型对第一信息进行处理后得到的特征信息,所述第二特征信息为所述第一特征信息转换后的特征信息;所述处理器用于基于第四AI网络模型对所述第二特征信息进行恢复处理,得到所述第一信息,其中,在所述第二信息包括所述第一特征信息的情况下,所述第二特征信息由所述第二节点对所述第一特征信息进行转换处理得到;或者,
所述通信接口用于获取第一AI网络模型对训练样本数据处理后得到的第三特征信息,以及获取第二AI网络模型对所述训练样本数据处理后得到的第四特征信息,其中,所述第一AI网络模型为第一节点具有的AI网络模型,所述第二AI网络模型为第二节点具有的AI网络模型;所述处理器用于根据所述第三特征信息和所述第四特征信息,确定转换信息,其中,所述转换信息用于将目标训练样本对应的第三特征信息转换为与所述目标训练样本对应的第四特征信息,所述目标训练样本为所述训练样本数据中的任一样本;所述通信接口还用于向所述第一节点和所述第二节点中的至少一个发送所述转换信息。
第九方面,提供了一种通信系统,包括:第一节点、第二节点和第三节点,所述第一节点可用于执行如第一方面所述的特征信息传输方法的步骤,所述第二节点可用于执行如第三方面所述的特征信息传输方法的步骤,所述第三节点可用于执行如第五方面所述的转换信息确定方法的步骤。
第十方面,提供了一种可读存储介质,所述可读存储介质上存储程序或指令,所述程序或指令被处理器执行时实现如第一方面所述的方法的步骤,或者实现如第三方面所述的方法的步骤,或者实现如第五方面所述的方法的步骤。
第十一方面,提供了一种芯片,所述芯片包括处理器和通信接口,所述通信接口和所述处理器耦合,所述处理器用于运行程序或指令,实现如第一方面所述的方法,或实现如第三方面所述的方法,或者、实现如第五方面所述的方法。
第十二方面,提供了一种计算机程序/程序产品,所述计算机程序/程序产品被存储在存储介质中,所述计算机程序/程序产品被至少一个处理器执行以实现如第一方面所述的特征信息传输方法的步骤,或者所述计算机程序/程序产品被至少一个处理器执行以实现如第三方面所述的特征信息传输方法的步骤,或者所述计算机程序/程序产品被至少一个处理器执行以实现如第五方面所述的转换信息确定方法的步骤。
在本申请实施例中,第一节点基于第一AI网络模型,将第一信息处理成第一特征信息;所述第一节点向第二节点发送第二信息,所述第二信息包括所述第一特征信息或第二特征信息,其中,所述第二特征信息为所述第一特征信息转换后的特征信息。在第一节点 和第二节点之间通过AI网络模型来压缩和恢复第一信息时,基于第一节点与第二节点使用的AI网络模型可能不匹配,从而造成第一节点压缩得到的第一特征信息不能够被第二节点使用或恢复时,第一节点可以对第一特征信息进行转换,以向第二节点发送其能够使用或恢复的第二特征信息,或者,第一节点也可以向第二节点发送第一特征信息,并由第二节点将第一特征信息转化为该第二节点具有的AI网络模型能够使用或恢复的第二特征信息后,利用AI网络模型对该第二特征信息进行恢复处理。这样,第一节点和第二节点能够独立使用或训练各自的AI网络模型,从而能够降低因传输AI网络模型而造成的开销,同时,第一节点与第二节点之间无需交互各自使用的AI网络模型,从而提升了第一节点与第二节点的信息安全性。
附图说明
图1是本申请实施例能够应用的一种无线通信系统的结构示意图;
图2是本申请实施例提供的一种特征信息传输方法的流程图;
图3是本申请实施例中第一节点和第二节点的AI网络模型的示意图;
图4是本申请实施例提供的另一种特征信息传输方法的流程图;
图5是本申请实施例提供的一种转换信息确定方法的流程图;
图6是本申请实施例提供的一种特征信息传输装置的结构示意图;
图7是本申请实施例提供的另一种特征信息传输装置的结构示意图;
图8是本申请实施例提供的一种转换信息确定装置的结构示意图;
图9是本申请实施例提供的一种通信设备的结构示意图。
具体实施方式
下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行清楚描述,显然,所描述的实施例是本申请一部分实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员所获得的所有其他实施例,都属于本申请保护的范围。
本申请的说明书和权利要求书中的术语“第一”、“第二”等是用于区别类似的对象,而不用于描述特定的顺序或先后次序。应该理解这样使用的术语在适当情况下可以互换,以便本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施,且“第一”、“第二”所区别的对象通常为一类,并不限定对象的个数,例如第一对象可以是一个,也可以是多个。此外,说明书以及权利要求中“和/或”表示所连接对象的至少其中之一,字符“/”一般表示前后关联对象是一种“或”的关系。
值得指出的是,本申请实施例所描述的技术不限于长期演进型(Long Term Evolution,LTE)/LTE的演进(LTE-Advanced,LTE-A)系统,还可用于其他无线通信系统,诸如码分多址(Code Division Multiple Access,CDMA)、时分多址(Time Division Multiple Access,TDMA)、频分多址(Frequency Division Multiple Access,FDMA)、正交频分多址(Orthogonal  Frequency Division Multiple Access,OFDMA)、单载波频分多址(Single-carrier Frequency Division Multiple Access,SC-FDMA)和其他系统。本申请实施例中的术语“系统”和“网络”常被可互换地使用,所描述的技术既可用于以上提及的系统和无线电技术,也可用于其他系统和无线电技术。以下描述出于示例目的描述了新空口(New Radio,NR)系统,并且在以下大部分描述中使用NR术语,但是这些技术也可应用于NR系统应用以外的应用,如第6代(6th Generation,6G)通信系统。
图1示出本申请实施例可应用的一种无线通信系统的框图。无线通信系统包括终端11和网络侧设备12。其中,终端11可以是手机、平板电脑(Tablet Personal Computer)、膝上型电脑(Laptop Computer)或称为笔记本电脑、个人数字助理(Personal Digital Assistant,PDA)、掌上电脑、上网本、超级移动个人计算机(ultra-mobile personal computer,UMPC)、移动上网装置(Mobile Internet Device,MID)、增强现实(augmented reality,AR)/虚拟现实(virtual reality,VR)设备、机器人、可穿戴式设备(Wearable Device)、车载设备(Vehicle User Equipment,VUE)、行人终端(Pedestrian User Equipment,PUE)、智能家居(具有无线通信功能的家居设备,如冰箱、电视、洗衣机或者家具等)、游戏机、个人计算机(personal computer,PC)、柜员机或者自助机等终端侧设备,可穿戴式设备包括:智能手表、智能手环、智能耳机、智能眼镜、智能首饰(智能手镯、智能手链、智能戒指、智能项链、智能脚镯、智能脚链等)、智能腕带、智能服装等。需要说明的是,在本申请实施例并不限定终端11的具体类型。网络侧设备12可以包括接入网设备或核心网设备,其中,接入网设备也可以称为无线接入网设备、无线接入网(Radio Access Network,RAN)、无线接入网功能或无线接入网单元。接入网设备可以包括基站、无线局域网(Wireless Local Area Network,WLAN)接入点或无线保真(Wireless Fidelity,WiFi)节点等,基站可被称为节点B、演进节点B(Evolved Node B,eNB)、接入点、基收发机站(Base Transceiver Station,BTS)、无线电基站、无线电收发机、基本服务集(Basic Service Set,BSS)、扩展服务集(Extended Service Set,ESS)、家用B节点、家用演进型B节点、发送接收点(Transmitting Receiving Point,TRP)或所述领域中其他某个合适的术语,只要达到相同的技术效果,所述基站不限于特定技术词汇,需要说明的是,在本申请实施例中仅以NR系统中的基站为例进行介绍,并不限定基站的具体类型。
人工智能目前在各个领域获得了广泛的应用。AI网络模型有多种实现方式,例如神经网络、决策树、支持向量机、贝叶斯分类器等。本申请以神经网络为例进行说明,但是并不限定AI网络模型的具体类型。
一般而言,根据需要解决的问题的不同类型,所选取的AI算法和采用的网络模型也有所差别。借助AI网络模型提升5G网络性能的主要方法是通过基于神经网络的算法和模型增强或者替代目前已有的算法或处理模块。在特定场景下,基于神经网络的算法和模型可以取得比基于确定性算法更好的性能。比较常用的神经网络包括深度神经网络、卷积神经网络和循环神经网络等。借助已有AI工具,可以实现神经网络的搭建、训练与验证 工作。
在相关技术中,AI网络模型包括编码部分和解码部分,该编码部分(该编码部分也可以称之为编码AI网络模型或编码器,即在编码器中使用的AI网络模型)和解码部分(该解码部分也可以称之为解码AI网络模型或解码器,即在解码器中使用的AI网络模型)通常是联合训练的,并在联合训练完成后,将编码部分发送给第一节点使用,将解码部分发送给第二节点使用,其中,第一节点可以是无线通信网络中的终端或网络侧设备,该第一节点基于编码部分对需要传输的信息进行编码,得到该信息的特征信息,并将该特征信息发送至第二节点,第二节点则可以基于解码部分来恢复接收到的特征信息,其中,第二节点可以是无线通信网络中与第一节点不同的终端或网络侧设备。
由上可知,相关技术中需要联合训练编码AI网络模型和解码AI网络模型,并对编码AI网络模型和/或解码AI网络模型进行传输,以将编码AI网络模型部署在第一节点,将解码AI网络模型部署在第二节点。
而本申请实施例中,第一节点所使用的编码AI网络模型和第二节点所使用的解码AI网络模型可以是独立训练的,这样,无需在第一节点和第二节点间传输编码AI网络模型和/或解码AI网络模型。同时,第一节点所使用的编码AI网络模型和第二节点所使用的解码AI网络模型可能并不匹配,例如:第一节点所使用的编码AI网络模型所输出的特征信息不能够被第二节点所使用的解码AI网络模型解码,或者第一节点所使用的编码AI网络模型所输出的特征信息的维度、尺寸等于第二节点所使用的解码AI网络模型的输入信息不匹配等。本申请实施例中,可以在第一节点或第二节点对第一节点所使用的编码AI网络模型所输出的特征信息进行转换处理,以使转换后的特征信息能够被第二节点使用或处理。
下面结合附图,通过一些实施例及其应用场景对本申请实施例提供的特征信息传输方法、特征信息传输装置及通信设备等进行详细地说明。
请参阅图2,本申请实施例提供的一种特征信息传输方法,其执行主体是第一节点,如图2所示,该第一节点执行的特征信息传输方法可以包括以下步骤:
步骤201、第一节点基于第一AI网络模型,将第一信息处理成第一特征信息。
步骤202、所述第一节点向第二节点发送第二信息,所述第二信息包括所述第一特征信息或第二特征信息,其中,所述第二特征信息为所述第一特征信息转换后的特征信息。
其中,第一节点可以是信息发送端,第二节点可以是信息接收端,第一信息为第一节点需要传递至第二节点的信息,例如:信道信息,该信道信息具体可以包括信道矩阵信息或预编码矩阵信息,为了便于说明,本申请实施例中通常以信道信息为信道矩阵信息为例进行举例说明,在此不构成具体限定。
在实施中,第一节点可以是终端或网络侧设备,其中,网络侧设备可以是接入网设备或者是核心网设备,第二节点可以是与第一节点不同的终端或网络侧设备。例如:第一节点是终端,第二节点是接入网设备;或者,第一节点是接入网设备,第二节点是核心网设 备;或者,第一节点是核心网设备,且第二节点是另一核心网设备等。为了便于说明,本申请实施例中以第一节点为终端,第二节点为接入网设备为例进行举例说明,在此不构成具体限定。
上述第一AI网络模型可以是编码AI网络模型,该编码AI网络模型用于对第一节点需要传输的第一信息进行压缩和/或编码处理,得到第一特征信息,并对该第一特征信息进行传输,这样,传输第一特征信息相较于传输第一信息而言,能够减少传递开销,并提升所传递信息的安全性。
在实施中,第二节点可以使用解码AI网络模型(以下实施例中将第二节点具有的解码AI网络模型称之为第四AI网络模型)对特征信息进行解压和/或解码处理,以恢复所述第一信息,但是,第一节点使用的编码AI网络模型与第二节点使用的解码AI网络模型不是联合训练得到的,因此,第一节点使用的编码AI网络模型与第二节点使用的解码AI网络模型可能并不匹配,此时,需要将第一节点使用的编码AI网络模型所输出的第一特征信息进行转换处理,以得到能够被第二节点使用的解码AI网络模型使用或恢复的第二特征信息。
在一种可能的实现方式中,可以在第一节点将第一特征信息转换为第二特征信息,并传输该第二特征信息至第二节点,这样,第二节点可以直接将第二特征信息输入至自身具有的解码AI网络模型,以恢复上述第一信息。
在一种特殊的实现方式下,可以将上述第一AI网络模型与转换AI网络模型融合成一个AI网络模型,例如:联合训练第一AI网络模型与转换AI网络模型,从而使得融合后的AI网络模型在输入第一信息时,能够输出第二特征信息,其中,转换AI网络模型能够将第一特征信息转换为第二特征信息。
当然,在实施中,第一节点还可以采用独立的转换AI网络模型,或者非AI的转换方式,将第一AI网络模型输出的第一特征信息转换为第二特征信息。
在另一种可能的实现方式中,第一节点可以直接将第一特征信息发送给第二节点,并由第二节点对该第一特征信息进行转换和恢复处理,这样,第二节点在将第一特征信息转换成第二特征信息后,可以将第二特征信息输入至自身具有的解码AI网络模型,以恢复上述第一信息。
在一种特殊的实现方式下,可以将上述第二节点的解码AI网络模型与转换AI网络模型融合成一个AI网络模型,例如:联合训练第四AI网络模型与转换AI网络模型,从而使得融合后的AI网络模型在输入第一特征信息时,能够输出第一信息,其中,转换AI网络模型能够将第一特征信息转换为第二特征信息。
当然,在实施中,第二节点还可以采用独立的转换AI网络模型,或者非AI的转换方式,将接收到的第一特征信息转换成与其具有的解码AI网络模型相匹配的第二特征信息。
可选地,在所述第一节点基于第一AI网络模型,将第一信息处理成第一特征信息之前,所述方法还包括:
所述第一节点训练得到所述第一AI网络模型。
本实施方式下,第一节点独立训练所述第一AI网络模型,例如:第一节点进行联合训练,以得到第一AI网络模型和与该第一AI网络模型匹配的解码AI网络模型(以下实施例中将与第一AI网络模型匹配的解码AI网络模型称之为第三AI网络模型),在使用中,第一节点可以仅保留第一AI网络模型,删除与该第一AI网络模型匹配的解码AI网络模型。
与之相似的,第二节点也可以独立训练解码AI网络模型,例如:第二节点进行联合训练,以得到编码AI网络模型和与该编码AI网络模型匹配的解码AI网络模型,以下实施例中将第二节点的编码AI网络模型称之为第二AI网络模型,将第二节点的解码AI网络模型称之为第四AI网络模型),在使用中,第二节点可以仅保留第四AI网络模型,删除与该第四AI网络模型匹配的编码AI网络模型。
例如:如图3所示,假设第一节点使用编码器1,第二节点使用解码器2,该解码器2与编码器2匹配,即编码器2输出的特征信息能够被解码器2恢复,但是编码器1输出的特征信息不能够被解码器2恢复,此时,可以利用转换模块对编码器1输出的特征信息进行转换处理,以得到与编码器2输出的特征信息相同的特征信息,其中,转换模块可以配置有用于将编码器1输出的特征信息转换为编码器2输出的特征信息的转换信息。
可选地,在第二节点是网络侧设备的情况下,该网络侧设备可以获取信道状态信息(Channel State Information,CSI)报告(report)触发的最接近的探测参考信号(Sounding Reference Signal,SRS)的上行信道估计结果作为上行信道矩阵,并基于该上行信道矩阵来训练解码AI网络模型,这样,网络侧设备可以获取同一个小区内的至少两个终端的上行信道矩阵,从而使训练得到的解码AI网络模型适用于该小区内的至少两个或全部终端的上行信道,基于上行信道与下行信道的互异性,该解码AI网络模型同样适用于该小区内的至少两个或全部终端的下行信道。
换而言之,网络侧设备可以采用同一个解码AI网络模型对同一小区内的不同终端上报的信道特征信息进行解码,但是,每一个终端可以对应不同的转换信息。例如:假设同一基站的小区内包括终端A和终端B,且终端A基于编码AI网络模型a对信道矩阵进行压缩编码处理,得到第一信道特征信息a,终端B基于编码AI网络模型b对信道矩阵进行压缩编码处理,得到第一信道特征信息b,则可以利用转换AI网络模型a对第一信道特征信息a进行转换处理,得到基站的解码AI网络模型能够处理的第二信道特征信息,以及利用转换AI网络模型b对第一信道特征信息b进行转换处理,得到基站的解码AI网络模型能够处理的第二信道特征信息。
当然,在实施中,第一节点还可能采用其他方式来获取所述第一AI网络模型,例如:获取核心网设备下发的第一AI网络模型等。与之相似的,第二节点的解码AI网络模型,可以由第二节点独立训练得到,或者,第二节点也可以从核心网设备获取解码AI网络模型,但是,该第二节点获取的解码AI网络模型可以与第一节点获取的编码AI网络模型不 匹配,例如:核心网设备可以按照第一节点的能力,给第一节点下发编码AI网络模型,并按照第二节点的能力,给第二节点下发解码AI网络模型,此时,第一节点和第二节点都不知道对方使用的AI网络模型。
作为一种可选的实施方式,在所述第一节点向第二节点发送第一信息之前,所述方法还包括:
所述第一节点获取转换信息,其中,所述转换信息用于将所述第一AI网络模型输出的特征信息转换成所述第二节点的AI网络模型匹配的特征信息;
所述第一节点根据所述转换信息将所述第一特征信息转换成所述第二特征信息。
其中,所述转换信息可以用于将第一特征信息转换为第二特征信息,例如,该转换信息可以包括以下至少一项:
第一指示信息,用于指示单位阵;
第二指示信息,用于指示稀疏矩阵;
第三指示信息,用于指示转换AI网络模型的相关信息,所述转换AI网络模型用于将所述第一特征信息转换成所述第二特征信息。
在一种可能的实现方式中,在第一特征信息与第二特征信息是相同比特(bit)的特征矩阵时,可以通过单位阵或稀疏矩阵来调整第一特征信息中的“1”和/或“0”的位置,以得到第二特征信息。
在另一种可能的实现方式中,可以采用AI网络模型来实现第一特征信息至第二特征信息的转换,该AI网络模型即称为转换AI网络模型。
值得提出的是,训练转换AI网络模型所需的迭代次数少于训练编码和/或解码AI网络模型的迭代次数,也就是说训练转换AI网络模型所需的训练样本数据较少,且训练复杂程度较低。此外,转换AI网络模型的模型尺寸较小,传递该转换AI网络模型时所需的开销较小,且传递转换AI网络模型并不涉及各个节点的编码和解码实现过程,即使在第一节点、第二节点甚至第三节点之间进行传递,也不会降低第一节点和第二节点之间的保密性。
本实施方式中,由第一节点根据获取的转换信息来实现第一特征信息至第二特征信息之间的转化。
在实施中上述转换信息可以由第一节点确定,或者第一节点从其他节点(第二节点或第三节点)接收该转换信息。
场景一
第一节点接收来自第三节点的所述转换信息,其中,第三节点可以是除了第一节点和第二节点之外的设备,例如:第三节点是给第一节点和第二节点下发AI网络模型的设备,或者是第一节点和第二节点共同信任的设备,这样该第三节点可以知道第一节点使用的AI网络模型的相关信息以及第二节点使用的AI网络模型的相关信息。
例如:第三节点可以知道第一节点和第二节点分别使用的AI网络模型的结构、参数、 网络模型文件、AI网络模型处理后的特征信息等,并据此确定如何转换第一特征信息和第二特征信息。
以转换信息包括转换AI网络模型为例,本实施方式下,可以由第三节点来训练得到该转换AI网络模型。其中,训练该转换AI网络模型所需的训练样数据可以来自第一节点、第二节点、专用于收集训练样本数据的节点以及不具有所述第一AI网络模型和所述第二AI网络模型的节点中的至少一项。
例如:第一节点向第三节点发送第一AI网络模型,第二节点向第三节点发送第二AI网络模型,这样,第三节点可以将已经获取的训练样数据分别输入第一AI网络模型和第二AI网络模型,并对第一AI网络模型和第二AI网络模型对同一训练样数据的处理结果进行分析,得到转换AI网络模型。假设第一AI网络模型对训练样数据的处理结果为第三特征信息,第二AI网络模型对训练样数据的处理结果为第四特征信息,则该转换AI网络模型输入第三特征信息时,能够输出与该第三特征信息对应同一训练样本数据的第四特征信息。
当然,在实施中,第一节点可能直接向第三节点发送第三特征信息,第二节点也可能直接向第三节点发送第四特征信息,或者第一节点和第二节点中的一个向第三节点发送训练样本数据和对该训练样本数据处理后的特征信息,第一节点和第二节点中的另一个向第三节点发送编码AI网络模型,其同样能够使第三节点获取第三特征信息和第四特征信息,其中,第三特征信息与由同一个训练样本数据处理得到的第四特征一一对应。这样,第三节点在训练转换AI网络模型时,能够将第三特征信息作为该转换AI网络模型的输入,并将该第三特征信息对应的第四特征信息作为转换AI网络模型的输出目标。
场景二
第一节点基于与第一AI网络模型和第二AI网络模型相关的辅助信息,确定所述转换信息。
以转换信息包括转换AI网络模型为例,本实施方式中,可以由第一节点来训练得到该转换AI网络模型。
在一种可能的实现方式中,第一节点也可以按照上述第三节点训练转换AI网络模型相同的过程来根据第三特征信息和第四特征信息训练转换AI网络模型。
在另一种可能的实现方式中,第一节点可以基于第四特征信息和训练样本数据对第一AI网络模型和转换AI网络模型进行联合训练,其中,联合训练过程中,AI网络模型的输入信息为训练样本数据,输出目标为第四特征信息。
可选地,所述第一节点训练得到所述转换AI网络模型,包括:
所述第一节点获取第三特征信息以及接收来自所述第二节点的第四特征信息,其中,所述第三特征信息为基于所述第一AI网络模型对训练样本数据进行处理后得到的特征信息,所述第四特征信息为基于所述第二节点具有的第二AI网络模型对所述训练样本数据进行处理后得到的特征信息;
所述第一节点根据所述第三特征信息和所述第四特征信息,训练得到所述转换AI网络模型,其中,所述第三特征信息作为所述转换AI网络模型的输入,所述第四特征信息作为所述转换AI网络模型的输出目标。
其中,上述第三特征信息可以是第一节点将训练样本数据输入第一AI网络模型时的第一AI网络模型的输出信息。
值得提出的是,第一节点在训练转换AI网络模型时还需要获取第四特征信息,该第四特征信息可以是来自第二节点的,例如:第二节点与第一节点提前获知了训练样本数据以及训练样本数据的排列顺序或标识,第二节点基于自身的编码AI网络模型将训练样本数据处理成第四特征信息,并向第一节点发送该第四特征信息,如按照上述训练样本数据的排列顺序向第一节点发送各个训练样本数据对应的第四特征信息,或者在向第一节点发送第四特征信息时,指示各个第四特征信息对应的训练样本数据或标识。这样,在第一节点可以基于第一AI网络模型来将上述训练样本数据处理成第三特征信息后,可以基于同一个训练样本数据、排列于同一位置的训练样本数据、具有相同标识的训练样本数据来确定第三特征信息与第四特征信息之间的对应关系。
当然,第一节点也可以提前获知第二节点的编码AI网络模型,并基于该第二节点的编码AI网络模型将训练样本数据处理成第四特征信息,在此对第一节点获取第四特征信息的方式不作具体限定。
场景三
第一节点接收来自第二节点的所述转换信息,其中,第二节点可以基于与第一AI网络模型和第二AI网络模型相关的辅助信息,确定所述转换信息;
以转换信息包括转换AI网络模型为例,本场景下,可以由第二节点来训练得到该转换AI网络模型。
需要说明的是,本实施方式中第二节点训练转换AI网络模型的过程与上述场景二中第一节点训练AI网络模型的过程相似,不同之处在于,本实施方式中,第二节点在训练转换AI网络模型时,需要从第一节点获取第三特征信息、训练样本数据、第一AI网络模型中的至少一项,第二节点训练转换AI网络模型的具体过程,可以参考场景二中,第一节点训练转换AI网络模型的过程,在此不再赘述。
可选地,在由第二节点或第三节点训练上述转换AI网络模型的情况下,第一节点需要向训练转换AI网络模型的节点提供辅助信息,该辅助信息用于辅助训练转换AI网络模型的节点获取第三特征信息。
例如:在所述第一节点接收来自第三节点或所述第二节点的所述第三指示信息之前,所述方法还包括:
所述第一节点向所述第三节点或所述第二节点发送第三信息,其中,所述第三信息包括所述第一AI网络模型的相关信息,或者,所述第三信息包括训练样本数据和第三特征信息,或者,所述第三信息包括所述第三特征信息;
其中,所述第三特征信息为基于所述第一AI网络模型对所述训练样本数据进行处理后得到的特征信息,所述转换AI网络模型基于所述第三特征信息和第四特征信息中的至少一项训练得到,所述第四特征信息为基于所述第二节点具有的第二AI网络模型对所述训练样本数据进行处理后得到的特征信息。
其中,所述第一AI网络模型的相关信息可以包括第一AI网络模型的模型结构信息、模型参数信息和模型文件中的至少一项,该训练转换AI网络模型的节点能够根据该第一AI网络模型的相关信息实现第一AI网络模型的模型推理,例如:基于第一AI网络模型的相关信息实现基于第一AI网络模型对训练样本数据进行处理,得到第三特征信息。
此外,在训练转换AI网络模型过程中的训练样本数据,可能是第三节点预先存储的,或者是来自第四节点的,其中,第四节点可以包括以下至少一项:
所述第一节点;
所述第二节点;
专用于收集训练样本数据的节点;
不具有所述第一AI网络模型和所述第二AI网络模型的节点。
其中,在训练样本数据来自第一节点的情况下,该训练样本数据可以是第一节点检测到的信道信息。
可选地,所述特征信息传输方法还包括:
所述第一节点估计第一信道的第一信道信息,其中,所述训练样本数据包括所述第一信道信息。
本实施方式中,可以基于第一节点估计到的实际信道的信道信息来训练转换AI网络模型,这样,可以提升转换AI网络模型与第一节点的实际信道状态的匹配程度。
需要说明的是,在第三节点或第二节点训练转换AI网络模型的情况下,如果第一节点只向训练转换AI网络模型的节点发送了第三特征信息,此时的训练样本数据的集合可以是固定的,且第一节点和训练转换AI网络模型的节点都已知该训练样本数据集合,例如:通过协议约定、定期更新的方式设置一个固定的训练样本数据集,该训练样本数据集中的训练样本数据的顺序是固定的,如:第一节点上报的第一个第三特征信息对应训练样本数据集中的第一个训练样本数据,并以此类推。这样,训练转换AI网络模型的节点能够根据第三特征信息对应的训练样本数据来训练转换AI网络模型,例如:根据第三特征信息与训练样本数据之间的对应关系,以及第四特征信息与训练样本数据之间的对应关系,将与同一个训练样本数据对应的第三特征信息和第四特征信息相互关联,以将相互关联的第三特征信息和第四特征信息分别作为训练转换AI网络模型的输入信息和标签。
当然,在实施中,第一节点还可以在上报第三特征信息的时候,同时上报该第三特征信息对应的训练样本数据的标识(Identifier,ID),该ID可以表示使用的训练样本数据在训练样本数据集合中的ID。这样,训练转换AI网络模型的节点能够根据这个ID寻找对应的训练样本数据,从而完成转换AI网络模型的训练。
相似的,在训练样本数据来自第二节点的情况下,该训练样本数据也可以是第二节点检测到的信道信息。例如:假设第一节点是终端,第二节点是基站,则训练样本数据可以包括终端估计的下行信道的信道信息,或者,训练样本数据可以包括基站估计的上行信道的信道信息,基于上行信道与下行信道的互异性,上行信道的信道信息与对应的下行信道的信道信息接近,从而使基于该上行信道的信道信息训练得到的转换AI网络模型与终端的下行信道状态相匹配。
当然,训练转换AI网络模型的节点还可以通过专用于收集训练样本数据的节点或者不具有所述第一AI网络模型和所述第二AI网络模型的节点获取训练样本数据,可以降低采集训练样本数据的复杂程度。
需要注意的是,在所述第一节点向转换AI网络模型的节点发送第三特征信息的情况下,转换AI网络模型的节点可以获取该第三特征信息对应的训练样本数据或与该第三特征信息对应相同的训练样本数据的第四特征信息。例如:假设由第三节点训练转换AI网络模型,则第三节点可以从第一节点获取第三特征信息,并从第二节点获取第四特征信息,该第三特征信息与第四特征信息基于相同的训练样本数据得到;或者,假设由第二节点联合训练转换AI网络模型和第四AI网络模型,则第二AI节点可以获取第三特征信息和训练样本数据,并据此联合训练能够将第三特征信息恢复成训练样本数据的AI网络模型。
在本申请实施例中,第一节点基于第一AI网络模型,将第一信息处理成第一特征信息;所述第一节点向第二节点发送第二信息,所述第二信息包括所述第一特征信息或第二特征信息,其中,所述第二特征信息为所述第一特征信息转换后的特征信息。在第一节点和第二节点之间通过AI网络模型来压缩和恢复第一信息时,基于第一节点与第二节点使用的AI网络模型可能不匹配,从而造成第一节点压缩得到的第一特征信息不能够被第二节点使用或恢复时,第一节点可以对第一特征信息进行转换,以向第二节点发送其能够使用或恢复的第二特征信息,或者,第一节点也可以向第二节点发送第一特征信息,并由第二节点将第一特征信息转化为该第二节点具有的AI网络模型能够使用或恢复的第二特征信息后,利用AI网络模型对该第二特征信息进行恢复处理。这样,第一节点和第二节点能够独立使用或训练各自的AI网络模型,从而能够降低因传输AI网络模型而造成的开销,同时,第一节点与第二节点之间无需交互各自使用的AI网络模型,从而提升了第一节点与第二节点的信息安全性。
请参阅图4,本申请实施例提供的另一种特征信息传输方法,其执行主体是第二节点,如图4所示,该第二节点执行的特征信息传输方法可以包括以下步骤:
步骤401、第二节点接收来自第一节点的第二信息,其中,所述第二信息包括第一特征信息或第二特征信息,其中,所述第一特征信息为基于所述第一节点具有的第一AI网络模型对第一信息进行处理后得到的特征信息,所述第二特征信息为所述第一特征信息转换后的特征信息。
步骤402、所述第二节点基于第四AI网络模型对所述第二特征信息进行恢复处理, 得到所述第一信息,其中,在所述第二信息包括所述第一特征信息的情况下,所述第二特征信息由所述第二节点对所述第一特征信息进行转换处理得到。
如图4所示方法实施例与如图2所示方法实施例相对应,不同之处在于,如图2所示方法实施例的执行主体是第一节点,而如图4所示方法实施例的执行主体是第二节点,且本实施例中的第一节点、第二节点、第二信息、第一特征信息、第二特征信息、第一AI网络模型、第四AI网络模型的具体含义可以参考如图2所示方法实施例中的解释说明,在此不再赘述。
作为一种可选的实施方式,在所述第二节点基于第四AI网络模型对所述第二特征信息进行恢复处理之前,所述特征信息传输方法还包括:
所述第二节点训练得到所述第四AI网络模型。
在实施中,第二节点可以联合训练第二AI网络模型和第四AI网络模型,并保留第四AI网络模型,用于对接收的特征信息进行解码处理;或者,第二节点可以独立训练第四AI网络模型;或者,第二节点可以联合训练转换AI网络模型和第四AI网络模型。
其中,第二AI网络模型和转换AI网络模型的含义可以参考如图2所示方法实施例中的解释说明,在此不再赘述。
可选地,所述第一AI网络模型与第三AI网络模型联合训练得到,所述第三AI网络模型与所述第一AI网络模型匹配,且用于对所述第一AI网络模型处理后的特征信息进行恢复处理;
和/或,
第二AI网络模型与所述第四AI网络模型联合训练得到,所述第二AI网络模型为所述第二节点具有的与所述第二特征信息匹配的AI网络模型,所述第四AI网络模型与所述第二AI网络模型匹配,且用于对所述第二AI网络模型处理后的特征信息进行恢复处理。
作为一种可选的实施方式,在所述第二节点基于第四AI网络模型对所述第二特征信息进行恢复处理之前,所述特征信息传输方法还包括:
所述第二节点获取转换信息,其中,所述转换信息用于将所述第一AI网络模型输出的特征信息转换成所述第二节点的AI网络模型匹配的特征信息;
所述第二节点根据所述转换信息将所述第一特征信息转换成所述第二特征信息。
本实施方式下,由第二节点实现第一特征信息与第二特征信息之间的转换。
作为一种可选的实施方式,所述转换信息包括以下至少一项:
第一指示信息,用于指示单位阵;
第二指示信息,用于指示稀疏矩阵;
第三指示信息,用于指示转换AI网络模型的相关信息,所述转换AI网络模型用于将所述第一特征信息转换成所述第二特征信息。
作为一种可选的实施方式,在所述转换信息包括所述第三指示信息的情况下,所述第二节点获取转换信息,包括:
所述第二节点训练得到所述转换AI网络模型;或者,
所述第二节点接收来自第三节点或所述第一节点的所述第三指示信息,其中,所述第三节点为除了所述第一节点和所述第二节点之外的设备。
作为一种可选的实施方式,所述第二节点训练得到所述转换AI网络模型,包括:
所述第二节点获取第四特征信息以及接收来自所述第一节点的第三特征信息,其中,所述第三特征信息为基于所述第一AI网络模型对训练样本数据进行处理后得到的特征信息,所述第四特征信息为基于所述第二节点具有的第二AI网络模型对所述训练样本数据进行处理后得到的特征信息;
所述第二节点根据所述第三特征信息和所述第四特征信息,训练得到所述转换AI网络模型,其中,所述第三特征信息作为所述转换AI网络模型的输入,所述第四特征信息作为所述转换AI网络模型的输出目标。
本实施方式下,由第二节点作为训练转换AI网络模型的节点,其中,训练转换AI网络模型所需的训练样本数据可以来自第一节点、第二节点和第三节点中的至少一项,且上述第二节点基于第三特征信息和第四特征信息训练转换AI网络模型的具体过程与如图2所示方法实施例中,第一节点基于第三特征信息和第四特征信息训练转换AI网络模型的具体过程相似,在此不再赘述。
作为一种可选的实施方式,在所述第二节点接收来自第三节点或所述第一节点的所述第三指示信息之前,所述特征信息传输方法还包括:
所述第二节点向所述第三节点或所述第一节点发送第四信息,其中,所述第四信息包括所述第二节点具有的第二AI网络模型的相关信息,或者,所述第四信息包括训练样本数据和第四特征信息,或者,所述第四信息包括所述第四特征信息;
其中,所述第四特征信息为基于所述第二AI网络模型对所述训练样本数据进行处理后得到的特征信息,所述转换AI网络模型基于第三特征信息和所述第四特征信息中的至少一项训练得到,所述第三特征信息为基于所述第一AI网络模型对所述训练样本数据进行处理后得到的特征信息。
本实施方式下,由第一节点或第三节点作为训练转换AI网络模型的节点,其具体过程可以参考如图2所示方法实施例中的相应说明,在此不再赘述。
作为一种可选的实施方式,所述特征信息传输方法还包括:
所述第二节点估计第二信道的第二信道信息,其中,所述训练样本数据包括所述第二信道信息。
作为一种可选的实施方式,在所述第二节点为网络侧设备的情况下,所述第二节点接收来自第一节点的第二信息,包括:
所述第二节点接收来自至少一个终端的目标信道状态信息CSI报告,所述第一特征信息包括所述目标CSI报告携带的目标信道的信道特征信息,其中,所述至少一个终端包括所述第一节点;
所述方法还包括:
所述第二节点获取与所述目标信道相关的SRS的第三信道信息;
所述第二节点根据所述第三信道信息训练所述第四AI网络模型。
其中,基于第三信道信息训练的第四AI网络模型,可以对终端上报的信道特征信息进行恢复处理,其中,各个终端使用各自的第一AI网络模型来对检测到的信道信息进行编码处理,得到信道特征信息。
本实施方式中,网络侧设备可以获取信道状态信息(Channel State Information,CSI)报告(report)触发的最接近的探测参考信号(Sounding Reference Signal,SRS)的上行信道估计结果作为上行信道矩阵,并基于该上行信道矩阵来训练解码AI网络模型,这样,网络侧设备可以获取同一个小区内的至少两个终端的上行信道矩阵,从而使训练得到的解码AI网络模型适用于该小区内的至少两个或全部终端的上行信道,基于上行信道与下行信道的互异性,该解码AI网络模型同样适用于该小区内的至少两个或全部终端的下行信道。
需要说明的是,虽然网络侧设备的同一个解码AI网络模型适用于不同终端上报的特征信,但是,仍然可以事先使用与各个终端一一对应的转换AI网络模型来对终端上报的特征信息进行转换处理,以得到该同一个解码AI网络模型能够处理的特征信息。
本申请实施例提供的由第二节点执行的特征信息传输方法与如图2所示方法实施例提供的由第一节点执行的特征信息传输方法相配合,能够共同实现使第一节点和第二节点能够独立使用或训练各自的AI网络模型,从而降低因传输AI网络模型而造成的开销的有益效果,同时,第一节点与第二节点之间无需交互各自使用的AI网络模型,从而提升了第一节点与第二节点的信息安全性。
本申请实施例提供的特征信息传输方法,执行主体可以为特征信息传输装置。本申请实施例中以特征信息传输装置执行特征信息传输方法为例,说明本申请实施例提供的特征信息传输装置。
请参阅图5,本申请实施例提供的一种特征信息传输装置,可以是第一节点内的装置,如图5所示,该特征信息传输装置500可以包括以下模块:
第一处理模块501,用于基于第一AI网络模型,将第一信息处理成第一特征信息;
第一发送模块502,用于向第二节点发送第二信息,所述第二信息包括所述第一特征信息或第二特征信息,其中,所述第二特征信息为所述第一特征信息转换后的特征信息。
可选地,特征信息传输装置500还包括:
第一训练模块,用于训练得到所述第一AI网络模型。
可选地,所述第一AI网络模型与第三AI网络模型联合训练得到,所述第三AI网络模型与所述第一AI网络模型匹配,且用于对所述第一AI网络模型处理后的特征信息进行恢复处理;
和/或,
第二AI网络模型与第四AI网络模型联合训练得到,所述第二AI网络模型为所述第二节点具有的与所述第二特征信息匹配的AI网络模型,所述第四AI网络模型与所述第二AI网络模型匹配,且用于对所述第二AI网络模型处理后的特征信息进行恢复处理。
可选地,特征信息传输装置500还包括:
第二获取模块,用于获取转换信息,其中,所述转换信息用于将所述第一AI网络模型输出的特征信息转换成所述第二节点的AI网络模型匹配的特征信息;
第一转换模块,用于根据所述转换信息将所述第一特征信息转换成所述第二特征信息。
可选地,所述转换信息包括以下至少一项:
第一指示信息,用于指示单位阵;
第二指示信息,用于指示稀疏矩阵;
第三指示信息,用于指示转换AI网络模型的相关信息,所述转换AI网络模型用于将所述第一特征信息转换成所述第二特征信息。
可选地,在所述转换信息包括所述第三指示信息的情况下,所述第二获取模块用于:
训练得到所述转换AI网络模型;或者,
接收来自第三节点或所述第二节点的所述第三指示信息,其中,所述第三节点为除了所述第一节点和所述第二节点之外的设备。
可选地,所述第二获取模块,包括:
第一获取单元,用于获取第三特征信息以及接收来自所述第二节点的第四特征信息,其中,所述第三特征信息为基于所述第一AI网络模型对训练样本数据进行处理后得到的特征信息,所述第四特征信息为基于所述第二节点具有的第二AI网络模型对所述训练样本数据进行处理后得到的特征信息;
第一训练单元,用于根据所述第三特征信息和所述第四特征信息,训练得到所述转换AI网络模型,其中,所述第三特征信息作为所述转换AI网络模型的输入,所述第四特征信息作为所述转换AI网络模型的输出目标。
可选地,特征信息传输装置500还包括:
第三发送模块,用于向所述第三节点或所述第二节点发送第三信息,其中,所述第三信息包括所述第一AI网络模型的相关信息,或者,所述第三信息包括训练样本数据和第三特征信息,或者,所述第三信息包括所述第三特征信息;
其中,所述第三特征信息为基于所述第一AI网络模型对所述训练样本数据进行处理后得到的特征信息,所述转换AI网络模型基于所述第三特征信息和第四特征信息中的至少一项训练得到,所述第四特征信息为基于所述第二节点具有的第二AI网络模型对所述训练样本数据进行处理后得到的特征信息。
可选地,特征信息传输装置500还包括:
第一信道估计模块,用于估计第一信道的第一信道信息,其中,所述训练样本数据包括所述第一信道信息。
本申请实施例提供的特征信息传输装置500,能够实现如图2所示方法实施例中第一节点实现的各个过程,且能够取得相同的有益效果,为避免重复,在此不再赘述。
请参阅图6,本申请实施例提供的另一种特征信息传输装置,可以是第二节点内的装置,如图6所示,该特征信息传输装置600可以包括以下模块:
第一接收模块601,用于接收来自第一节点的第二信息,其中,所述第二信息包括第一特征信息或第二特征信息,其中,所述第一特征信息为基于所述第一节点具有的第一AI网络模型对第一信息进行处理后得到的特征信息,所述第二特征信息为所述第一特征信息转换后的特征信息;
第二处理模块602,用于基于第四AI网络模型对所述第二特征信息进行恢复处理,得到所述第一信息,其中,在所述第二信息包括所述第一特征信息的情况下,所述第二特征信息由所述第二节点对所述第一特征信息进行转换处理得到。
可选地,特征信息传输装置600还包括:
第二训练模块,用于训练得到所述第四AI网络模型。
可选地,所述第一AI网络模型与第三AI网络模型联合训练得到,所述第三AI网络模型与所述第一AI网络模型匹配,且用于对所述第一AI网络模型处理后的特征信息进行恢复处理;
和/或,
第二AI网络模型与所述第四AI网络模型联合训练得到,所述第二AI网络模型为所述第二节点具有的与所述第二特征信息匹配的AI网络模型,所述第四AI网络模型与所述第二AI网络模型匹配,且用于对所述第二AI网络模型处理后的特征信息进行恢复处理。
可选地,特征信息传输装置600还包括:
第三获取模块,用于获取转换信息,其中,所述转换信息用于将所述第一AI网络模型输出的特征信息转换成所述第二节点的AI网络模型匹配的特征信息;
第二转换模块,用于根据所述转换信息将所述第一特征信息转换成所述第二特征信息。
可选地,所述转换信息包括以下至少一项:
第一指示信息,用于指示单位阵;
第二指示信息,用于指示稀疏矩阵;
第三指示信息,用于指示转换AI网络模型的相关信息,所述转换AI网络模型用于将所述第一特征信息转换成所述第二特征信息。
可选地,在所述转换信息包括所述第三指示信息的情况下,所述第三获取模块用于:
训练得到所述转换AI网络模型;或者,
接收来自第三节点或所述第一节点的所述第三指示信息,其中,所述第三节点为除了所述第一节点和所述第二节点之外的设备。
可选地,所述第三获取模块,包括:
第二获取单元,用于获取第四特征信息以及接收来自所述第一节点的第三特征信息, 其中,所述第三特征信息为基于所述第一AI网络模型对训练样本数据进行处理后得到的特征信息,所述第四特征信息为基于所述第二节点具有的第二AI网络模型对所述训练样本数据进行处理后得到的特征信息;
第二训练单元,用于根据所述第三特征信息和所述第四特征信息,训练得到所述转换AI网络模型,其中,所述第三特征信息作为所述转换AI网络模型的输入,所述第四特征信息作为所述转换AI网络模型的输出目标。
可选地,特征信息传输装置600还包括:
第四发送模块,用于向所述第三节点或所述第一节点发送第四信息,其中,所述第四信息包括所述第二节点具有的第二AI网络模型的相关信息,或者,所述第四信息包括训练样本数据和第四特征信息,或者,所述第四信息包括所述第四特征信息;
其中,所述第四特征信息为基于所述第二AI网络模型对所述训练样本数据进行处理后得到的特征信息,所述转换AI网络模型基于第三特征信息和所述第四特征信息中的至少一项训练得到,所述第三特征信息为基于所述第一AI网络模型对所述训练样本数据进行处理后得到的特征信息。
可选地,特征信息传输装置600还包括:
第二信道估计模块,用于估计第二信道的第二信道信息,其中,所述训练样本数据包括所述第二信道信息。
可选地,在所述第二节点为网络侧设备的情况下,第一接收模块601用于:
接收来自至少一个终端的目标信道状态信息CSI报告,所述第一特征信息包括所述目标CSI报告携带的目标信道的信道特征信息,其中,所述至少一个终端包括所述第一节点;
特征信息传输装置600还包括:
第四获取模块,用于获取与所述目标信道相关的SRS的第三信道信息;
第三训练模块,用于根据所述第三信道信息训练所述第四AI网络模型。
本申请实施例提供的特征信息传输装置600,能够实现如图4所示方法实施例中第二节点实现的各个过程,且能够取得相同的有益效果,为避免重复,在此不再赘述。
本申请实施例中的特征信息传输装置可以是电子设备,例如具有操作系统的电子设备,也可以是电子设备中的部件,例如集成电路或芯片。该电子设备可以是终端或网络侧设备,也可以为除终端和网络侧设备之外的其他设备。示例性的,终端可以包括但不限于上述所列举的终端11的类型,网络侧设备可以包括但不限于上述所列举的网络侧设备12的类型。其他设备可以为服务器、网络附属存储器(Network Attached Storage,NAS)等,本申请实施例不作具体限定。
本申请实施例提供的特征信息传输装置能够实现图2或图4所示方法实施例实现的各个过程,并达到相同的技术效果,为避免重复,这里不再赘述。
请参阅图7,本申请实施例提供的一种转换信息确定方法,其执行主体是第三节点,如图7所示,该第三节点执行的转换信息确定方法可以包括以下步骤:
步骤701、第三节点获取第一AI网络模型对训练样本数据处理后得到的第三特征信息,以及获取第二AI网络模型对所述训练样本数据处理后得到的第四特征信息,其中,所述第一AI网络模型为第一节点具有的AI网络模型,所述第二AI网络模型为第二节点具有的AI网络模型。
步骤702、所述第三节点根据所述第三特征信息和所述第四特征信息,确定转换信息,其中,所述转换信息用于将目标训练样本对应的第三特征信息转换为与所述目标训练样本对应的第四特征信息,所述目标训练样本为所述训练样本数据中的任一样本。
步骤703、所述第三节点向所述第一节点和所述第二节点中的至少一个发送所述转换信息。
如图7所示方法实施例中的第一节点、第二节点、第三节点、第一AI网络模型、第二AI网络模型、第三特征信息、第四特征信息、转换信息、目标训练样本的含义和作用可以参考如图2所示方法实施例中的解释说明。
本申请实施例中,主要针对第三节点确定转换信息的实施方式,例如:由第三节点更新第三特征信息和第四特征信息来训练转换AI网络模型,并发送给第一节点或第二节点,,其中,第三节点确定转换信息的过程可以参考如图2所示方法实施例中,第三节点确定转换信息的解释说明,在此不再赘述。
作为一种可选的实施方式,所述转换信息包括以下至少一项:
第一指示信息,用于指示单位阵;
第二指示信息,用于指示稀疏矩阵;
第三指示信息,用于指示转换AI网络模型的相关信息,所述转换AI网络模型用于将第一特征信息转换成第二特征信息,其中,所述第一特征信息与所述第一AI网络模型匹配,所述第二特征信息与所述第二AI网络模型匹配。
作为一种可选的实施方式,所述第三节点获取第一AI网络模型对训练样本数据处理后得到的第三特征信息,包括:
所述第三节点接收来自所述第一节点的第五信息,所述第五信息包括所述第一AI网络模型的相关信息;
所述第三节点将训练样本数据输入所述第一AI网络模型,并获取所述第一AI网络模型输出的所述第三特征信息;
和/或,
所述第三节点获取第二AI网络模型对所述训练样本数据处理后得到的第四特征信息,包括:
所述第三节点接收来自所述第二节点的第六信息,所述第六信息包括所述第二AI网络模型的相关信息;
所述第三节点将训练样本数据输入所述第二AI网络模型,并获取所述第二AI网络模型输出的所述第四特征信息。
本实施方式下,第三节点可以获取第一节点和第二节点各自的编码AI网络模型,并基于第一节点和第二节点的编码AI网络模型分别对训练样本数据进行处理,以得到确定转换信息所需的第三特征信息和第四特征信息。
作为一种可选的实施方式,所述训练样本数据预先存储于所述第三节点,或者,所述训练样本数据来自第四节点,所述第四节点包括以下至少一项:
所述第一节点;
所述第二节点;
专用于收集训练样本数据的节点;
不具有所述第一AI网络模型和所述第二AI网络模型的节点。
作为一种可选的实施方式,所述第三节点获取第一AI网络模型对训练样本数据处理后得到的第三特征信息,包括:
所述第三节点接收来自所述第一节点的训练样本数据和所述第三特征信息,其中,所述第三特征信息为将所述训练样本数据输入所述第一AI网络模型后,由所述第一AI网络模型输出的特征信息;
所述第三节点获取第二AI网络模型对训练样本数据处理后得到的第四特征信息,包括:
所述第三节点将所述训练样本数据输入所述第二AI网络模型,并获取所述第二AI网络模型输出的所述第四特征信息。
作为一种可选的实施方式,所述第三节点获取第二AI网络模型对训练样本数据处理后得到的第四特征信息,包括:
所述第三节点接收来自所述第二节点的训练样本数据和所述第四特征信息,其中,所述第四特征信息为将所述训练样本数据输入所述第二AI网络模型后,由所述第二AI网络模型输出的特征信息;
所述第三节点获取第一AI网络模型对训练样本数据处理后得到的第三特征信息,包括:
所述第三节点将所述训练样本数据输入所述第一AI网络模型,并获取所述第一AI网络模型输出的所述第三特征信息。
例如:假设第一节点是终端,第二节点是基站,第三节点可以是该终端和该基站共同信任的设备。此时,第三节点训练AI网络模型的方案可以是:
终端和基站将各自的编码AI网络模型的相关信息发送给第三节点,第三节点使用自身保存的信道信息分别基于终端和基站的编码AI网络模型获得编码后的第三特征信息和第四特征信息,将第三特征信息作为输入,将第四特征信息所谓标签,对转换AI网络模型进行训练;或者,
终端将实际估计的信道信息和这个信道信息通过终端的编码AI网络模型处理后的第三特征信息发送给第三节点,基站将自身的编码AI网络模型的相关信息发送给第三节点, 第三节点通过基站的编码AI网络模型对终端上报的信道信息进行处理,得到第四特征信息,并将第三特征信息作为输入,将第四特征信息所谓标签,对转换AI网络模型进行训练。
本申请实施例中,第三节点可以获取第一节点的第一AI网络模型对训练样本数据进行处理后得到的第三特征信息,并获取第二节点的第二AI网络模型对训练样本数据进行处理后得到的第四特征信息,并根据两者的差异来确定转换信息,以基于该转换信息,能够将第一AI网络模型的输出结果转换成第二AI网络模型的输出结果。
本申请实施例提供的特征信息传输方法,执行主体可以为特征信息传输装置。本申请实施例中以特征信息传输装置执行特征信息传输方法为例,说明本申请实施例提供的特征信息传输装置。
请参阅图8,本申请实施例提供的一种转换信息确定装置,可以是第三节点内的装置,如图8所示,该转换信息确定装置800可以包括以下模块:
第一获取模块801,用于获取第一AI网络模型对训练样本数据处理后得到的第三特征信息,以及获取第二AI网络模型对所述训练样本数据处理后得到的第四特征信息,其中,所述第一AI网络模型为第一节点具有的AI网络模型,所述第二AI网络模型为第二节点具有的AI网络模型;
第一确定模块802,用于根据所述第三特征信息和所述第四特征信息,确定转换信息,其中,所述转换信息用于将目标训练样本对应的第三特征信息转换为与所述目标训练样本对应的第四特征信息,所述目标训练样本为所述训练样本数据中的任一样本;
第二发送模块803,用于向所述第一节点和所述第二节点中的至少一个发送所述转换信息。
可选地,所述转换信息包括以下至少一项:
第一指示信息,用于指示单位阵;
第二指示信息,用于指示稀疏矩阵;
第三指示信息,用于指示转换AI网络模型的相关信息,所述转换AI网络模型用于将第一特征信息转换成第二特征信息,其中,所述第一特征信息与所述第一AI网络模型匹配,所述第二特征信息与所述第二AI网络模型匹配。
可选地,第一获取模块801,包括:
第一接收单元,用于接收来自所述第一节点的第五信息,所述第五信息包括所述第一AI网络模型的相关信息;
第一处理单元,用于将训练样本数据输入所述第一AI网络模型,并获取所述第一AI网络模型输出的所述第三特征信息;
和/或,
第一获取模块801,还包括:
第二接收单元,用于接收来自所述第二节点的第六信息,所述第六信息包括所述第二 AI网络模型的相关信息;
第二处理单元,用于将训练样本数据输入所述第二AI网络模型,并获取所述第二AI网络模型输出的所述第四特征信息。
可选地,所述训练样本数据预先存储于所述第三节点,或者,所述训练样本数据来自第四节点,所述第四节点包括以下至少一项:
所述第一节点;
所述第二节点;
专用于收集训练样本数据的节点;
不具有所述第一AI网络模型和所述第二AI网络模型的节点。
可选地,第一获取模块801,包括:
第三接收单元,用于接收来自所述第一节点的训练样本数据和所述第三特征信息,其中,所述第三特征信息为将所述训练样本数据输入所述第一AI网络模型后,由所述第一AI网络模型输出的特征信息;
第三处理单元,用于将所述训练样本数据输入所述第二AI网络模型,并获取所述第二AI网络模型输出的所述第四特征信息。
可选地,第一获取模块801,包括:
第四接收单元,用于接收来自所述第二节点的训练样本数据和所述第四特征信息,其中,所述第四特征信息为将所述训练样本数据输入所述第二AI网络模型后,由所述第二AI网络模型输出的特征信息;
第四处理单元,用于将所述训练样本数据输入所述第一AI网络模型,并获取所述第一AI网络模型输出的所述第三特征信息。
本申请实施例提供的转换信息确定装置800,能够实现如图7所示方法实施例中第三节点实现的各个过程,且能够取得相同的有益效果,为避免重复,在此不再赘述。
可选地,如图9所示,本申请实施例还提供一种通信设备900,包括处理器901和存储器902,存储器902上存储有可在所述处理器901上运行的程序或指令,例如,该通信设备900作为第一节点时,该程序或指令被处理器901执行时实现如图2所示方法实施例的各个步骤,且能达到相同的技术效果。该通信设备900作为第二节点时,该程序或指令被处理器901执行时实现如图4所示方法实施例的各个步骤,且能达到相同的技术效果。该通信设备900作为第三节点时,该程序或指令被处理器901执行时实现如图7所示方法实施例的各个步骤,且能达到相同的技术效果,为避免重复,这里不再赘述。
本申请实施例还提供一种通信设备,包括处理器和通信接口。
在一种实施方式中,在该通信设备作为第一节点时,所述处理器用于基于第一AI网络模型,将第一信息处理成第一特征信息;所述通信接口用于向第二节点发送第二信息,所述第二信息包括所述第一特征信息或第二特征信息,其中,所述第二特征信息为所述第一特征信息转换后的特征信息;或者,
在另一种实施方式中,在该通信设备作为第二节点时,所述通信接口用于接收来自第一节点的第二信息,其中,所述第二信息包括第一特征信息或第二特征信息,其中,所述第一特征信息为基于所述第一节点具有的第一AI网络模型对第一信息进行处理后得到的特征信息,所述第二特征信息为所述第一特征信息转换后的特征信息;所述处理器用于基于第四AI网络模型对所述第二特征信息进行恢复处理,得到所述第一信息,其中,在所述第二信息包括所述第一特征信息的情况下,所述第二特征信息由所述第二节点对所述第一特征信息进行转换处理得到。
在另一种实施方式中,在该通信设备作为第三节点时,所述通信接口用于获取第一AI网络模型对训练样本数据处理后得到的第三特征信息,以及获取第二AI网络模型对所述训练样本数据处理后得到的第四特征信息,其中,所述第一AI网络模型为第一节点具有的AI网络模型,所述第二AI网络模型为第二节点具有的AI网络模型;所述处理器用于根据所述第三特征信息和所述第四特征信息,确定转换信息,其中,所述转换信息用于将目标训练样本对应的第三特征信息转换为与所述目标训练样本对应的第四特征信息,所述目标训练样本为所述训练样本数据中的任一样本;所述通信接口还用于向所述第一节点和所述第二节点中的至少一个发送所述转换信息。
该通信设备实施例与如图2或图4或图7所示方法实施例对应,图2或图4或图7所示方法实施例的各个实施过程和实现方式均可适用于该通信设备实施例中,且能达到相同的技术效果。
本申请实施例还提供一种可读存储介质,所述可读存储介质上存储有程序或指令,该程序或指令被处理器执行时实现如图2或图4或图7所示方法实施例的各个过程,且能达到相同的技术效果,为避免重复,这里不再赘述。
其中,所述处理器为上述实施例中所述的终端中的处理器。所述可读存储介质,包括计算机可读存储介质,如计算机只读存储器ROM、随机存取存储器RAM、磁碟或者光盘等。
本申请实施例另提供了一种芯片,所述芯片包括处理器和通信接口,所述通信接口和所述处理器耦合,所述处理器用于运行程序或指令,实现如图2或图4或图7所示方法实施例的各个过程,且能达到相同的技术效果,为避免重复,这里不再赘述。
应理解,本申请实施例提到的芯片还可以称为系统级芯片,系统芯片,芯片系统或片上系统芯片等。
本申请实施例另提供了一种计算机程序/程序产品,所述计算机程序/程序产品被存储在存储介质中,所述计算机程序/程序产品被至少一个处理器执行以实现如图2或图4或图7所示方法实施例的各个过程,且能达到相同的技术效果,为避免重复,这里不再赘述。
本申请实施例还提供了一种通信系统,包括:第一节点、第二节点和第三节点,所述第一节点可用于执行如图2所示的特征信息传输方法的步骤,所述第二节点可用于执行如图4所示的特征信息传输方法的步骤,所述第三节点可用于执行如图7所示的转换信息确 定方法的步骤。
需要说明的是,在本文中,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、物品或者装置不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、物品或者装置所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、方法、物品或者装置中还存在另外的相同要素。此外,需要指出的是,本申请实施方式中的方法和装置的范围不限按示出或讨论的顺序来执行功能,还可包括根据所涉及的功能按基本同时的方式或按相反的顺序来执行功能,例如,可以按不同于所描述的次序来执行所描述的方法,并且还可以添加、省去、或组合各种步骤。另外,参照某些示例所描述的特征可在其他示例中被组合。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以计算机软件产品的形式体现出来,该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中,包括若干指令用以使得一台终端(可以是手机,计算机,服务器,空调器,或者网络设备等)执行本申请各个实施例所述的方法。
上面结合附图对本申请的实施例进行了描述,但是本申请并不局限于上述的具体实施方式,上述的具体实施方式仅仅是示意性的,而不是限制性的,本领域的普通技术人员在本申请的启示下,在不脱离本申请宗旨和权利要求所保护的范围情况下,还可做出很多形式,均属于本申请的保护之内。

Claims (30)

  1. 一种特征信息传输方法,包括:
    第一节点基于第一AI网络模型,将第一信息处理成第一特征信息;
    所述第一节点向第二节点发送第二信息,所述第二信息包括所述第一特征信息或第二特征信息,其中,所述第二特征信息为所述第一特征信息转换后的特征信息。
  2. 根据权利要求1所述的方法,其中,在所述第一节点基于第一AI网络模型,将第一信息处理成第一特征信息之前,所述方法还包括:
    所述第一节点训练得到所述第一AI网络模型。
  3. 根据权利要求2所述的方法,其中,所述第一AI网络模型与第三AI网络模型联合训练得到,所述第三AI网络模型与所述第一AI网络模型匹配,且用于对所述第一AI网络模型处理后的特征信息进行恢复处理;
    和/或,
    第二AI网络模型与第四AI网络模型联合训练得到,所述第二AI网络模型为所述第二节点具有的与所述第二特征信息匹配的AI网络模型,所述第四AI网络模型与所述第二AI网络模型匹配,且用于对所述第二AI网络模型处理后的特征信息进行恢复处理。
  4. 根据权利要求1至3中任一项所述的方法,其中,在所述第一节点向第二节点发送第一信息之前,所述方法还包括:
    所述第一节点获取转换信息,其中,所述转换信息用于将所述第一AI网络模型输出的特征信息转换成所述第二节点的AI网络模型匹配的特征信息;
    所述第一节点根据所述转换信息将所述第一特征信息转换成所述第二特征信息。
  5. 根据权利要求4所述的方法,其中,所述转换信息包括以下至少一项:
    第一指示信息,用于指示单位阵;
    第二指示信息,用于指示稀疏矩阵;
    第三指示信息,用于指示转换AI网络模型的相关信息,所述转换AI网络模型用于将所述第一特征信息转换成所述第二特征信息。
  6. 根据权利要求5所述的方法,其中,在所述转换信息包括所述第三指示信息的情况下,所述第一节点获取转换信息,包括:
    所述第一节点训练得到所述转换AI网络模型;或者,
    所述第一节点接收来自第三节点或所述第二节点的所述第三指示信息,其中,所述第三节点为除了所述第一节点和所述第二节点之外的设备。
  7. 根据权利要求6所述的方法,其中,所述第一节点训练得到所述转换AI网络模型,包括:
    所述第一节点获取第三特征信息以及接收来自所述第二节点的第四特征信息,其中,所述第三特征信息为基于所述第一AI网络模型对训练样本数据进行处理后得到的特征信 息,所述第四特征信息为基于所述第二节点具有的第二AI网络模型对所述训练样本数据进行处理后得到的特征信息;
    所述第一节点根据所述第三特征信息和所述第四特征信息,训练得到所述转换AI网络模型,其中,所述第三特征信息作为所述转换AI网络模型的输入,所述第四特征信息作为所述转换AI网络模型的输出目标。
  8. 根据权利要求6所述的方法,其中,在所述第一节点接收来自第三节点或所述第二节点的所述第三指示信息之前,所述方法还包括:
    所述第一节点向所述第三节点或所述第二节点发送第三信息,其中,所述第三信息包括所述第一AI网络模型的相关信息,或者,所述第三信息包括训练样本数据和第三特征信息,或者,所述第三信息包括所述第三特征信息;
    其中,所述第三特征信息为基于所述第一AI网络模型对所述训练样本数据进行处理后得到的特征信息,所述转换AI网络模型基于所述第三特征信息和第四特征信息中的至少一项训练得到,所述第四特征信息为基于所述第二节点具有的第二AI网络模型对所述训练样本数据进行处理后得到的特征信息。
  9. 根据权利要求7或8所述的方法,其中,所述方法还包括:
    所述第一节点估计第一信道的第一信道信息,其中,所述训练样本数据包括所述第一信道信息。
  10. 一种特征信息传输方法,包括:
    第二节点接收来自第一节点的第二信息,其中,所述第二信息包括第一特征信息或第二特征信息,其中,所述第一特征信息为基于所述第一节点具有的第一AI网络模型对第一信息进行处理后得到的特征信息,所述第二特征信息为所述第一特征信息转换后的特征信息;
    所述第二节点基于第四AI网络模型对所述第二特征信息进行恢复处理,得到所述第一信息,其中,在所述第二信息包括所述第一特征信息的情况下,所述第二特征信息由所述第二节点对所述第一特征信息进行转换处理得到。
  11. 根据权利要求10所述的方法,其中,在所述第二节点基于第四AI网络模型对所述第二特征信息进行恢复处理之前,所述方法还包括:
    所述第二节点训练得到所述第四AI网络模型。
  12. 根据权利要求11所述的方法,其中,所述第一AI网络模型与第三AI网络模型联合训练得到,所述第三AI网络模型与所述第一AI网络模型匹配,且用于对所述第一AI网络模型处理后的特征信息进行恢复处理;
    和/或,
    第二AI网络模型与所述第四AI网络模型联合训练得到,所述第二AI网络模型为所述第二节点具有的与所述第二特征信息匹配的AI网络模型,所述第四AI网络模型与所述第二AI网络模型匹配,且用于对所述第二AI网络模型处理后的特征信息进行恢复处理。
  13. 根据权利要求10至12中任一项所述的方法,其中,在所述第二节点基于第四AI网络模型对所述第二特征信息进行恢复处理之前,所述方法还包括:
    所述第二节点获取转换信息,其中,所述转换信息用于将所述第一AI网络模型输出的特征信息转换成所述第二节点的AI网络模型匹配的特征信息;
    所述第二节点根据所述转换信息将所述第一特征信息转换成所述第二特征信息。
  14. 根据权利要求13所述的方法,其中,所述转换信息包括以下至少一项:
    第一指示信息,用于指示单位阵;
    第二指示信息,用于指示稀疏矩阵;
    第三指示信息,用于指示转换AI网络模型的相关信息,所述转换AI网络模型用于将所述第一特征信息转换成所述第二特征信息。
  15. 根据权利要求14所述的方法,其中,在所述转换信息包括所述第三指示信息的情况下,所述第二节点获取转换信息,包括:
    所述第二节点训练得到所述转换AI网络模型;或者,
    所述第二节点接收来自第三节点或所述第一节点的所述第三指示信息,其中,所述第三节点为除了所述第一节点和所述第二节点之外的设备。
  16. 根据权利要求15所述的方法,其中,所述第二节点训练得到所述转换AI网络模型,包括:
    所述第二节点获取第四特征信息以及接收来自所述第一节点的第三特征信息,其中,所述第三特征信息为基于所述第一AI网络模型对训练样本数据进行处理后得到的特征信息,所述第四特征信息为基于所述第二节点具有的第二AI网络模型对所述训练样本数据进行处理后得到的特征信息;
    所述第二节点根据所述第三特征信息和所述第四特征信息,训练得到所述转换AI网络模型,其中,所述第三特征信息作为所述转换AI网络模型的输入,所述第四特征信息作为所述转换AI网络模型的输出目标。
  17. 根据权利要求15所述的方法,其中,在所述第二节点接收来自第三节点或所述第一节点的所述第三指示信息之前,所述方法还包括:
    所述第二节点向所述第三节点或所述第一节点发送第四信息,其中,所述第四信息包括所述第二节点具有的第二AI网络模型的相关信息,或者,所述第四信息包括训练样本数据和第四特征信息,或者,所述第四信息包括所述第四特征信息;
    其中,所述第四特征信息为基于所述第二AI网络模型对所述训练样本数据进行处理后得到的特征信息,所述转换AI网络模型基于第三特征信息和所述第四特征信息中的至少一项训练得到,所述第三特征信息为基于所述第一AI网络模型对所述训练样本数据进行处理后得到的特征信息。
  18. 根据权利要求16或17所述的方法,其中,所述方法还包括:
    所述第二节点估计第二信道的第二信道信息,其中,所述训练样本数据包括所述第 二信道信息。
  19. 根据权利要求16或17所述的方法,其中,在所述第二节点为网络侧设备的情况下,所述第二节点接收来自第一节点的第二信息,包括:
    所述第二节点接收来自至少一个终端的目标信道状态信息CSI报告,所述第一特征信息包括所述目标CSI报告携带的目标信道的信道特征信息,其中,所述至少一个终端包括所述第一节点;
    所述方法还包括:
    所述第二节点获取与所述目标信道相关的SRS的第三信道信息;
    所述第二节点根据所述第三信道信息训练所述第四AI网络模型。
  20. 一种转换信息确定方法,包括:
    第三节点获取第一AI网络模型对训练样本数据处理后得到的第三特征信息,以及获取第二AI网络模型对所述训练样本数据处理后得到的第四特征信息,其中,所述第一AI网络模型为第一节点具有的AI网络模型,所述第二AI网络模型为第二节点具有的AI网络模型;
    所述第三节点根据所述第三特征信息和所述第四特征信息,确定转换信息,其中,所述转换信息用于将目标训练样本对应的第三特征信息转换为与所述目标训练样本对应的第四特征信息,所述目标训练样本为所述训练样本数据中的任一样本;
    所述第三节点向所述第一节点和所述第二节点中的至少一个发送所述转换信息。
  21. 根据权利要求20所述的方法,其中,所述转换信息包括以下至少一项:
    第一指示信息,用于指示单位阵;
    第二指示信息,用于指示稀疏矩阵;
    第三指示信息,用于指示转换AI网络模型的相关信息,所述转换AI网络模型用于将第一特征信息转换成第二特征信息,其中,所述第一特征信息与所述第一AI网络模型匹配,所述第二特征信息与所述第二AI网络模型匹配。
  22. 根据权利要求20所述的方法,其中,所述第三节点获取第一AI网络模型对训练样本数据处理后得到的第三特征信息,包括:
    所述第三节点接收来自所述第一节点的第五信息,所述第五信息包括所述第一AI网络模型的相关信息;
    所述第三节点将训练样本数据输入所述第一AI网络模型,并获取所述第一AI网络模型输出的所述第三特征信息;
    和/或,
    所述第三节点获取第二AI网络模型对所述训练样本数据处理后得到的第四特征信息,包括:
    所述第三节点接收来自所述第二节点的第六信息,所述第六信息包括所述第二AI网络模型的相关信息;
    所述第三节点将训练样本数据输入所述第二AI网络模型,并获取所述第二AI网络模型输出的所述第四特征信息。
  23. 根据权利要求22所述的方法,其中,所述训练样本数据预先存储于所述第三节点,或者,所述训练样本数据来自第四节点,所述第四节点包括以下至少一项:
    所述第一节点;
    所述第二节点;
    专用于收集训练样本数据的节点;
    不具有所述第一AI网络模型和所述第二AI网络模型的节点。
  24. 根据权利要求20所述的方法,其中,所述第三节点获取第一AI网络模型对训练样本数据处理后得到的第三特征信息,包括:
    所述第三节点接收来自所述第一节点的训练样本数据和所述第三特征信息,其中,所述第三特征信息为将所述训练样本数据输入所述第一AI网络模型后,由所述第一AI网络模型输出的特征信息;
    所述第三节点获取第二AI网络模型对训练样本数据处理后得到的第四特征信息,包括:
    所述第三节点将所述训练样本数据输入所述第二AI网络模型,并获取所述第二AI网络模型输出的所述第四特征信息。
  25. 根据权利要求20所述的方法,其中,所述第三节点获取第二AI网络模型对训练样本数据处理后得到的第四特征信息,包括:
    所述第三节点接收来自所述第二节点的训练样本数据和所述第四特征信息,其中,所述第四特征信息为将所述训练样本数据输入所述第二AI网络模型后,由所述第二AI网络模型输出的特征信息;
    所述第三节点获取第一AI网络模型对训练样本数据处理后得到的第三特征信息,包括:
    所述第三节点将所述训练样本数据输入所述第一AI网络模型,并获取所述第一AI网络模型输出的所述第三特征信息。
  26. 一种特征信息传输装置,应用于第一节点,所述装置包括:
    第一处理模块,用于基于第一AI网络模型,将第一信息处理成第一特征信息;
    第一发送模块,用于向第二节点发送第二信息,所述第二信息包括所述第一特征信息或第二特征信息,其中,所述第二特征信息为所述第一特征信息转换后的特征信息。
  27. 一种特征信息传输装置,应用于第二节点,所述装置包括:
    第一接收模块,用于接收来自第一节点的第二信息,其中,所述第二信息包括第一特征信息或第二特征信息,其中,所述第一特征信息为基于所述第一节点具有的第一AI网络模型对第一信息进行处理后得到的特征信息,所述第二特征信息为所述第一特征信息转换后的特征信息;
    第二处理模块,用于基于第四AI网络模型对所述第二特征信息进行恢复处理,得到所述第一信息,其中,在所述第二信息包括所述第一特征信息的情况下,所述第二特征信息由所述第二节点对所述第一特征信息进行转换处理得到。
  28. 一种转换信息确定装置,应用于第三节点,所述装置包括:
    第一获取模块,用于获取第一AI网络模型对训练样本数据处理后得到的第三特征信息,以及获取第二AI网络模型对所述训练样本数据处理后得到的第四特征信息,其中,所述第一AI网络模型为第一节点具有的AI网络模型,所述第二AI网络模型为第二节点具有的AI网络模型;
    第一确定模块,用于根据所述第三特征信息和所述第四特征信息,确定转换信息,其中,所述转换信息用于将目标训练样本对应的第三特征信息转换为与所述目标训练样本对应的第四特征信息,所述目标训练样本为所述训练样本数据中的任一样本;
    第二发送模块,用于向所述第一节点和所述第二节点中的至少一个发送所述转换信息。
  29. 一种通信设备,包括处理器和存储器,所述存储器存储可在所述处理器上运行的程序或指令,所述程序或指令被所述处理器执行时实现如权利要求1至9中任一项所述的特征信息传输方法的步骤,或者实现如权利要求10至19中任一项所述的特征信息传输方法的步骤,或者实现如权利要求20至25中任一项所述的转换信息确定方法的步骤。
  30. 一种可读存储介质,所述可读存储介质上存储程序或指令,所述程序或指令被处理器执行时实现如权利要求1至9中任一项所述的特征信息传输方法的步骤,或者实现如权利要求10至19中任一项所述的特征信息传输方法的步骤,或者实现如权利要求20至25中任一项所述的转换信息确定方法的步骤。
PCT/CN2023/110480 2022-08-04 2023-08-01 特征信息传输方法、转换信息确定方法、装置和通信设备 WO2024027682A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210933644.0 2022-08-04
CN202210933644.0A CN117560251A (zh) 2022-08-04 2022-08-04 特征信息传输方法、转换信息确定方法、装置和通信设备

Publications (1)

Publication Number Publication Date
WO2024027682A1 true WO2024027682A1 (zh) 2024-02-08

Family

ID=89822068

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/110480 WO2024027682A1 (zh) 2022-08-04 2023-08-01 特征信息传输方法、转换信息确定方法、装置和通信设备

Country Status (2)

Country Link
CN (1) CN117560251A (zh)
WO (1) WO2024027682A1 (zh)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111464220A (zh) * 2020-03-10 2020-07-28 西安交通大学 一种基于深度学习的信道状态信息重建方法
WO2022040055A1 (en) * 2020-08-18 2022-02-24 Qualcomm Incorporated Processing timeline considerations for channel state information

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111464220A (zh) * 2020-03-10 2020-07-28 西安交通大学 一种基于深度学习的信道状态信息重建方法
WO2022040055A1 (en) * 2020-08-18 2022-02-24 Qualcomm Incorporated Processing timeline considerations for channel state information

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
HUAWEI, HISILICON: "Discussion on general aspects of AI/ML framework", 3GPP DRAFT; R1-2203139, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. RAN WG1, no. e-Meeting; 20220509 - 20220520, 29 April 2022 (2022-04-29), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France, XP052143957 *
MODERATOR (HUAWEI): "Summary#1 of [109-e-R18-AI/ML-03]", 3GPP DRAFT; R1-2205222, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. RAN WG1, no. e-Meeting; 20220509 - 20220520, 11 May 2022 (2022-05-11), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France, XP052146776 *
VIVO: "Other aspects on AI/ML for CSI feedback enhancement", 3GPP DRAFT; R1-2203551, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. RAN WG1, no. e-Meeting; 20220509 - 20220520, 29 April 2022 (2022-04-29), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France, XP052153026 *

Also Published As

Publication number Publication date
CN117560251A (zh) 2024-02-13

Similar Documents

Publication Publication Date Title
RU2011102566A (ru) Индексирование ретрансляционных антенн для связи посредством совместно используемых антенн
US20230262728A1 (en) Communication Method and Communication Apparatus
CN114390519B (zh) 一种无线信道密钥生成方法、装置、设备及存储介质
WO2023246618A1 (zh) 信道矩阵处理方法、装置、终端及网络侧设备
WO2024027682A1 (zh) 特征信息传输方法、转换信息确定方法、装置和通信设备
US20230136416A1 (en) Neural network obtaining method and apparatus
EP3148251A1 (en) Data transmission method and device
WO2023179570A1 (zh) 信道特征信息传输方法、装置、终端及网络侧设备
WO2024051564A1 (zh) 信息传输方法、ai网络模型训练方法、装置和通信设备
WO2023070675A1 (zh) 数据处理的方法及装置
WO2023060503A1 (zh) 信息处理方法、装置、设备、介质、芯片、产品及程序
WO2023283785A1 (zh) 信号处理的方法及接收机
WO2023185980A1 (zh) 信道特征信息传输方法、装置、终端及网络侧设备
WO2024055974A1 (zh) Cqi传输方法、装置、终端及网络侧设备
WO2023185995A1 (zh) 信道特征信息传输方法、装置、终端及网络侧设备
WO2023185790A1 (zh) 信息传输方法、装置、终端及网络侧设备
WO2024027683A1 (zh) 模型匹配方法、装置、通信设备及可读存储介质
WO2023179460A1 (zh) 信道特征信息传输方法、装置、终端及网络侧设备
WO2023097645A1 (zh) 数据获取方法、装置、设备、介质、芯片、产品及程序
WO2024032775A1 (zh) 量化方法和装置
WO2023125423A1 (zh) 编码方法、设备及可读存储介质
WO2023125424A1 (zh) 解码方法、设备及可读存储介质
WO2023004563A1 (zh) 获取参考信号的方法及通信设备
WO2024007949A1 (zh) Ai模型处理方法、装置、终端及网络侧设备
WO2024140422A1 (zh) Ai单元的性能监测方法、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: 23849389

Country of ref document: EP

Kind code of ref document: A1